2024-03-21 09:30:47 -04:00
|
|
|
// SPDX-FileCopyrightText: 2024 Redict Contributors
|
|
|
|
// SPDX-FileCopyrightText: 2024 Salvatore Sanfilippo <antirez at gmail dot com>
|
|
|
|
//
|
|
|
|
// SPDX-License-Identifier: BSD-3-Clause
|
2024-03-21 15:11:44 -04:00
|
|
|
// SPDX-License-Identifier: LGPL-3.0-only
|
2024-03-21 09:30:47 -04:00
|
|
|
|
2021-02-05 12:51:31 -05:00
|
|
|
#define _XOPEN_SOURCE 700
|
2024-03-21 05:49:18 -04:00
|
|
|
#include "redictmodule.h"
|
2021-01-29 08:38:30 -05:00
|
|
|
#include <stdio.h>
|
|
|
|
#include <stdlib.h>
|
|
|
|
#include <pthread.h>
|
|
|
|
#include <time.h>
|
|
|
|
|
|
|
|
#define UNUSED(x) (void)(x)
|
|
|
|
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
typedef struct {
|
2024-03-25 07:41:50 -04:00
|
|
|
/* Mutex for protecting RedictModule_BlockedClientMeasureTime*() API from race
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
* conditions due to timeout callback triggered in the main thread. */
|
|
|
|
pthread_mutex_t measuretime_mutex;
|
|
|
|
int measuretime_completed; /* Indicates that time measure has ended and will not continue further */
|
|
|
|
int myint; /* Used for replying */
|
|
|
|
} BlockPrivdata;
|
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
void blockClientPrivdataInit(RedictModuleBlockedClient *bc) {
|
|
|
|
BlockPrivdata *block_privdata = RedictModule_Calloc(1, sizeof(*block_privdata));
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
block_privdata->measuretime_mutex = (pthread_mutex_t)PTHREAD_MUTEX_INITIALIZER;
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModule_BlockClientSetPrivateData(bc, block_privdata);
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
}
|
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
void blockClientMeasureTimeStart(RedictModuleBlockedClient *bc, BlockPrivdata *block_privdata) {
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
pthread_mutex_lock(&block_privdata->measuretime_mutex);
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModule_BlockedClientMeasureTimeStart(bc);
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
pthread_mutex_unlock(&block_privdata->measuretime_mutex);
|
|
|
|
}
|
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
void blockClientMeasureTimeEnd(RedictModuleBlockedClient *bc, BlockPrivdata *block_privdata, int completed) {
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
pthread_mutex_lock(&block_privdata->measuretime_mutex);
|
|
|
|
if (!block_privdata->measuretime_completed) {
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModule_BlockedClientMeasureTimeEnd(bc);
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
if (completed) block_privdata->measuretime_completed = 1;
|
|
|
|
}
|
|
|
|
pthread_mutex_unlock(&block_privdata->measuretime_mutex);
|
|
|
|
}
|
|
|
|
|
2021-01-29 08:38:30 -05:00
|
|
|
/* Reply callback for blocking command BLOCK.DEBUG */
|
2024-03-25 07:41:50 -04:00
|
|
|
int HelloBlock_Reply(RedictModuleCtx *ctx, RedictModuleString **argv, int argc) {
|
2021-01-29 08:38:30 -05:00
|
|
|
UNUSED(argv);
|
|
|
|
UNUSED(argc);
|
2024-03-25 07:41:50 -04:00
|
|
|
BlockPrivdata *block_privdata = RedictModule_GetBlockedClientPrivateData(ctx);
|
|
|
|
return RedictModule_ReplyWithLongLong(ctx,block_privdata->myint);
|
2021-01-29 08:38:30 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Timeout callback for blocking command BLOCK.DEBUG */
|
2024-03-25 07:41:50 -04:00
|
|
|
int HelloBlock_Timeout(RedictModuleCtx *ctx, RedictModuleString **argv, int argc) {
|
2021-01-29 08:38:30 -05:00
|
|
|
UNUSED(argv);
|
|
|
|
UNUSED(argc);
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModuleBlockedClient *bc = RedictModule_GetBlockedClientHandle(ctx);
