v1.2.3.b8 - fix: perf percentiles and duplicate entity log lines
- #### Debug
- **PerfLog** - `p50`, `p95` and `p99` returned the upper bound of the histogram bucket instead of a real sample, so they could report a value higher than `max`. It happened in 74% of the windows in a 6.5 hour session (e.g. `avg=0.06ms p99=8.00ms max=4.68ms`). Percentiles are now taken from the actual samples of the window, and the bucket estimate is only used as a fallback for windows that overflow the buffer - interpolated inside the bucket and clamped to `max`.
- **EntityManager** - An entity that was already tracked logged a second `+` line, so creates and destroys stopped matching and the entity log read like a leak. Re-registering now logs `~ oldType -> newType` when the type changes and nothing when it does not.
- **Ghost, Ninja, C4Camouflage** - The hiding prop was registered twice, first as `prop_dynamic` and then as `empty_prop`. It is now tracked under the right name from creation through the new `trackAs` parameter of `CreateTrackedDynamicProp`.
This commit is contained in:
parent
b95c52028e
commit
68d311e130
8 changed files with 59 additions and 16 deletions
1
.gitignore
vendored
1
.gitignore
vendored
|
|
@ -30,6 +30,7 @@ project.fragment.lock.json
|
|||
*.userosscache
|
||||
|
||||
# Logs, caches, temp
|
||||
log/
|
||||
*.log
|
||||
debug_*.txt
|
||||
perf_*.txt
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue