Adjust stackprof sampling intervals

- sample CPU at 99Hz instead of 100Hz to avoid lock-stepping
- sample objects at 100 (1000 was much too coarse)

config/initializers/stackprof.rb

@@ -9,10 +9,10 @@ module Gitlab
     DEFAULT_FILE_PREFIX = Dir.tmpdir
     DEFAULT_TIMEOUT_SEC = 30
     DEFAULT_MODE = :cpu
-    # Sample interval as a frequency in microseconds (~100hz); appropriate for CPU profiles
-    DEFAULT_INTERVAL_US = 10_000
+    # Sample interval as a frequency in microseconds (~99hz); appropriate for CPU profiles
+    DEFAULT_INTERVAL_US = 10_100
     # Sample interval in event occurrences (n = every nth event); appropriate for allocation profiles
-    DEFAULT_INTERVAL_EVENTS = 1_000
+    DEFAULT_INTERVAL_EVENTS = 100
 
     # this is a workaround for sidekiq, which defines its own SIGUSR2 handler.
     # by defering to the sidekiq startup event, we get to set up our own

doc/development/performance.md

@@ -257,8 +257,8 @@ The following configuration options can be configured:
   Defaults to `cpu`.
 - `STACKPROF_INTERVAL`: Sampling interval. Unit semantics depend on `STACKPROF_MODE`.
   For `object` mode this is a per-event interval (every `nth` event is sampled)
-  and defaults to `1000`.
-  For other modes such as `cpu` this is a frequency and defaults to `10000` μs (100hz).
+  and defaults to `100`.
+  For other modes such as `cpu` this is a frequency interval and defaults to `10100` μs (99hz).
 - `STACKPROF_FILE_PREFIX`: File path prefix where profiles are stored. Defaults
   to `$TMPDIR` (often corresponds to `/tmp`).
 - `STACKPROF_TIMEOUT_S`: Profiling timeout in seconds. Profiling will