feat(discovery)!: spread joins across lobbies instead of packing one - #71
Merged
Conversation
Joins now go to the least-occupied lobby, always. The previous policy packed the fullest lobby below a soft cap so that a network of 50 would feel like one lobby of 50 rather than five of ten — a real concern, but the effect was that every player in a region sat in one process, and the cap defaulted to 400, which no region ever reached. Measured on stage, the lobby was never the reason to split: 228 players cost 0.29 cores and a 2 ms average tick against a 50 ms budget. The reason is blast radius — one restart took every player in the region with it. That argues for spreading always, not for a threshold nobody reaches. A newly autoscaled lobby needs no special case: it is empty, so it is the least occupied, so it takes joins until it has caught up. Priority filling falls out of the same rule that does the spreading. BREAKING CHANGE: GROUNDS_AGONES_LOBBY_SOFT_CAP is gone. Deployments setting it keep working, the value is simply ignored; groundsgg/deploy should drop it.
|
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Joins go to the least-occupied lobby, always.
Why
The previous policy packed the fullest lobby below a soft cap, with a good
reason stated in the code: a network of 50 should feel like one lobby of 50,
not five of ten. In practice the cap defaulted to 400 and no region ever came
close, so every player in a region sat in a single process.
Measured on stage under load, the lobby was never the reason to split:
The reason to split is blast radius, not load — one restart took the whole
region's players with it. That argues for spreading always rather than for a
threshold nobody reaches.
Autoscaling falls out of the same rule
A lobby added by the FleetAutoscaler is empty, so it is the least occupied, so
it takes joins until it has caught up. No special case for "new lobby".
Verified on stage before this change, using the soft cap as a stand-in: with
the cap lowered to 150, joins moved to the second lobby (152/48) and the Fleet
scaled 2 → 3 on its own. This change makes that the default behaviour without a
threshold.
Breaking
GROUNDS_AGONES_LOBBY_SOFT_CAPis removed. Deployments that set it keepworking — the value is ignored — but
groundsgg/deployshould drop it, sinceit was only just added for the experiment above.
LobbyPackingbecomesLobbySelection; the name was describing the old policy.Trade-off worth naming
This reverses a deliberate product decision. With the Buffer FleetAutoscaler
keeping one Ready lobby spare, always-spreading means that spare starts taking
players immediately, which allocates it, which makes the autoscaler add
another — so a region will tend to sit at
maxReplicas(4 today) rather thanat the minimum. That is more small lobbies rather than one big one, which is
exactly the intent, but it is a visible change to how the network feels.