server,sql,ui: disable system.eventlog by default, remove event box#84983
server,sql,ui: disable system.eventlog by default, remove event box#84983knz wants to merge 2 commits intocockroachdb:masterfrom
Conversation
|
This change is |
andreimatei
left a comment
There was a problem hiding this comment.
should we do something to the events box in the right pane of the Metrics pane?
Reviewable status:
complete! 0 of 0 LGTMs obtained
Do we like it still? |
|
I kinda like it because it's one of the few things in the console that gives a user the impression that the console has any clue about stuff that's happening to a cluster. It doesn't seem very important to me, but between having it and not having it I'd vote for having it. Not at the cost of inter-region writes for schema changes, though. So if that motivation is burning, I'm not against getting rid of it (but then I think it's up to this PR to do something to the UI). If it's not burning, I would leave it alone until 23.1, when hopefully the Obs Service should be deployed and ready to serve it. |
|
how do you feel about using gossip entries (with a ~week TTL) for this? |
|
I think I'd rather remove the functionality than add gossip on top of structured logging + optional event log table + the obs service export that I'm adding. |
|
Done. RFAL (I'd like this to merge for 22.2) |
system.eventlog by defaultWe're going to remove the population of system.eventlog for structured events: - it adds a 5-way write to many small "maintainance" transactions. - it forces a global roundtrip in multi-region schema changes. - it does not significantly help when something goes wrong, as when something goes wrong there may not be a way to access the system table any more. - it is not sufficiently tamper-proof, and thus unsuitable for audit logging. The deprecation had been announced as of 21.1. It's time to move forward with the removal: we're making the feature opt-in instead of opt-out for one release cycle, then remove it in the next cycle. Release justification: stop released code from relying on deprecated features Release note (backward-incompatible change): The storage of operational events in `system.eventlog` had been deprecated since v21.1. It is now disabled by default. Deployments should now transition to structured logging events, stored outside of CockroachDB. It is still possible to manually re-enable the feature via the cluster setting `server.eventlog.enabled`, but the feature will be removed altogether in an upcoming version of CockroachDB. Release note (ui change): The cluster event box has been removed from the DB Console metrics page. Administrators are invited to refer to their OPS logs instead.
Release justification: test-only changes Release note: None
|
@knz can we stop writing schema change events to the table as a compromise for this release? Then we can point people to information about where they can find that info if they're looking for it. I think people rely on that box for the node events. It provides helpful reassurance in a running cluster. |
|
Yes we can do what you suggest as a compromise. However:
Are you sure of this? Do we have proof? |
|
Okay it looks like we've collected proof through discussion. I'll make the recommended change. |
|
Let's go for the following PR instead: #86174 If folk like it, I'll close this one and also the other issues i've created around it. |
86174: sql: async writes to system eventlog r=andreimatei a=knz Informs #73308. Epic: CRDB-18596 This PR aims to remove the synchronous write to system.eventlog for structured events, which is undesirable: - it adds a 5-way write to many small "maintainance" transactions. - it forces a global roundtrip in multi-region schema changes. Instead it makes the writes asynchronous. This PR subsumes #84983. See individual commits for details. Release justification: bug fixes Co-authored-by: Raphael 'kena' Poss <knz@thaumogen.net>
|
Subsumed by #86174. |
Fixes #69772.
Fixes #72325.
Informs #73308.
We're going to remove the population of system.eventlog for structured
events:
something goes wrong there may not be a way to access the system
table any more.
audit logging.
The deprecation had been announced as of 21.1. It's time to move
forward with the removal: we're making the feature opt-in instead of
opt-out for one release cycle, then remove it in the next cycle.
Release justification: stop released code from relying on deprecated features
Release note (backward-incompatible change): The storage of
operational events in
system.eventloghad been deprecated sincev21.1. It is now disabled by default. Deployments should now
transition to structured logging events, stored outside of
CockroachDB. It is still possible to manually re-enable the feature
via the cluster setting
server.eventlog.enabled, but the featurewill be removed altogether in an upcoming version of CockroachDB.