阅读中文版本

GC Pause Tuning Workflow

Reduce garbage collection pauses by finding allocation hotspots first, then tuning collector settings as a second step.

Overview

Reduce GC pause time by fixing allocation patterns before tuning collector flags.

This workflow is reusable across teams and environments; adapt the commands to your tooling while keeping the sequence and checkpoints intact.

Environment and prerequisites

• GC logging
• An allocation profiler
• A representative load test

Procedure

1. Quantify the pauses

Collect GC logs and quantify pause frequency and duration under load.

Expected state: The pause problem is measured.

2. Find allocation hotspots

Profile allocations to find the code creating most garbage.

Expected state: The main allocation sources are known.

3. Reduce allocations

Fix the hotspots by reusing buffers or avoiding needless object creation.

Expected state: Allocation rate drops measurably.

4. Tune the collector

Adjust heap and collector settings only after allocation fixes, then verify.

Expected state: Pauses fall and throughput holds.

# analyse GC pause distribution
jq -r 'select(.gcId) | .pause' gc.log | sort -n | tail -20

Verification

Confirm pause p99 falls and throughput holds, with allocation rate reduced by the code fixes.

Decision points

Tune flags or fix allocations?

• Fix allocations first
• Tune flags first

Recommendation: Fix allocations first; tuning around high garbage only hides the cause.

Bigger heap or better collection?

• Match heap to live set
• Maximise heap

Recommendation: Size the heap to the live set; an oversized heap increases pause length.

Common failures

• Tuning flags while allocation churns high wastes effort.
• Oversized heaps lengthen pauses.
• No GC logs means guessing at pauses.