Automatic heap dump on OOM versus manual on-demand dump for intermittent UI slowdown under a fixed heap limit

Answer

Keep automatic heap‑dump on OutOfMemoryError disabled in a regular development workflow. Enable it only temporarily when you suspect the slowdown is caused by actual heap exhaustion, and pair it with a retention policy or temporary filesystem to limit disk use and protect sensitive data.

Likely explanation

UI slowdown often appears before the heap is exhausted, so an automatic dump will not fire unless an OutOfMemoryError actually occurs. Manual dumps can capture the pre‑OOM state but require you to trigger them during the symptom window.

Confirmed facts

• With a fixed -Xmx limit, automatic dumps generate a file each time the limit is exceeded, causing steady disk growth and potential exposure of source code or credentials.
• The manual dump action (Help → Diagnostic Tools → Dump Memory Heap) location varies between IDE versions, so users must relearn the path after upgrades.
• If the slowdown stems from UI‑thread blocking, plugin deadlocks, or GC pauses unrelated to heap size, a heap dump (automatic or manual) provides limited diagnostic value.

Steps for this case

1. Check idea.log for OutOfMemoryError entries around the time of slowdown. If none are present, an automatic dump will not have been created.
2. If you can reliably reproduce the slowdown, use the manual dump action at the onset of the symptom to capture the heap state.
3. If reproduction is unreliable, consider enabling automatic dump temporarily with a short‑lived dump directory (e.g., a tmpfs mount) and a cleanup policy that removes files older than a few hours.
4. After collecting dumps, analyze them with Eclipse MAT or jhat to look for large retained sets, thread stacks, or GC patterns.

Missing diagnostic detail

Can you reliably reproduce the UI slowdown so that you can trigger a manual heap dump during the symptom window? If the answer is no, the automatic dump (with retention controls) becomes the more reliable option despite its disk and privacy trade‑offs.