1,945 Commits over 488 Days - 0.17cph!
Merge: from baseplayer_serverupdateparallel
- Optim for OcclusionV2 - remove extra hash lookup in a busy loop
Tests: with UsePlayerUpdateJobs 2 - toggled invis in close proximity and far away, suicided away from spawn and flew back, teleported around
Optim: OcclusionV2 - skip resetting lastPlayerVisibility every frame
Turns out didn't need to make the other half of changes
Tests: used invis close and far away
Update: UseOcclusionV2 - simplify toggling invis
- when enabling invis, notify occlusion tracking that we destroyed entity on subscriber's end
- when disabling invis, skip sending snapshots - occlusion will handle it
Halfway to remove the constant hashSet.Remove calls in GatherPairs loop
Tests: UsePlayerUpdateJobs 2, toggled invis in front of player. went invis, flew outside network range, flew back, disabled invis - spawned correctly.
Merge: from baseplayer_serverupdateparallel
- Bugfix for UsePlayerUpdateJobs 2 not replicating players for spectators
- Bugfix for leaking players in occlusion groups
Tests: spectated players from far away. Inspected server state after killing a sleeper
Bugfix: ServerOcclusion - clean up players from all of their occlusion groups on server destroy
Turns out we were leaving a lot of "zombie" players once sleepers were killed. This would slow down server occlusion eventually.
Tests: on Craggy, killed a sleeper, moved to a different network group, and checked in debugger if the occlusion group only contains my player
Bugfix: UseOcclusionV2 - handle spectating players
Since we no longer do bidirectional checks when iterating occlusion groups, we ended up missing out on spectators and wouldn't allow replication to them.
Tests: on craggy with UsePlayerUpdateJobs 2, flew far away to get outside of network range then started spectating - could see target moving.
Merge: from spectate_dontfloodsnapshots
- Bugfix for occlusion not destroying all players in UsePlayerUpdateJobs 2 mode
Tests: 2p on craggy - used hill as obstruction, checked entity lists on both clients in editor
Bugfix: UsePlayerTasks - calculate end of batch correctly in SendEntityDestroyMessages_AsyncState
This was causing server occlusion to skip sending some of destroy commands (from last batch), leaving a phantom player
Tests: on Craggy went on behind hill, then checked editor entity list - confirmed that replicated players got removed on both ends
Merge: from spectate_dontfloodsnapshots
- Fixes spectate not working in UsePlayerUpdateJobs 2 case
Tests: tried to spectate occluded player
Bugfix: Ensure initial player snapshot gets sent to spectator in UsePlayerUpdateJobs 2 case
Tests: on craggy with UsePlayerUpdateJobs 2, took 2nd player behind a hill to occlude, then tried to spectate them - confirmed that the first snapshot got sent.
Clean: post-submit feedback
- rename entity_prefabId field to entity_prefab_id
- rename AddShortEntity to AddShortEntityField
More consistent with other examples
Tests: compiles in editor
Merge: from analytics_entity_createdestroy
- minor code clean
Tests: editor compiles
Merge: from analytics_entity_createdestroy
- we now collect entity spawn and destroy events once the server has started up (part of `gameplay_analytics` servervar)
Tests: observed logs of analytics in CLIENT+SERVER editor
Clean: fix formatting in Analytics.Azure.cs
Most of file was offset with 2 tabs. Sorry to anyone who has conflicting changes in same file (hopefully no-one)
Tests: editor compiles
Update: Analytics - track all entity spawn/destroy events after server initializes
- change the format of these events to only send vital data
- HackableLockedCrate no longer has it's own EntityCreated event
Tests: logged analytics on Craggy - saw events being appended after server initialized
Merge: from server_occlusion_poppin
- Reduce player poppin by making occlusion grid generation more conservative
- Bumped occlusion version, will cause regeneration on next server boot
Tests: visualized in editor, ran previously failed queries
Update: ServerOcclusion - only block occlusion cell if it's fully under terrain
- bumped occlusion file version
Previously we considered it was blocked if one of sample points was in terrain. This allows more occlusion queries to pass, which should reduce poppin.
Tests: visualized new grid, ran previously failing query - they now pass
Merge: from hide_getinternalarrayunsafe
- Replaces dangerous API with a safer, updates existing use cases
Tests: unit tests
Update: replace invalid usage of GetInternalArrayUnsafe
Tests: none, trivial change
Update: Hide List.GetInternalArrayUnsafe
- introduce ListAsReadOnlySpanOf to help deal with casting up (since we don't have ReadOnlySpan<T>.CastUp<U>)
- added simple unit tests
Tests: ran unit tests
Merge: from analytics_v2_expose_telem
- Fixes CLient crash when running analytics.tablestats or uploaderstats
Tests: built standalone and ran the commands - no crash
▊▉▆█▋▌: ▇█▍▊ ▌▋▉▋▍▅▆▍▇▊▌▊▌▇▇▊▋▌▉▋▇▋ ▊█ ▅▉▍██▉▋▉▄
█▌▉▅▊▍▍▆▆▌▋ ▋▄▊▌▅ ▄▄█▌▅▇▄ ▅▅▊▇▆▌, ▉▋▌▆▉ ██▋▉▅ ▍▊█▄▍ ▌▊▄▌▄ ▆▇ █▆ ▊▌▅▌█▍ ▄▉▋█▋▅▌ █▉▆█ █▉▍ ▉▅▋▆▅▆█▉█▅ ▉▅▇▉ ▊▊▆▊▆▉▄▇▍.
