Ratekeeper throttling aggressively when unable to fetch storage server list #1858
Conversation
…ge server list for some configurable amount of time.
fdbserver/Ratekeeper.actor.cpp
Outdated
| totalDiskUsageBytes += s.value.lastReply.storageBytes.used; | ||
|
|
||
| if (now() - self->lastSSListFetchedTimestamp > SERVER_KNOBS->STORAGE_SERVER_LIST_FETCH_TIMEOUT) { | ||
| limits->tpsLimit = 1; |
There was a problem hiding this comment.
If the tpsLimit happens to be less than 1, we probably shouldn't increase it or change the reason, even if we can't fetch the storage server list.
Also, is there a particular reason 1 was chosen instead of 0 as the limit?
There was a problem hiding this comment.
No particular reason. Just a small value. Should i change it to 0?
There was a problem hiding this comment.
My gut says there isn't a lot to be gained by allowing 1 vs 0, but that the slow trickle of transactions could in some cases cause additional damage. I'd be inclined to clamp it all the way down in this case and only allow immediate priority transactions. You could also potentially check with @etschannen for another opinion.
There was a problem hiding this comment.
I agree that it should be set to 0
3 participants
Fixes #1642