mo-service profiling flags

This page describes main/latest. Defaults can differ in older releases; check mo-service --help on the target binary. This page does not infer the first release containing these flags.

Flag

Current default

Meaning

--block-profile-rate

0

Calls runtime.SetBlockProfileRate(rate): on average, one event per rate nanoseconds spent blocked; 1 records every blocking event and 0 disables collection. This is not one per rate events.

--mutex-profile-fraction

0

Calls runtime.SetMutexProfileFraction(rate): on average, one in rate mutex contention events; 1 records every event and 0 disables collection.

The units differ. Do not directly compare their values or treat 100 as a universal production setting. Dense sampling adds overhead; use bounded diagnostic windows and measure workload impact.

Enable and collect

Use your deployment’s configuration. The CN path below exists in the source launch configuration. Bind debug HTTP to loopback; use controlled forwarding for remote collection.

mo-service --cfg etc/launch/cn.toml   --debug-http 127.0.0.1:12345   --block-profile-rate 1 --mutex-profile-fraction 1

The CN configuration must point to ready TN/LogService nodes; this does not deploy a complete cluster. Analyze collected profiles with the matching binary:

curl -fsS http://127.0.0.1:12345/debug/pprof/block -o block.pprof
curl -fsS http://127.0.0.1:12345/debug/pprof/mutex -o mutex.pprof
go tool pprof -top ./mo-service block.pprof
go tool pprof -top ./mo-service mutex.pprof

HTTP 200 does not prove contention samples exist. Check sample type, totals and hot stacks in pprof. Values across versions or sampling rates are not directly comparable. These are startup flags; restore the original sampling configuration through your normal restart procedure after diagnosis.