[MXS-2576] Columnstore Monitor inaccurately labels a UM as slave Created: 2019-06-24 Updated: 2020-08-25 Resolved: 2019-08-13 |
|
| Status: | Closed |
| Project: | MariaDB MaxScale |
| Component/s: | Monitor |
| Affects Version/s: | 2.3.8 |
| Fix Version/s: | 2.3.12 |
| Type: | Bug | Priority: | Major |
| Reporter: | Geoff Montee (Inactive) | Assignee: | markus makela |
| Resolution: | Fixed | Votes: | 0 |
| Labels: | None | ||
| Sprint: | MXS-SPRINT-87 |
| Description |
|
Columnstore Monitor can inaccurately label a server as a "slave" if the monitor's connection gets dropped in the get_cs_version() function: The reason is that this function is called inside CsMonitor::update_server_status to determine if the server supports the mcsSystemPrimary function: If the get_cs_version function returns a value lower than 10200, then MaxScale checks the server's configuration for the "primary" parameter. However, if the monitor's connection gets disconnected in the get_cs_version function, then the function will return 0. Therefore, since this value is lower than 10200, MaxScale will check the server's configuration for the primary parameter. But since the DBA expects MaxScale to use the mcsSystemPrimary function instead, this parameter is most likely not going to be set at all for the server. This causes MaxScale to set the server as a "Slave", since it thinks that some other server is the primary server. Instead of allowing the server to be incorrectly set as a "Slave", Columnstore Monitor should detect that the connection died, and it should label the server as "Down". Here's some relevant entries from a MaxScale error log that shows this happening:
|
| Comments |
| Comment by markus makela [ 2019-06-25 ] |
|
Increasing query_retries should reduce the likelihood of this happening. |