Loading...

XML

Word

Printable

JSON

Type: Improvement
Resolution: Fixed
Priority: Major - P3
Fix Version/s: WT11.2.0, 7.1.0-rc0
Affects Version/s: None
Component/s: Checkpoints
Labels:
- supportability

Assigned Teams:

Storage Engines

Story Points:
8
Sprint:
2023-06-27 Lord of the Sprints, 2023-07-11 WiredTractor, 2023-07-25 Absolute unit, StorEng - 2023-08-08, ASeasonTooMany-2023-08-22
Downstream Team Attention:
Needed
Teams Impacted:

Triage and Release
Downstream Changes Summary:

Hide
Checkpoint statistics were altered:
* Checkpoint-related stats got moved out of the big "transaction" category into their own "checkpoint" category
* The descriptions were altered to remove the redundant "transaction checkpoint" text (since them being in the checkpoint category already adds the "checkpoint" text to the description automatically)
* "checkpoint running" became "checkpoint state", and works largely the same except that any non-zero value is equivalent to the old "1" value.
* "checkpoint currently running for history store file" got folded into the checkpoint state and is no longer a separate statistic

Show
Checkpoint statistics were altered: * Checkpoint-related stats got moved out of the big "transaction" category into their own "checkpoint" category * The descriptions were altered to remove the redundant "transaction checkpoint" text (since them being in the checkpoint category already adds the "checkpoint" text to the description automatically) * "checkpoint running" became "checkpoint state", and works largely the same except that any non-zero value is equivalent to the old "1" value. * "checkpoint currently running for history store file" got folded into the checkpoint state and is no longer a separate statistic

Checkpointing is one of the most user-visibly disruptive maintenance operations WiredTiger does. Sometimes a lot of time is spent doing checkpoints, and it isn't clear what it is doing with that time.

We should add metrics and/or log lines that give better insight.

An example from a customer shows checkpoints completing in between 15 and 20 minutes, spending 4 minutes of that time writing content (2 at the start, 2 at the end), with no meaningful indication of what happens for the remaining 11 to 16 minutes. There is no meaningful time being spent flushing checkpoint content to disk.

A starting point for this is probably building or finding a workload that takes a long time to create a new checkpoint.