Streamline AsyncShardFetch#getNumberOfInFlightFetches #96545
Conversation
Avoids an O(#nodes) iteration by tracking the number of fetches directly. Backport of elastic#93632 to 7.17
DaveCTurner
added
>bug
:Distributed/Allocation
|
This backports cleanly and is a good improvement in large clusters. We've seen no related trouble since it was introduced in 8.5 so I see no good reason not to backport it, but just checking with @henningandersen before I merge it. |
Correction, sorry, this was only introduced in 8.8 (I was thinking of #96546). But IMO the assertions in this area are good and we haven't seen any related test failures since it was merged. |
| @@ -131,7 +128,7 @@ public synchronized FetchResult<T> fetchData(DiscoveryNodes nodes, Set<String> i | |||
| } | |||
|
|
|||
There was a problem hiding this comment.
nit: ideally we'd also assert consistency between fetchingCount and cache here (or inside the brace one line up).
| assert assertFetchingCountConsistent(); |
There was a problem hiding this comment.
++ thanks, yes that'd make sense. I'll do that in a follow up (and apply it in main too)
Avoids an O(#nodes) iteration by tracking the number of fetches directly.
Backport of #93632 to 7.17