|
|
|
|
BlockPrivdata *block_privdata = RedictModule_GetBlockedClientPrivateData(ctx);
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
blockClientMeasureTimeEnd(bc, block_privdata, 1);
|
2024-03-25 07:41:50 -04:00
|
|
|
return RedictModule_ReplyWithSimpleString(ctx,"Request timedout");
|
2021-01-29 08:38:30 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Private data freeing callback for BLOCK.DEBUG command. */
|
2024-03-25 07:41:50 -04:00
|
|
|
void HelloBlock_FreeData(RedictModuleCtx *ctx, void *privdata) {
|
2021-01-29 08:38:30 -05:00
|
|
|
UNUSED(ctx);
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
BlockPrivdata *block_privdata = privdata;
|
|
|
|
pthread_mutex_destroy(&block_privdata->measuretime_mutex);
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModule_Free(privdata);
|
2021-01-29 08:38:30 -05:00
|
|
|
}
|
|
|
|
|
2021-09-14 08:14:09 -04:00
|
|
|
/* Private data freeing callback for BLOCK.BLOCK command. */
|
2024-03-25 07:41:50 -04:00
|
|
|
void HelloBlock_FreeStringData(RedictModuleCtx *ctx, void *privdata) {
|
|
|
|
RedictModule_FreeString(ctx, (RedictModuleString*)privdata);
|
2021-09-14 08:14:09 -04:00
|
|
|
}
|
|
|
|
|
2021-01-29 08:38:30 -05:00
|
|
|
/* The thread entry point that actually executes the blocking part
|
|
|
|
* of the command BLOCK.DEBUG. */
|
|
|
|
void *BlockDebug_ThreadMain(void *arg) {
|
|
|
|
void **targ = arg;
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModuleBlockedClient *bc = targ[0];
|
2021-01-29 08:38:30 -05:00
|
|
|
long long delay = (unsigned long)targ[1];
|
|
|
|
long long enable_time_track = (unsigned long)targ[2];
|
2024-03-25 07:41:50 -04:00
|
|
|
BlockPrivdata *block_privdata = RedictModule_BlockClientGetPrivateData(bc);
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
|
2021-01-29 08:38:30 -05:00
|
|
|
if (enable_time_track)
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
blockClientMeasureTimeStart(bc, block_privdata);
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModule_Free(targ);
|
2021-01-29 08:38:30 -05:00
|
|
|
|
|
|
|
struct timespec ts;
|
|
|
|
ts.tv_sec = delay / 1000;
|
|
|
|
ts.tv_nsec = (delay % 1000) * 1000000;
|
|
|
|
nanosleep(&ts, NULL);
|
|
|
|
if (enable_time_track)
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
blockClientMeasureTimeEnd(bc, block_privdata, 0);
|
|
|
|
block_privdata->myint = rand();
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModule_UnblockClient(bc,block_privdata);
|
2021-01-29 08:38:30 -05:00
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* The thread entry point that actually executes the blocking part
|
2021-02-05 12:51:31 -05:00
|
|
|
* of the command BLOCK.DOUBLE_DEBUG. */
|
2021-01-29 08:38:30 -05:00
|
|
|
void *DoubleBlock_ThreadMain(void *arg) {
|
|
|
|
void **targ = arg;
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModuleBlockedClient *bc = targ[0];
|
2021-01-29 08:38:30 -05:00
|
|
|
long long delay = (unsigned long)targ[1];
|
2024-03-25 07:41:50 -04:00
|
|
|
BlockPrivdata *block_privdata = RedictModule_BlockClientGetPrivateData(bc);
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
blockClientMeasureTimeStart(bc, block_privdata);
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModule_Free(targ);
|
2021-01-29 08:38:30 -05:00
|
|
|
struct timespec ts;
|
|
|
|
ts.tv_sec = delay / 1000;
|
|
|
|
ts.tv_nsec = (delay % 1000) * 1000000;
|
|
|
|
nanosleep(&ts, NULL);
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
blockClientMeasureTimeEnd(bc, block_privdata, 0);
|
2024-03-25 07:41:50 -04:00
|
|
|
/* call again RedictModule_BlockedClientMeasureTimeStart() and
|
|
|
|
* RedictModule_BlockedClientMeasureTimeEnd and ensure that the
