[MXS-3881] Maxscale killing connection to backend node when load data infile is run with transaction replay Created: 2021-11-16 Updated: 2022-02-04 Resolved: 2021-11-19 |
|
| Status: | Closed |
| Project: | MariaDB MaxScale |
| Component/s: | readwritesplit |
| Affects Version/s: | 6.1.4 |
| Fix Version/s: | 6.2.0 |
| Type: | Bug | Priority: | Major |
| Reporter: | Isaac Venn (Inactive) | Assignee: | markus makela |
| Resolution: | Fixed | Votes: | 0 |
| Labels: | triage | ||
| Description |
|
Tested in skysql. What happens is, with transaction replay enabled (and also causal reads, but we aren't sure if that's related), when a load data infile operation is run, somewhere in the execution the connection is killed to the backend node and the load fails. Here is an example of what is seen from the command line interface:
And this is what is seen in the maxscale log:
To reproduce: Spin up a new HA cluster.
Load data from attached file as csv like so:
|
| Comments |
| Comment by markus makela [ 2021-11-17 ] |
|
Does the problem consistently go away if you disable both transaction_replay and causal reads? |
| Comment by Isaac Venn (Inactive) [ 2021-11-17 ] |
|
Just disabling transaction replay solves the issue. |
| Comment by Isaac Venn (Inactive) [ 2021-11-18 ] |
|
Removed. |