Hi folks,
We have an application that we are profiling. It's not getting anywhere near the throughput we would expect. As I look at this flame graph, it seems to be spending the bulk of the time in LuaJIT and lpeg (lua parser). This doesn't match expectations because (1) the data was collected over a 5 minute load test; (2) we have the lua_code_cache on.
So I'm thinking we either screwed something up in the config, or I'm completely misreading the flame graph.
I appreciate any tips this community may offer. Thanks!
-Scott
NOTE: perf data was captured using ...
perf record -F 99 -a -g -- sleep 300
perf script | ./stackcollapse-perf.pl > out.perf-folded
./flamegraph.pl out.perf-folded > perf-kernel-nologging.svg
Attachment:
perf-kernel-nologging.svg
Description: image/svg