fix(serializing): reuse larger pooled writers - #1054
Merged
FirstGearGames merged 1 commit intoAug 9, 2026
Merged
FirstGearGames merged 1 commit into
FirstGearGames merged 1 commit into
Conversation
Length-pooled writers can be stored under a larger capacity bucket than the requested length bucket. Reusing the nearest larger bucket prevents pooled writers from staying rooted in the static length pool under variable payload sizes.
sqdbruh
marked this pull request as ready for review
June 25, 2026 12:29
Owner
|
Good find, thank you! |
FirstGearGames
pushed a commit
that referenced
this pull request
Sep 2, 2026
- Fixed ZigZagDecode improperly decoding values in IL2CPP with specific backing CPP libraries. - Fixed Scene object expecting prefab ID (#1032). - Fixed eliminated or significantly reduced wobble in local reconcile resimulations on server-authoritative objects. - Fixed NetworkAnimator delay/desync on host owned object, observed by clients (#1062). - Fixed NetworkCollider/2D tick comparison error (#1041). - Fixed prefab objects generator cannot include folders with spaces in the path (#1045). - Fixed large writers pooling incorrectly, causing them to not be re-used (#1054). - Fixed LiteNetLib wrong data length after BasePacketLayer modification (#1064). - Fixed NetworkTrigger2D by uncommenting entire file. - Fixed nested NetworkObjects not spawning recursively when spawned via OnStartServer of their parent. - Fixed SyncType values being reset to default before OnStopServer as clientHost (#1036). - Fixed disconnecting clients not being cleared on Transport shutdown (#1044). - Fixed Quaternion ScaleToFloat helper math, cause of marginal rotational errors (#1067). - Fixed loading scenes by path not loading the scene for clients under certain conditions (#1060). - Obsoleted SceneLookupData.Name in favor of SceneLookupData.FullName. - Added NetworkBehaviour.ToString null safety (#1046). - Added Unity 6.5 scene handle change support (#1052 / #1051 merge). - Added Unity 6.5 GetEntityId/GetInstanceId support. - Added NetworkCollider2D use PhysicsScene2D.OverlapCollider as a fallback (#1042). - Added NetworkCollider/2D.RemoveEnteredCollider. - Added NetworkCollider/2D InvokeStayOnEnter. - Added NetworkAnimator channel option - under Beta menu, Animator Channel (#1039).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This changes
WriterPool.Retrieve(length)so it can reuse a writer from the closest larger length bucket when the exact bucket is empty.The exact-bucket path stays the same. The fallback only runs when there is no writer available for the requested bucket, and empty buckets are removed after popping.
Background
I ran into this on a long-running Minecraft-like server with chunk streaming and large, variable payload sizes. Each player reconnect caused FishNet to retain some memory even after all player-owned resources were cleaned up. After enough reconnects, memory usage kept climbing until the server eventually ran out of RAM.
The issue is that a writer can be returned to the length pool under a bucket based on its actual capacity, which may be larger than the bucket requested later. Since
Retrieve(length)only checked the exact bucket, larger cached writers could sit in_lengthPooland remain strongly referenced instead of being reused.Reusing the nearest larger bucket fixed the retained allocation growth in that workload and kept memory usage stable across repeated reconnects.