kv, gossip: remove misc deprecated system config code #82329
kv, gossip: remove misc deprecated system config code #82329craig[bot] merged 1 commit intocockroachdb:masterfrom
Conversation
ajwerner
left a comment
There was a problem hiding this comment.
A thing of beauty! Let's get somebody on KV to have a look. Maybe @irfansharif or @nvanbenschoten?
There was a problem hiding this comment.
should we be deleting this?
There was a problem hiding this comment.
I think we can delete this but I'll look into it into a follow up.
12ac40d to
21ddc67
Compare
kvoli
left a comment
There was a problem hiding this comment.
LGTM, +@nvanbenschoten
nit: I think there should be a commit message. Referencing #76279 for posterity.
irfansharif
left a comment
There was a problem hiding this comment.
I've paged a lot of this out, surprisingly, but the broad strokes LGTM. I'm curious what's left in order to get rid of all this code:
cockroach/pkg/config/system.go
Lines 76 to 84 in 609825e
| @@ -2193,7 +2193,6 @@ func (s *Store) GetConfReader(ctx context.Context) (spanconfig.StoreReader, erro | |||
|
|
|||
| if s.cfg.SpanConfigsDisabled || | |||
There was a problem hiding this comment.
Should we get rid of all these knobs now that we're no longer gossipping the system config span: SpanConfigsDisabled, spanconfigstore.EnabledSetting, s.TestingKnobs().UseSystemConfigSpanForQueues.
There was a problem hiding this comment.
Did you mean to resolve this comment? If we want to get rid of these in future commits, worth leaving a TODO.
There was a problem hiding this comment.
I swear I replied but the message disappeared.
I tried removing it but some tests were failing and it wasn't obvious to me why. I added two more TODOs to remove SpanConfigsDisabled and spanconfigstore.EnabledSetting and you already had one for s.TestingKnobs().UseSystemConfigSpanForQueues
21ddc67 to
c122f24
Compare
c122f24 to
4b1be2e
Compare
|
@irfansharif do you want to take another look here? I have some follow work here as well |
irfansharif
left a comment
There was a problem hiding this comment.
or at least update
SystemConfigSpanto contain just system.descriptor and
system.zones.
I'm not sure why we need an unsplittable span to contain those two tables specifically but happy to learn in your subsequent PRs. This is really great stuff, thanks for cleaning up the mess.
| SystemConfigSpan: true, | ||
| }, | ||
| } | ||
| etArgsWithTrigger.InternalCommitTrigger = &roachpb.InternalCommitTrigger{} |
There was a problem hiding this comment.
We don't need this line at all, right?
There was a problem hiding this comment.
We still do, it changes how the lock span is populated below.
I'm just gonna update it to use a NodeLiveness span instead in the Trigger.
| @@ -2193,7 +2193,6 @@ func (s *Store) GetConfReader(ctx context.Context) (spanconfig.StoreReader, erro | |||
|
|
|||
| if s.cfg.SpanConfigsDisabled || | |||
There was a problem hiding this comment.
Did you mean to resolve this comment? If we want to get rid of these in future commits, worth leaving a TODO.
Added a comment for this. |
There is still some work left to actually remove `SystemConfigSpan` or at least update `SystemConfigSpan` to contain just system.descriptor and system.zones. cockroachdb#76279 Release note: None
4b1be2e to
dc80d3c
Compare
|
Thanks for reviewing! bors r+ |
|
Build succeeded: |
140160: sql: rollback transaction eagerly upon error r=rafiss a=rafiss ### sql: rollback transaction eagerly upon error If we have not used any savepoints, it's safe to rollback the transaction immediately when a non-retryable error is encountered. This releases locks sooner, which is useful so we avoid needing to wait until the client sends a ROLLBACK statement. Release note (bug fix): Transactions that enter the aborted state will now release locks they are holding immediately, as long as there is no SAVEPOINT active in the transaction. ### sql: make transaction_timeout abort transaction eagerly Previously, the transaction_timeout setting would only cause the transaction to abort if the timeout had already expired by the time the next statement is received by CRDB. Now we create a timer that fires if the timeout expires while we are waiting for the next statement. There's no release note, since up until the previous commit, which changed the behavior so transactions eagerly release locks, there was no functional difference caused by lazily timing out the transaction. Release note: None ### sql: add test for implicit txn holding locks across retries Release note: None --- fixes #140234 141815: systemconfigwatcher: remove NewWithAdditionalProvider r=rafiss a=rafiss This was needed in the 21.2-22.1 mixed version state, before tenants had a spanconfig reconciliation job. Now that the job exists, there's no need to pass in an additional provider for tenants. This takes the work in #82329 a bit further. informs #76625 Release note: None Co-authored-by: Rafi Shamim <rafi@cockroachlabs.com>
Release note: None