Attempt to match against an empty key | Crash in UNIV_DEBUG
Affects | Status | Importance | Assigned to | Milestone | ||
---|---|---|---|---|---|---|
Galera |
Fix Released
|
Undecided
|
Alex Yurchenko | |||
Percona XtraDB Cluster moved to https://jira.percona.com/projects/PXC | Status tracked in 5.6 | |||||
5.5 |
Invalid
|
Undecided
|
Unassigned | |||
5.6 |
Fix Released
|
Undecided
|
Unassigned |
Bug Description
[root@ip-
Percona-
Percona-
Percona-
percona-
Percona-
percona-
Percona-
Scenario is 2 node cluster.
node2 is wsrep_desync=ON and FLUSH TABLES WITH READ LOCK;
node1 runs: sysbench --db-driver=mysql --test=
Once sysbench finishes on node1, execute UNLOCK TABLES on node2. Node2 gets this error:
Version: '5.5.34-55' socket: '/var/lib/
140129 15:25:09 [Note] WSREP: Service thread queue flushed.
140129 15:25:09 [Note] WSREP: Synchronized with group, ready for connections
140129 15:25:09 [Note] WSREP: wsrep_notify_cmd is not defined, skipping notification.
140129 15:25:22 [Note] WSREP: Node 1 (ip-10-29-254-10) desyncs itself from group
140129 15:25:22 [Note] WSREP: Shifting SYNCED -> DONOR/DESYNCED (TO: 2424911)
140129 15:26:36 [Note] WSREP: Provider paused at 3f186ea4-
140129 15:27:16 [Note] WSREP: Provider resumed.
140129 16:52:06 [Note] WSREP: Provider paused at 3f186ea4-
140129 16:53:18 [Note] WSREP: Created page /var/lib/
140129 16:54:26 [Note] WSREP: Created page /var/lib/
140129 16:55:35 [Note] WSREP: Created page /var/lib/
140129 16:56:42 [Note] WSREP: Created page /var/lib/
140129 17:07:50 [Note] WSREP: Provider resumed.
140129 17:07:51 [ERROR] WSREP: Attempt to match against an empty key (1,0): 22 (Invalid argument)
at galera/
140129 17:07:51 [Note] WSREP: applier thread exiting (code:7)
140129 17:07:51 [ERROR] WSREP: node consistency compromised, aborting
140129 17:07:51 [Note] WSREP: starting shutdown
140129 17:07:51 [Note] /usr/sbin/mysqld: Normal shutdown
140129 17:07:51 [Note] WSREP: Stop replication
140129 17:07:51 [Note] WSREP: Closing send monitor...
140129 17:07:51 [Note] WSREP: Closed send monitor.
140129 17:07:51 [Note] WSREP: gcomm: terminating thread
140129 17:07:51 [Note] WSREP: gcomm: joining thread
140129 17:07:51 [Note] WSREP: gcomm: closing backend
140129 17:07:52 [Note] WSREP: view(view_
} joined {
} left {
} partitioned {
})
140129 17:07:52 [Note] WSREP: New COMPONENT: primary = no, bootstrap = no, my_idx = 0, memb_num = 1
140129 17:07:52 [Note] WSREP: view((empty))
140129 17:07:52 [Note] WSREP: gcomm: closed
140129 17:07:52 [Note] WSREP: Flow-control interval: [1000000, 1000000]
140129 17:07:52 [Note] WSREP: Received NON-PRIMARY.
140129 17:07:52 [Note] WSREP: Shifting DONOR/DESYNCED -> OPEN (TO: 3624911)
140129 17:07:52 [Note] WSREP: Received self-leave message.
140129 17:07:52 [Note] WSREP: Flow-control interval: [0, 0]
140129 17:07:52 [Note] WSREP: Received SELF-LEAVE. Closing connection.
140129 17:07:52 [Note] WSREP: Shifting OPEN -> CLOSED (TO: 3624911)
140129 17:07:52 [Note] WSREP: RECV thread exiting 0: Success
140129 17:07:52 [Note] WSREP: recv_thread() joined.
140129 17:07:52 [Note] WSREP: Closing replication queue.
140129 17:07:52 [Note] WSREP: Closing slave action queue.
140129 17:07:54 [Note] WSREP: killing local connection: 18
Note that node2 is also hung at this point: mysqld is there, but it does not accept connections. I'm unsure if this will happen consistently or not.
Related branches
- David Bennett: Pending requested
- Diff: 0 lines
Changed in galera: | |
status: | Fix Committed → Fix Released |
Tested this with 5.6 and sysbench, couldn't replicate it there. Need to check with 5.5.
@Jay,
One more thing, I see 5.5 has been used with 3.x, were you able to replicate this with 5.5 + 2.x as well?