:TX_MEDIATOR_TIMECAST CRIT: Actor# [...] got update from Mediator# 72075186224037894 with LatestStep# 0 previous LatestStep# 1739943505740

[snaury]

I often see messages like these under nemesis:

:TX_MEDIATOR_TIMECAST CRIT: Actor# [50004:7473000454368397226:2064] got update from Mediator# 72075186224037894 with LatestStep# 0 previous LatestStep# 1739943505740

This doesn't actually look like a problem, probably a restarted mediator reporting LatestStep# 0? Doesn't look like something that needs a CRIT log level.

[snaury]

Actually it looks like some strange problem, since we're supposed to receive events with ever increasing SubscriptionId and we're filtering old messages. We also set WatchSynced = false every time we resubscribe and we don't set WatchSynced = true until latest step matches what we have in memory. Since mediator sends its AcceptedStep as latest step it's not supposed to go backwards ever, so a message like this is totally unexpected.

[snaury]

Oh, we don't set WatchSynced = true when subscribing to bucket after an initial reconnect. So we could:

• Have all tablets in a bucket to unregister
• Lose connection with mediator due to its restart
• Reconnect without subscribing to the bucket above
• A new tablet starting very quickly and subscribing to bucket

We would erroneously think the bucket is synced (based on the previous connection) when in reality it is not. The code still treats it the same as unsynchronized case, but this spams with an unwanted CRIT in the log, need to fix.