▄▆▌▄▆: ▉▊▊▌▌ █▋█▅▌▋▆▉▉▄ ▉▍▊▋▆▌, ▌▍▌ ▅▍▌▆▆▊▍▇▆.▅▍▌█▊▇▌▌▅▉ - █▍ ▄▄▉▆ ▇▅▊▄▋█▋
Merge: from spectate_dontfloodsnapshots
- reduce potential delay when starting to spectate a far-away player
Tests: flew around 2k procgen map, got to the other side of it then spectated player on the opposite end.
Update: spectator's snapshot queue is cleared when switching between far-away targets
This should speed up switching around on a server. Doesn't help if the network write queue is backed up
Tests: on a 2k procgen map and fast noclip, flew around the island to load up the snapshot queue, then spectated on a player that's on the other side of the island. Observed player moving.
Undo:
135590 - ISubscriberStrategy.GatherSubscriptions outputs whether it has fully or partially gathered subscriptions
After implementing throttling realized there's a better way(invalidate grid sbuscriptions for spectator) - thottling solves only half of problem
Tests: editor compiles
Update: ISubscriberStrategy.GatherSubscriptions outputs whether it has fully or partially gathered subscriptions
- all existing implementations report "full" gathers
This feeds into whether we'll stop running UpdateSubscriptions or not. Will use this to throttle-and-prioritize cell streaming to spectators
Tests: editor compiles
Merge: from networkpositiontick_reduce
- Optim: buoys, containers, water-junkpiles and vines no longer always replicate their position. Reduces number of NetworkPositionTick invokes from 2309 down to 522 on 4.5k map
Tests: interacted with each prefab type, observed up close and from afar
Update: codegen
Tests: editor compiles
Bugfix: profile.CountSyncMoveEntities - check if invoke handlers are not instantiated
- Also add a total count
Tests: used it while swinging from vines
Optim: VineSwingMountable no longer syncs positions by default
Should save another 0.7ms
Tests: Swung across multiple vines in playground_vineswinging, descended on one to ground
Optim: crate_underwater_* - only sync positions once freed
Should save us 0.7ms on 4.5k server
Tests: untied basic and advanced crates on Craggy and observed them raise up
Bugfix: disable Drop To Ground on spawnpoints for junkpile_water_* barrel spawns
Some were incorrectly placed leading to barrel spawning on the ocean floor.
Tests: spawned on craggy, inspected a bunch of junkpile_water_*, none had underwater barrels
Optim: junkpile_water_* - enable client interpolation and use InvokeHandler instead of InvokeHandlerFixedTime
Fixed time handler fires off 10x more by default, and it would generate 3x more position tick messages than InvokeHandler. Will need validating on naval_update
Tests: observed transofrm values of a junkpile in editor while close and at a distance - interpolation didn't throw it anywhere
Buildfix: remove stale debug code
Tests: editor compiles
Optim: OceanWreckBuoy/-Small skips syncing position if there are no nearby players
- set Sync Position to false by default
- marking buoys to always interpolate (need to do the same to junkpiles_water)
Tests: observed transform in editor from craggy and from nearby. Buoys didn't sink/fly off due to continual interpolation (need to double check what's stopping it)
Optim: set junkpile_water_a/b/c Sync Position to false
It is now script controlled, and having it off by default saves allocating an action per entity
Tests: observed transform in editor on craggy at spawn then when close
Optim: junkpile_water_a/b/c replicate positions only if player is in sight
- also switched them to the fixed time invoke handler, since their positionsare ran per fixed update
Tests: booted craggy, validated they weren't moving while I was on island. Flew close, observed them moving and checked in inspector that values change
Update: bring over jarryd's
124139 - BasePlayer.ToggleNetworkPositionTick
Slightly different form - lets us save on the IsInvoking lookups, fixed a bug with fixed time callbacks
Tests: booted on craggy, pushed a car, observed buoys and junkpile_water movement
Add: profile.CountSyncMoveEntities server var
Reports number of "Sync Positions" on the server. Some entities don't get into saves, so this is the only way to see what's putting pressure on InvokeHandler
Tests: ran it in editor on craggy
Bugfix(editor): if invoke throws, don't lose Profiler.EndSample
There's bigger problems out there, but this was making profiling a bit annoying
Tests: loaded playtest save and profiled it in editor
Update: Enable AnalyticsV2 by default
Staging servers ran this for a couple days now and this'll switch over clients as well.
Tests: none, trivial change
Merge: from save_viewer_extras
- Save viewer can now filter for mobile(Sync Position) prefabs
- It can also aggregate by prefab
Tests: opened a save from the playtest
Update(editor): SaveViewer - adjust what we display if we're in aggregate mode or not
Shows count in aggregate mode, otherwise shows ID and Parent ID
Tests: opened a save from playtest
