caleb-brown/redict
1
0
Fork 0
mirror of https://codeberg.org/redict/redict.git synced 2025-01-23 00:28:26 -05:00
Code Issues Actions Packages Projects Releases Wiki Activity

Cluster: PFAIL -> FAIL transition allowed for slaves.

Browse Source 
First change: now there is no need to be a master in order to detect a
failure, however the majority of masters signaling PFAIL or FAIL is needed.

This change is important because it allows slaves rejoining the cluster
after a partition to sense the FAIL condition so that eventually all the
nodes agree on failures.
This commit is contained in:
antirez 2013-09-20 11:26:44 +02:00
parent 925ea9f858
commit 3c9bb8751a
1 changed files with 23 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
src/cluster.c
View File

@ -594,25 +594,36 @@ void clusterRenameNode(clusterNode *node, char *newname) {
/* This function checks if a given node should be marked as FAIL. /* This function checks if a given node should be marked as FAIL.
* It happens if the following conditions are met: * It happens if the following conditions are met:
* *
* 1) We are a master node. Only master nodes can mark a node as failing. * 1) We received enough failure reports from other master nodes via gossip.
* 2) We received enough failure reports from other nodes via gossip. * Enough means that the majority of the masters signaled the node is
* Enough means that the majority of the masters believe the node is * down recently.
* down. * 2) We believe this node is in PFAIL state.
* 3) We believe this node is in PFAIL state.
* *
* If a failure is detected we also inform the whole cluster about this * If a failure is detected we also inform the whole cluster about this
* event trying to force every other node to set the FAIL flag for the node. * event trying to force every other node to set the FAIL flag for the node.
*
* Note that the form of agreement used here is weak, as we collect the majority
* of masters state during some time, and even if we force agreement by
* propagating the FAIL message, because of partitions we may not reach every
* node. However:
*
* 1) Either we reach the majority and eventually the FAIL state will propagate
* to all the cluster.
* 2) Or there is no majority so no slave promotion will be authorized and the
* FAIL flag will be cleared after some time.
*/ */
void markNodeAsFailingIfNeeded(clusterNode *node) { void markNodeAsFailingIfNeeded(clusterNode *node) {
int failures; int failures;
int needed_quorum = (server.cluster->size / 2) + 1; int needed_quorum = (server.cluster->size / 2) + 1;
if (!(server.cluster->myself->flags & REDIS_NODE_MASTER)) return;
if (!(node->flags & REDIS_NODE_PFAIL)) return; /* We can reach it. */ if (!(node->flags & REDIS_NODE_PFAIL)) return; /* We can reach it. */
if (node->flags & REDIS_NODE_FAIL) return; /* Already FAILing. */ if (node->flags & REDIS_NODE_FAIL) return; /* Already FAILing. */
failures = 1 + clusterNodeFailureReportsCount(node); /* +1 is for myself. */ failures = clusterNodeFailureReportsCount(node);
if (failures < needed_quorum) return; /* Also count myself as a voter if I'm a master. */
if (server.cluster->myself->flags & REDIS_NODE_MASTER)
failures += 1;
if (failures < needed_quorum) return; /* No weak agreement from masters. */
redisLog(REDIS_NOTICE, redisLog(REDIS_NOTICE,
"Marking node %.40s as failing (quorum reached).", node->name); "Marking node %.40s as failing (quorum reached).", node->name);
@ -622,8 +633,10 @@ void markNodeAsFailingIfNeeded(clusterNode *node) {
node->flags |= REDIS_NODE_FAIL; node->flags |= REDIS_NODE_FAIL;
node->fail_time = time(NULL); node->fail_time = time(NULL);
/* Broadcast the failing node name to everybody */ /* Broadcast the failing node name to everybody, forcing all the other
clusterSendFail(node->name); * reachable nodes to flag the node as FAIL. */
if (server.cluster->myself->flags & REDIS_NODE_MASTER)
clusterSendFail(node->name);
clusterUpdateState(); clusterUpdateState();
clusterSaveConfigOrDie(); clusterSaveConfigOrDie();
} }