Closed realFlowControl closed 3 weeks ago
Benchmark execution time: 2024-06-10 11:38:05
Comparing candidate commit 9e04353bc99a9baa3077645f8adca48317b11a37 in PR branch florian/thread-name-id
with baseline commit e63625078edc52eb92e4c6104f0daaab3ec8fa63 in branch master
.
Found 0 performance improvements and 0 performance regressions! Performance is the same for 27 metrics, 9 unstable metrics.
All modified and coverable lines are covered by tests :white_check_mark:
Project coverage is 79.32%. Comparing base (
e636250
) to head (be0af97
). Report is 2 commits behind head on master.
@morrisonlevi I can not approve but I approve
Description
Turns out the profiling-backend assumes that all pprof based profiles emit a
thread id
andthread name
label with every sample, so this PR aims to do this. Edit: profiling-backend was updated to not require the thread name, but we should add at leastthread id
anyway. Since thread names are uncommon (we don't know of a single PHP program/library which names threads), we omitthread name
for now.Took this as an opportunity to clean up a few things:
message_labels
was renamed tocommon_labels
and now sets the "fiber" tag (was inprepare_message
before). It also accepts a parameter for how much extra space to allocate in theVec
, so callers can avoid reallocations.lazy_static
labels, and removed un-necessaryvec!
usages to reduce allocations.send_message(prepare_message(...))
, so I made a helper for this calledprepare_and_send_message
.PROF-9887
Reviewer checklist