This addresses two problems, one where infinite (negative) repeat count is broken for all types for Redis,
and another specific to cluster mode where redirection is needed.
Now allows and works correctly for negative (i.e. -1) repeat values passed with `-r` argument to redis-cli
as documented here https://redis.io/topics/rediscli#continuously-run-the-same-command which seems to have
regressed as a feature in 95b988 (though that commit removed bad integer wrap-around to `0` behaviour).
This broken behaviour exists currently (e50458), and redis-cli will just exit immediately with repeat `-r <= 0`
as opposed to send commands indefinitely as it should with `-r < 0`
Additionally prevents a repeat * interval seconds hang/time spent doing nothing at the start before issuing
commands in cluster mode (`-c`), where the command needed to redirect to a slot on another node, as commands
where failing and waiting to be reissued but this was fully repeated before being reissued. For example,
redis-cli -c -r 10 -i 0.5 INCR test_key_not_on_6379
Would hang and show nothing for 5 seconds (10 * 0.5) before showing
(integer) 1
(integer) 2
(integer) 3
(integer) 4
(integer) 5
(integer) 6
(integer) 7
(integer) 8
(integer) 9
(integer) 10
at half second intervals as intended.
If a key exists in the target node during a migration (BUSYKEY),
the value of the key on both nodes (source and target) will be compared.
If the key has the same value on both keys, the migration will be
automatically retried with the REPLACE argument in order to override
the target's key.
If the key has different values, the behaviour will depend on such
cases:
- In case of 'fix' command, the migration will stop and the user
will be warned to manually check the key(s).
- In other cases (ie. reshard), if the user launched the command
with the --cluster-replace option, the migration will be
retried with the REPLACE argument, elsewhere the migration will
stop and the user will be warned.
When loading data, we call processEventsWhileBlocked
to process events and execute commands.
But if we are loading AOF it's dangerous, because
processCommand would call freeMemoryIfNeeded to evict,
and that will break data consistency, see issue #5686.
Thanks to @soloestoy for discovering this issue in #5667.
This is an alternative fix in order to avoid both cycling the clients
and also disconnecting clients just having valid read-only transactions
pending.
these metrics become negative when RSS is smaller than the used_memory.
This can easily happen when the program allocated a lot of memory and haven't
written to it yet, in which case the kernel doesn't allocate any pages to the process
In MEMORY USAGE command, we count the key argv[2] into usage,
but the argument in command may contains free spaces because of
sdsMakeRoomFor. But the key in db never contains free spaces
because we use sdsdup when dbAdd, so using the real key to
count the usage is more accurate.
If we encounter an unsupported protocol in the "bind" list, don't
ipso-facto consider it a fatal error. We continue to abort startup if
there are no listening sockets at all.
This ensures that the lack of IPv6 support does not prevent Redis from
starting on Debian where we try to bind to the ::1 interface by default
(via "bind 127.0.0.1 ::1"). A machine with IPv6 disabled (such as some
container systems) would simply fail to start Redis after the initiall
call to apt(8).
This is similar to the case for where "bind" is not specified:
https://github.com/antirez/redis/issues/3894
... and was based on the corresponding PR:
https://github.com/antirez/redis/pull/4108
... but also adds EADDRNOTAVAIL to the list of errors to catch which I
believe is missing from there.
This issue was raised in Debian as both <https://bugs.debian.org/900284>
& <https://bugs.debian.org/914354>.
This really helps spot it in the logs, otherwise it does not look like a
warning/error. For example:
Creating Server TCP listening socket ::1:6379: bind: Cannot assign requested address
... is not nearly as clear as:
Could not create server TCP listening listening socket ::1:6379: bind: Cannot assign requested address
This commit fixes#5570. It is a similar bug to one fixed a few weeks
ago and is due to the range API to be called with NULL as "end ID"
parameter instead of repeating again the start ID, to be sure that we
selectively issue the entry with a given ID, or we get zero returned
(and we know we should emit a NULL reply).
This fixes the issue reported in #5570.
This was fixed the hard way, that is, propagating more information to
the lower level API about this being a request to read just the history,
so that the code is simpler and less likely to regress.
- clusterManagerFixOpenSlot: ensure that the
slot is unassigned before ADDSLOTS
- clusterManagerFixSlotsCoverage: after cold
migration, the slot configuration
is now updated on all the nodes.
This bug had a double effect:
1. Sometimes entries may not be emitted, producing broken protocol where
the array length was greater than the emitted entires, blocking the
client waiting for more data.
