View Issue Details
| ID | Project | Category | View Status | Date Submitted | Last Update |
|---|---|---|---|---|---|
| 0000424 | Pgpool-II | Bug | public | 2018-08-16 16:40 | 2019-05-21 10:15 |
| Reporter | nagata | Assigned To | t-ishii | ||
| Priority | normal | Severity | minor | Reproducibility | have not tried |
| Status | closed | Resolution | open | ||
| Summary | 0000424: pcp_recovery_nod in follow_master_command possibly fails with an error | ||||
| Description | One of our clients is getting this error each time follow_master calls pcp_recovery_node on a downed standby: ERROR: failed to process PCP request at the moment DETAIL: failover is in progress In the current implementation, pcp_recovery_node can not be run during failover/failback. (This is checked in pcp_process_command()). So, it seams that pcp_recovery_node is being called before the failover_command completes. The ideas to run pcp_recovery_node safely in follow_master are as below. 1. Insert a sleep before running pcp_recovery_node. 2. Retry pcp_recovery_node if this fails due to the error. However, can this be handled in Pgpool-II itself? If running pcp_recovery_node in follow_master_command is a expected use case, I think Pgpool-II should provide the safe way to do this. Any idea to resolve this? | ||||
| Tags | No tags attached. | ||||
|
|
Probably you are misunderstanding the usage of follow master command. The command should be run *after* failover done. Users can issue SQL to Pgpool-II while follow master commands are running. See follow_master_command.sh generated by pgpool_setup. |
|
|
On my understanding, the follow_master_command are triggered in failover() and this is before Req_info->switching is cleared (= set to false). |
|
|
Don't you miss the next line? if(Req_info->request_queue_tail != Req_info->request_queue_head) |
|
|
Oops, sorry, I missed this. I'll looking into this, again. |
|
|
OK - Req_info->request_queue_tail is incremented when failover or failover request is registered by register_node_operation_request(). - Req_info->request_queue_head is incremented at the top of the loop in failover(), that is, at the start point of processing each failback/failover request. So, (Req_info->request_queue_tail != Req_info->request_queue_head) is true when multiple failover or failback were registered but some part of them is not processed yet. This may happen if failover is requested twice or more quickly, or the "failback" request is registered at pcp_recovery_node in the first follow_master_command, for example. I'll report more when we get a log messages from the client or when I can reproduce this in my machine. |
|
|
Can we close this issue? |
|
|
No response from the reporter over 1 month. I am going to close this issue. |
| Date Modified | Username | Field | Change |
|---|---|---|---|
| 2018-08-16 16:40 | nagata | New Issue | |
| 2018-08-16 17:03 | t-ishii | Note Added: 0002161 | |
| 2018-08-16 17:34 | nagata | Note Added: 0002162 | |
| 2018-08-16 17:39 | t-ishii | Note Added: 0002163 | |
| 2018-08-16 17:43 | nagata | Note Added: 0002164 | |
| 2018-08-16 18:20 | nagata | Note Added: 0002165 | |
| 2019-01-30 10:08 | administrator | Assigned To | => t-ishii |
| 2019-01-30 10:08 | administrator | Status | new => assigned |
| 2019-02-04 08:59 | t-ishii | Status | assigned => feedback |
| 2019-02-24 20:21 | t-ishii | Note Added: 0002402 | |
| 2019-05-21 10:15 | t-ishii | Note Added: 0002609 | |
| 2019-05-21 10:15 | t-ishii | Status | feedback => closed |
