cockroachdb / cockroachdb/cockroach

sql/stats: forecast closer to the current time

Open
#130,710 4 comments 0 reactions 0 assignees View on GitHub
A-sql-table-stats C-performance O-postmortem O-support P-2 T-sql-queries
Dominant language
Go
Stars
32.5k
Forks
4.1k
PR merge metrics
PR metrics pending

Description

Pulling out [this comment](https://github.com/cockroachdb/cockroach/issues/119967#issuecomment-1985109828) into a separate issue. We should try to pick a forecast time closer to the current time.

> I think both this and https://github.com/cockroachlabs/support/issues/2792 are strong arguments for picking something closer to the current time for the forecast, rather than the predicted time of the next automatic statistics collection. I've resisted doing this because it makes forecasts nondeterministic, but I think there are ways we could reduce the pain of nondeterminism:
>
> 1. round the current time down to the nearest hour / minute / some known interval, and use this for the forecast time, to get pseudo-determinism
> 2. for AOST queries use the calculated txn time instead of the current time
> 3. log changes in forecasts if they're significant, or plot them in some time-series chart
> 4. add a time to SHOW STATISTICS WITH FORECAST so that we can see what was happening at a certain time
> 5. add current forecasts to statement diagnostics bundles

Also see these slack threads:
- https://cockroachlabs.slack.com/archives/C01RX2G8LT1/p1705421378616609
- https://cockroachlabs.slack.com/archives/G01R3EYPQKA/p1709587805437729
- https://cockroachlabs.slack.com/archives/G01R3EYPQKA/p1726164326086669

Jira issue: CRDB-42187

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.