[MXS-2625] Transaction Replay: Trying to execute statement before Clustrix cluster is up, gets the statement stuck Created: 2019-08-07 Updated: 2019-09-13 Resolved: 2019-09-06 |
|
| Status: | Closed |
| Project: | MariaDB MaxScale |
| Component/s: | xpandmon |
| Affects Version/s: | None |
| Fix Version/s: | 2.4.1 |
| Type: | Bug | Priority: | Major |
| Reporter: | Rahul Joshi (Inactive) | Assignee: | markus makela |
| Resolution: | Fixed | Votes: | 0 |
| Labels: | None | ||
| Environment: |
MaxScale server karma197: |
||
| Attachments: |
|
| Sprint: | MXS-SPRINT-88, MXS-SPRINT-89 |
| Description |
|
Connected to MaxScale from client machine and run following:
Restart clustrix here by running: Before the cluster forms, run following.
Expected: Actual: Relevant logs:
Full logs and config file attached. Clustrix log for service shutwon and cluster formation times:
|
| Comments |
| Comment by Johan Wikman [ 2019-08-21 ] | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
Could repeat the behaviour using 658aae6b6c77774c34d0f52c03f63edc2d44969e | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Comment by Johan Wikman [ 2019-08-21 ] | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
rahul.joshi@mariadb.com With 2.4.1, MaxScale does not hang but the connection is closed. I think the closing is caused by session commands (use test;) being executed within the transaction. If perform the same test like:
Restart cluster
then the transaction is replayed successfully. Could you retry after
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Comment by Rahul Joshi (Inactive) [ 2019-08-22 ] | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
Hi johan.wikman, Moved the use test; session command out of the trx.
Restarted the cluster and run following before it can form group
conf file is the same:
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Comment by Johan Wikman [ 2019-09-02 ] | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
I have now tried this numerous times:
I think this is related to a new command being received while a transaction replay (that will fail due to no nodes being up) is in progress. The following works: First
Then on one Clustrix node:
Wait until the restart has been performed (it's possible to sudo mysql) on the node.
That results in successful-trx-replay.log However, when done like: First
Then on one Clustrix node:
and immediately when the command has returned
the mysql-client will hang and we get the log unsuccessful-trx-replay.log If the hanging of the client is interrupted by 3 ctrl-C:s
we further get into the log interruption.log It appears that if a transaction replay fails, the internal book-keeping of RWS gets confused so that the transaction is not replayed correctly and a reply is not sent to the client. There is some additional logging in the logs. | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Comment by markus makela [ 2019-09-05 ] | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
rahul.joshi@mariadb.com Can you try reproducing this with a build from the latest 2.4 branch? I've tested this numerous times and I suspect this might've been fixed by this commit that's in 2.4.1.
The problem is easily reproducible when that commit is reverted. | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Comment by Johan Wikman [ 2019-09-05 ] | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
rahul.joshi@mariadb.com I just verified that with commit 658aae6b6c77774c34d0f52c03f63edc2d44969e the behaviour repeats itselfs consistently, but using tag maxscale-2.4.2 it does not. | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Comment by Rahul Joshi (Inactive) [ 2019-09-06 ] | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
Hi,
Logs show checksum error. |