2. Some other time the right entry was claimed, but a wrong entry was
returned to the client.
This fix should correct both the instances.
server.hz was uninitialized between initServerConfig and initServer.
this can lead to someone (e.g. queued modules) doing createObject,
and accessing an uninitialized variable, that can potentially be 0,
and lead to a crash.
So far it was not possible to setup Sentinel with authentication
enabled. This commit introduces this feature: every Sentinel will try to
authenticate with other sentinels using the same password it is
configured to accept clients with.
So for instance if a Sentinel has a "requirepass" configuration
statemnet set to "foo", it will use the "foo" password to authenticate
with every other Sentinel it connects to. So basically to add the
"requirepass" to all the Sentinels configurations is enough in order to
make sure that:
1) Clients will require the password to access the Sentinels instances.
2) Each Sentinel will use the same password to connect and authenticate
with every other Sentinel in the group.
Related to #3279 and #3329.
Fake clients are used in special situations and are not linked to the
normal clients list, freeing them will always result in Redis crashing
in one way or the other.
It's not common to send replies to fake clients, but we have one usage
in the modules API. When a client is blocked, we associate to the
blocked client object (that is safe to manipulate in a thread), a fake
client that accumulates replies. So because of this bug there was
the problem described in issue #5443.
The fix was verified to work with the provided example module. To write
a regression is very hard and unlikely to be triggered in the future.
This adds support for passing a password through a REDISCLI_AUTH
environment variable (which is safer than the CLI), which might often be
safer than passing it through a CLI argument.
Passing a password this way does not trigger the warning about passing a
password through CLI arguments, and CLI arguments take precedence over
it.
The fix was removed by c8ca71d40 attempting to fix the stack generation
on ARM64, without testing if it would still work on ARM32.
Now it should work both sides.
They play better with Lua scripting, otherwise Lua will see status
replies as "ok" = "string" which is very odd, and actually as @oranagra
reasoned in issue #5456 in the rest of the Redis code base there was no
such concern as saving a few bytes when the protocol is emitted.
Related to #4840.
Note that when we re-enter the event loop with aeProcessEvents() we
don't process timers, nor before/after sleep callbacks, so we should
never end calling freeClientsInAsyncFreeQueue() when re-entering the
loop.
Related to #5201.
I removed the !!! Warning part since compared to the other errors, a
missing EXEC is in theory a normal happening in the AOF file, at least
in theory: may happen in a differnet number of situations, and it's
probably better to don't give the user the feeling that something really
bad happened.
This is useful in order to spot bugs where we fail
at updating the pointer returned by the insertion
function. Normally often the same pointer is returned,
making it harder than needed to spot bugs.
Related to #5210.
When HAVE_MALLOC_SIZE is false, each call to zrealloc causes used_memory
to increase by PREFIX_SIZE more than it should, due to mis-matched
accounting between the original zmalloc (which includes PREFIX size in
its increment) and zrealloc (which misses it from its decrement).
I've also supplied a command-line test to easily demonstrate the
problem. It's not wired into the test framework, because I don't know
TCL so I'm not sure how to automate it.
sdsZmallocSize assumes a dynamically allocated SDS. When given a string
object created by createEmbeddedStringObject, it calls zmalloc_size on a
pointer that isn't the one returned by zmalloc
There are two problems if we use lastcmd:
1. BRPOPLPUSH cannot be rewrited as RPOPLPUSH in multi/exec
In mulit/exec context, the lastcmd is exec.
2. Redis will crash when execute RPOPLPUSH loading from AOF
In fakeClient, the lastcmd is NULL.
Storing the context is useless, because we can't really reuse that
later. For instance in the API RM_DictNext() that returns a
RedisModuleString for the next key iterated, the user should pass the
new context, because we may run the keys of the dictionary in a
different context of the one where the dictionary was created. Also the
dictionary may be created without a context, but we may still demand
automatic memory management for the returned strings while iterating.
By using the "C" suffix for functions getting pointer/len, we can do the
same in the future for other modules APIs that need a variant with
pointer/len and that are now accepting a RedisModuleString.
The burden of having to always create RedisModuleString objects within a
module context was too much, especially now that we have threaded
operations and modules are doing more interesting things. The context in
the string API is currently only used for automatic memory managemnet,
so now the API was modified so that the user can opt-out this feature by
passing a NULL context.
During the full database resync we may still have unsaved changes
on the receiving side. This causes a race condition between
synced data rename/load and the rename of rdbSave tempfile.
