Perf/dont redownload downloaded code #6873
Conversation
| @@ -29,6 +30,9 @@ public class SnapProvider : ISnapProvider | |||
|
|
|||
| private readonly ProgressTracker _progressTracker; | |||
|
|
|||
| // This is actually close to 97% effective. | |||
| private readonly LruKeyCache<ValueHash256> _codeExistKeyCache = new(1024 * 16, ""); | |||
There was a problem hiding this comment.
512kb of Keccaks; does/can this get dumped after sync?
There was a problem hiding this comment.
I think we can drop/clean it after the sync finishes.
There was a problem hiding this comment.
We found lately with @obasekiosa that acquiring lock in LRU can take quite some time in very congested scenarios. Would be good to create LRU that can avoid waiting for locks or maybe avoid LRU altogether and use NonBlocking.ConcurrentDictionary as hash?
I assume shouldn't be that contended here? |
This is in parallel: https://github.com/NethermindEth/nethermind/pull/6873/files#diff-59cbf3228de3b06fc97ff45e63cc32f17a5194574fe42e34ccf863515adb1b17R96-R103 and could produce high lock contention. It is fine for now - not blocking this PR, I will try to propose a nonblocking LRU cache :) |
| @@ -205,6 +205,7 @@ public override void SyncModeSelectorOnChanged(SyncMode current) | |||
| if (_disposed) return; | |||
| if (CurrentState == SyncFeedState.Dormant) | |||
| { | |||
| _snapProvider.Dispose(); | |||
There was a problem hiding this comment.
won't this be called over and over again?
There was a problem hiding this comment.
Ow, I think I made a mistake here.
Changes
Types of changes
What types of changes does your code introduce?
Testing
Requires testing
If yes, did you write tests?