|
2021-01-29 08:38:30 -05:00
|
|
|
* total execution time is 2x the delay. */
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
blockClientMeasureTimeStart(bc, block_privdata);
|
2021-01-29 08:38:30 -05:00
|
|
|
nanosleep(&ts, NULL);
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
blockClientMeasureTimeEnd(bc, block_privdata, 0);
|
|
|
|
block_privdata->myint = rand();
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModule_UnblockClient(bc,block_privdata);
|
2021-01-29 08:38:30 -05:00
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
void HelloBlock_Disconnected(RedictModuleCtx *ctx, RedictModuleBlockedClient *bc) {
|
|
|
|
RedictModule_Log(ctx,"warning","Blocked client %p disconnected!",
|
2021-01-29 08:38:30 -05:00
|
|
|
(void*)bc);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* BLOCK.DEBUG <delay_ms> <timeout_ms> -- Block for <count> milliseconds, then reply with
|
|
|
|
* a random number. Timeout is the command timeout, so that you can test
|
|
|
|
* what happens when the delay is greater than the timeout. */
|
2024-03-25 07:45:47 -04:00
|
|
|
int HelloBlock_RedictCommand(RedictModuleCtx *ctx, RedictModuleString **argv, int argc) {
|
2024-03-25 07:41:50 -04:00
|
|
|
if (argc != 3) return RedictModule_WrongArity(ctx);
|
2021-01-29 08:38:30 -05:00
|
|
|
long long delay;
|
|
|
|
long long timeout;
|
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
if (RedictModule_StringToLongLong(argv[1],&delay) != REDICTMODULE_OK) {
|
|
|
|
return RedictModule_ReplyWithError(ctx,"ERR invalid count");
|
2021-01-29 08:38:30 -05:00
|
|
|
}
|
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
if (RedictModule_StringToLongLong(argv[2],&timeout) != REDICTMODULE_OK) {
|
|
|
|
return RedictModule_ReplyWithError(ctx,"ERR invalid count");
|
2021-01-29 08:38:30 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
pthread_t tid;
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModuleBlockedClient *bc = RedictModule_BlockClient(ctx,HelloBlock_Reply,HelloBlock_Timeout,HelloBlock_FreeData,timeout);
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
blockClientPrivdataInit(bc);
|
2021-01-29 08:38:30 -05:00
|
|
|
|
|
|
|
/* Here we set a disconnection handler, however since this module will
|
|
|
|
* block in sleep() in a thread, there is not much we can do in the
|
|
|
|
* callback, so this is just to show you the API. */
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModule_SetDisconnectCallback(bc,HelloBlock_Disconnected);
|
2021-01-29 08:38:30 -05:00
|
|
|
|
|
|
|
/* Now that we setup a blocking client, we need to pass the control
|
|
|
|
* to the thread. However we need to pass arguments to the thread:
|
|
|
|
* the delay and a reference to the blocked client handle. */
|
2024-03-25 07:41:50 -04:00
|
|
|
void **targ = RedictModule_Alloc(sizeof(void*)*3);
|
2021-01-29 08:38:30 -05:00
|
|
|
targ[0] = bc;
|
|
|
|
targ[1] = (void*)(unsigned long) delay;
|
|
|
|
// pass 1 as flag to enable time tracking
|
|
|
|
targ[2] = (void*)(unsigned long) 1;
|
|
|
|
|
|
|
|
if (pthread_create(&tid,NULL,BlockDebug_ThreadMain,targ) != 0) {
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModule_AbortBlock(bc);
|
|
|
|
return RedictModule_ReplyWithError(ctx,"-ERR Can't start thread");
|
2021-01-29 08:38:30 -05:00
|
|
|
}
|
2024-03-25 07:41:50 -04:00
|
|
|
return REDICTMODULE_OK;
|
2021-01-29 08:38:30 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
/* BLOCK.DEBUG_NOTRACKING <delay_ms> <timeout_ms> -- Block for <count> milliseconds, then reply with
|
|
|
|
* a random number. Timeout is the command timeout, so that you can test
|
|
|
|
* what happens when the delay is greater than the timeout.
