Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Duplicate
-
10.5.12
Description
MariaDB 10.5 does not checkpoint as I would expect (i.e same as in MariaDB 10.4 or PS 5.7 ) .
Attached is the graph of how MariaDB 10.5 checkpoints, i.e it let's redo log get to max checkpoint age, then checkpoints everything , then start over.
It is easily repeatable by creating a new instance from scratch with all the defaults and insert a few records, checking that checkpoint age does not move.
SEIS On MariaDB :
Log sequence number 1850969
Log flushed up to 1850969
Pages flushed up to 1846760
Last checkpoint at 1846748
SEIS on PS 5.7:
Log sequence number 4441325019
Log flushed up to 4441325019
Pages flushed up to 4441325019
Last checkpoint at 4441325010
It can be seen that "Last checkpoint at xxxx" is close to LSN on PS, but on MariaDB 10.5 it will remain with a fixed value until max checkpoint age.
In the attachment it can be seen that every ~4 hours the redo logs gets full, does an almost complete checkpoint and starts again. This causes an stall of a few seconds while the disks (which luckily are fast) do the checkpoint
On MariaDB 10.4 this does not happen and I have also compared configuration between 10.4 and 10.5 and could not find anything.
This other bug report MDEV-25093 mentions that any flushing is skipped but it seems something still broken.
Attachments
Issue Links
- is duplicated by
-
-
- Closed
-
- relates to
-
-
- Closed
-
Activity
| Field | Original Value | New Value |
|---|---|---|
| Link |
This issue relates to |
| Assignee | Marko Mäkelä [ marko ] | |
| Labels | need_feedback performance |
| Component/s | Storage Engine - InnoDB [ 10129 ] | |
| Fix Version/s | 10.6.5 [ 26034 ] | |
| Fix Version/s | 10.5.13 [ 26026 ] | |
| Resolution | Duplicate [ 3 ] | |
| Status | Open [ 1 ] | Closed [ 6 ] |
| Link |
This issue is duplicated by |
| Link |
This issue relates to |
| Workflow | MariaDB v3 [ 126392 ] | MariaDB v4 [ 159789 ] |
| Link |
This issue relates to |
| Description |
MariaDB 10.5 does not checkpoint as I would expect (i.e same as in MariaDB 10.4 or PS 5.7 ) .
Attached is the graph of how MariaDB 10.5 checkpoints, i.e it let's redo log get to max checkpoint age, then checkpoints everything , then start over. It is easily repeatable by creating a new instance from scratch with all the defaults and insert a few records, checking that checkpoint age does not move. SEIS On MariaDB : Log sequence number 1850969 Log flushed up to 1850969 Pages flushed up to 1846760 Last checkpoint at 1846748 SEIS on PS 5.7: Log sequence number 4441325019 Log flushed up to 4441325019 Pages flushed up to 4441325019 Last checkpoint at 4441325010 It can be seen that "Last checkpoint at xxxx" is close to LSN on PS, but on MariaDB 10.5 it will remain with a fixed value until max checkpoint age. In the attachment it can be seen that every ~4 hours the redo logs gets full, does an almost complete checkpoint and starts again. This causes an stall of a few seconds while the disks (which luckily are fast) do the checkpoint On MariaDB 10.4 this does not happen and I have also compared configuration between 10.4 and 10.5 and could not find anything. This other bug report https://jira.mariadb.org/browse/MDEV-25093 mentions that any flushing is skipped but it seems something still broken. |
MariaDB 10.5 does not checkpoint as I would expect (i.e same as in MariaDB 10.4 or PS 5.7 ) .
Attached is the graph of how MariaDB 10.5 checkpoints, i.e it let's redo log get to max checkpoint age, then checkpoints everything , then start over. It is easily repeatable by creating a new instance from scratch with all the defaults and insert a few records, checking that checkpoint age does not move. SEIS On MariaDB : Log sequence number 1850969 Log flushed up to 1850969 Pages flushed up to 1846760 Last checkpoint at 1846748 SEIS on PS 5.7: Log sequence number 4441325019 Log flushed up to 4441325019 Pages flushed up to 4441325019 Last checkpoint at 4441325010 It can be seen that "Last checkpoint at xxxx" is close to LSN on PS, but on MariaDB 10.5 it will remain with a fixed value until max checkpoint age. In the attachment it can be seen that every ~4 hours the redo logs gets full, does an almost complete checkpoint and starts again. This causes an stall of a few seconds while the disks (which luckily are fast) do the checkpoint On MariaDB 10.4 this does not happen and I have also compared configuration between 10.4 and 10.5 and could not find anything. This other bug report |
Thank you for the report. This feels like a duplicate of
MDEV-26626. Can you please test if the problem is repeatable with a recent 10.5 development snapshot from http://hasky.askmonty.org/archive/10.5/ (or with recent 10.5 source code)?