pg,abci: unblock notice subscribers on DB shutdown #1021
Conversation
| // will still be deterministic so nbd to not halt here | ||
| a.log.Errorf("failed to parse notice (%.20s...): %v", log, err) | ||
| continue // since txid is invalid and won't match any result.TxHash |
There was a problem hiding this comment.
Pls double check me on this continue @brennanjl . I tend to agree it not necessary to halt, but I don't think we should attempt to use the txid (which is empty or generally invalid) as a map key below, just continue to receive in the next iteration of the loop
There was a problem hiding this comment.
Yeah this seems correct.
| // will still be deterministic so nbd to not halt here | ||
| a.log.Errorf("failed to parse notice (%.20s...): %v", log, err) | ||
| continue // since txid is invalid and won't match any result.TxHash |
There was a problem hiding this comment.
Yeah this seems correct.
| wg.Wait() | ||
| // wait for all logs to be received, or a premature shutdown | ||
| select { | ||
| case <-ctx.Done(): |
There was a problem hiding this comment.
Do we need to worry about this not being supported in comet v0.38? I don't think so because we close the notice subscribers when the database shuts down, but just want to double check that you see it the same
There was a problem hiding this comment.
No it's ok. With 0.38 we only have one way to break out of here, and that's the notice signaling that the DB dies. With 1.0 we can also get a signal from cometbft, but that's just going to be a bonus.
jchappelow
force-pushed
the
unblock-logs-shutdown
branch
from
e205a27
to
cd9a85f
Compare
jchappelow
force-pushed
the
unblock-logs-shutdown
branch
from
cd9a85f
to
a6ee3b5
Compare
|
gofmt'd |
|
Resolved conflicts. |
This resolves a possible hang in
FinalizeBlockif the DB shuts down / dies while waiting for the completion of the notice stream atwg.Wait(). When thepg.DBinstance closes up in response to the replication connection terminating, we also make sure to close the subscriber channels to unblock the receivers. To ensure the receivers handle this as an error rather than assuming all notices have been received successfully, an empty string is first sent on the channel. Since all notice messages have a special prefix (pgtx:), this is easily recognized as an exception.