|
|
|
|
* this command does not track background time so the background time should no appear in stats*/
|
2024-03-25 07:45:47 -04:00
|
|
|
int HelloBlockNoTracking_RedictCommand(RedictModuleCtx *ctx, RedictModuleString **argv, int argc) {
|
2024-03-25 07:41:50 -04:00
|
|
|
if (argc != 3) return RedictModule_WrongArity(ctx);
|
2021-01-29 08:38:30 -05:00
|
|
|
long long delay;
|
|
|
|
long long timeout;
|
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
if (RedictModule_StringToLongLong(argv[1],&delay) != REDICTMODULE_OK) {
|
|
|
|
return RedictModule_ReplyWithError(ctx,"ERR invalid count");
|
2021-01-29 08:38:30 -05:00
|
|
|
}
|
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
if (RedictModule_StringToLongLong(argv[2],&timeout) != REDICTMODULE_OK) {
|
|
|
|
return RedictModule_ReplyWithError(ctx,"ERR invalid count");
|
2021-01-29 08:38:30 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
pthread_t tid;
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModuleBlockedClient *bc = RedictModule_BlockClient(ctx,HelloBlock_Reply,HelloBlock_Timeout,HelloBlock_FreeData,timeout);
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
blockClientPrivdataInit(bc);
|
2021-01-29 08:38:30 -05:00
|
|
|
|
|
|
|
/* Here we set a disconnection handler, however since this module will
|
|
|
|
* block in sleep() in a thread, there is not much we can do in the
|
|
|
|
* callback, so this is just to show you the API. */
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModule_SetDisconnectCallback(bc,HelloBlock_Disconnected);
|
2021-01-29 08:38:30 -05:00
|
|
|
|
|
|
|
/* Now that we setup a blocking client, we need to pass the control
|
|
|
|
* to the thread. However we need to pass arguments to the thread:
|
|
|
|
* the delay and a reference to the blocked client handle. */
|
2024-03-25 07:41:50 -04:00
|
|
|
void **targ = RedictModule_Alloc(sizeof(void*)*3);
|
2021-01-29 08:38:30 -05:00
|
|
|
targ[0] = bc;
|
|
|
|
targ[1] = (void*)(unsigned long) delay;
|
|
|
|
// pass 0 as flag to enable time tracking
|
|
|
|
targ[2] = (void*)(unsigned long) 0;
|
|
|
|
|
|
|
|
if (pthread_create(&tid,NULL,BlockDebug_ThreadMain,targ) != 0) {
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModule_AbortBlock(bc);
|
|
|
|
return RedictModule_ReplyWithError(ctx,"-ERR Can't start thread");
|
2021-01-29 08:38:30 -05:00
|
|
|
}
|
2024-03-25 07:41:50 -04:00
|
|
|
return REDICTMODULE_OK;
|
2021-01-29 08:38:30 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
/* BLOCK.DOUBLE_DEBUG <delay_ms> -- Block for 2 x <count> milliseconds,
|
|
|
|
* then reply with a random number.
|
2024-03-25 07:41:50 -04:00
|
|
|
* This command is used to test multiple calls to RedictModule_BlockedClientMeasureTimeStart()
|
|
|
|
* and RedictModule_BlockedClientMeasureTimeEnd() within the same execution. */
|
2024-03-25 07:45:47 -04:00
|
|
|
int HelloDoubleBlock_RedictCommand(RedictModuleCtx *ctx, RedictModuleString **argv, int argc) {
|
2024-03-25 07:41:50 -04:00
|
|
|
if (argc != 2) return RedictModule_WrongArity(ctx);
|
2021-01-29 08:38:30 -05:00
|
|
|
long long delay;
|
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
if (RedictModule_StringToLongLong(argv[1],&delay) != REDICTMODULE_OK) {
|
|
|
|
return RedictModule_ReplyWithError(ctx,"ERR invalid count");
|
2021-01-29 08:38:30 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
pthread_t tid;
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModuleBlockedClient *bc = RedictModule_BlockClient(ctx,HelloBlock_Reply,HelloBlock_Timeout,HelloBlock_FreeData,0);
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 08:12:49 -05:00
|
|
|
blockClientPrivdataInit(bc);
|
2021-01-29 08:38:30 -05:00
|
|
|
|
|
|
|
/* Now that we setup a blocking client, we need to pass the control
|
|
|
|
* to the thread. However we need to pass arguments to the thread:
|
|
|
|
* the delay and a reference to the blocked client handle. */
|
2024-03-25 07:41:50 -04:00
|
|
|
void **targ = RedictModule_Alloc(sizeof(void*)*2);
|
2021-01-29 08:38:30 -05:00
|
|
|
targ[0] = bc;
|
|
|
|
targ[1] = (void*)(unsigned long) delay;
|
|
|
|
|
|
|
|
if (pthread_create(&tid,NULL,DoubleBlock_ThreadMain,targ) != 0) {
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModule_AbortBlock(bc);
|
|
|
|
return RedictModule_ReplyWithError(ctx,"-ERR Can't start thread");
|
2021-01-29 08:38:30 -05:00
|
|
|
}
|
2024-03-25 07:41:50 -04:00
|
|
|
return REDICTMODULE_OK;
|
2021-01-29 08:38:30 -05:00
|
|
|
}
|
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModuleBlockedClient *blocked_client = NULL;
|
2021-07-01 10:11:27 -04:00
|
|
|
|
|
|
|
/* BLOCK.BLOCK [TIMEOUT] -- Blocks the current client until released
|
|
|
|
* or TIMEOUT seconds. If TIMEOUT is zero, no timeout function is
|
|
|
|
* registered.
|
|
|
|
*/
|
2024-03-25 07:45:47 -04:00
|
|
|
int Block_RedictCommand(RedictModuleCtx *ctx, RedictModuleString **argv, int argc) {
|
2024-03-25 07:41:50 -04:00
|
|
|
if (RedictModule_IsBlockedReplyRequest(ctx)) {
|
|
|
|
RedictModuleString *r = RedictModule_GetBlockedClientPrivateData(ctx);
|
|
|
|
return RedictModule_ReplyWithString(ctx, r);
|
|
|
|
} else if (RedictModule_IsBlockedTimeoutRequest(ctx)) {
|
|
|
|
RedictModule_UnblockClient(blocked_client, NULL); /* Must be called to avoid leaks. */
|
2021-07-01 10:11:27 -04:00
|
|
|
blocked_client = NULL;
|
2024-03-25 07:41:50 -04:00
|
|
|
return RedictModule_ReplyWithSimpleString(ctx, "Timed out");
|
2021-07-01 10:11:27 -04:00
|
|
|
}
|
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
if (argc != 2) return RedictModule_WrongArity(ctx);
|
2021-07-01 10:11:27 -04:00
|
|
|
long long timeout;
|
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
if (RedictModule_StringToLongLong(argv[1], &timeout) != REDICTMODULE_OK) {
|
|
|
|
return RedictModule_ReplyWithError(ctx, "ERR invalid timeout");
|
2021-07-01 10:11:27 -04:00
|
|
|
}
|
|
|
|
if (blocked_client) {
|
2024-03-25 07:41:50 -04:00
|
|
|
return RedictModule_ReplyWithError(ctx, "ERR another client already blocked");
|
2021-07-01 10:11:27 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Block client. We use this function as both a reply and optional timeout
|
|
|
|
* callback and differentiate the different code flows above.
|
|
|
|
*/
|
2024-03-25 07:45:47 -04:00
|
|
|
blocked_client = RedictModule_BlockClient(ctx, Block_RedictCommand,
|
|
|
|
timeout > 0 ? Block_RedictCommand : NULL, HelloBlock_FreeStringData, timeout);
|
2024-03-25 07:41:50 -04:00
|
|
|
return REDICTMODULE_OK;
|
2021-07-01 10:11:27 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
/* BLOCK.IS_BLOCKED -- Returns 1 if we have a blocked client, or 0 otherwise.
|
|
|
|
*/
|
2024-03-25 07:45:47 -04:00
|
|
|
int IsBlocked_RedictCommand(RedictModuleCtx *ctx, RedictModuleString **argv, int argc) {
|
2021-07-01 10:11:27 -04:00
|
|
|
UNUSED(argv);
|
|
|
|
UNUSED(argc);
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModule_ReplyWithLongLong(ctx, blocked_client ? 1 : 0);
|
|
|
|
return REDICTMODULE_OK;
|
2021-07-01 10:11:27 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
/* BLOCK.RELEASE [reply] -- Releases the blocked client and produce the specified reply.
|
|
|
|
*/
|
2024-03-25 07:45:47 -04:00
|
|
|
int Release_RedictCommand(RedictModuleCtx *ctx, RedictModuleString **argv, int argc) {
|
2024-03-25 07:41:50 -04:00
|
|
|
if (argc != 2) return RedictModule_WrongArity(ctx);
|
2021-07-01 10:11:27 -04:00
|
|
|
if (!blocked_client) {
|
2024-03-25 07:41:50 -04:00
|
|
|
return RedictModule_ReplyWithError(ctx, "ERR No blocked client");
|
2021-07-01 10:11:27 -04:00
|
|
|
}
|
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModuleString *replystr = argv[1];
|
|
|
|
RedictModule_RetainString(ctx, replystr);
|
|
|
|
RedictModule_UnblockClient(blocked_client, replystr);
|
2021-07-01 10:11:27 -04:00
|
|
|
blocked_client = NULL;
|
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
RedictModule_ReplyWithSimpleString(ctx, "OK");
|
2021-07-01 10:11:27 -04:00
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
return REDICTMODULE_OK;
|
2021-07-01 10:11:27 -04:00
|
|
|
}
|
2021-01-29 08:38:30 -05:00
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
int RedictModule_OnLoad(RedictModuleCtx *ctx, RedictModuleString **argv, int argc) {
|
2021-01-29 08:38:30 -05:00
|
|
|
UNUSED(argv);
|
|
|
|
UNUSED(argc);
|
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
if (RedictModule_Init(ctx,"block",1,REDICTMODULE_APIVER_1)
|
|
|
|
== REDICTMODULE_ERR) return REDICTMODULE_ERR;
|
2021-01-29 08:38:30 -05:00
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
if (RedictModule_CreateCommand(ctx,"block.debug",
|
2024-03-25 07:45:47 -04:00
|
|
|
HelloBlock_RedictCommand,"",0,0,0) == REDICTMODULE_ERR)
|
2024-03-25 07:41:50 -04:00
|
|
|
return REDICTMODULE_ERR;
|
2021-01-29 08:38:30 -05:00
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
if (RedictModule_CreateCommand(ctx,"block.double_debug",
|
2024-03-25 07:45:47 -04:00
|
|
|
HelloDoubleBlock_RedictCommand,"",0,0,0) == REDICTMODULE_ERR)
|
2024-03-25 07:41:50 -04:00
|
|
|
return REDICTMODULE_ERR;
|
2021-01-29 08:38:30 -05:00
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
if (RedictModule_CreateCommand(ctx,"block.debug_no_track",
|
2024-03-25 07:45:47 -04:00
|
|
|
HelloBlockNoTracking_RedictCommand,"",0,0,0) == REDICTMODULE_ERR)
|
2024-03-25 07:41:50 -04:00
|
|
|
return REDICTMODULE_ERR;
|
2021-01-29 08:38:30 -05:00
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
if (RedictModule_CreateCommand(ctx, "block.block",
|
2024-03-25 07:45:47 -04:00
|
|
|
Block_RedictCommand, "", 0, 0, 0) == REDICTMODULE_ERR)
|
2024-03-25 07:41:50 -04:00
|
|
|
return REDICTMODULE_ERR;
|
2021-07-01 10:11:27 -04:00
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
if (RedictModule_CreateCommand(ctx,"block.is_blocked",
|
2024-03-25 07:45:47 -04:00
|
|
|
IsBlocked_RedictCommand,"",0,0,0) == REDICTMODULE_ERR)
|
2024-03-25 07:41:50 -04:00
|
|
|
return REDICTMODULE_ERR;
|
2021-07-01 10:11:27 -04:00
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
if (RedictModule_CreateCommand(ctx,"block.release",
|
2024-03-25 07:45:47 -04:00
|
|
|
Release_RedictCommand,"",0,0,0) == REDICTMODULE_ERR)
|
2024-03-25 07:41:50 -04:00
|
|
|
return REDICTMODULE_ERR;
|
2021-07-01 10:11:27 -04:00
|
|
|
|
2024-03-25 07:41:50 -04:00
|
|
|
return REDICTMODULE_OK;
|
2021-01-29 08:38:30 -05:00
|
|
|
}
|