Tackling Memory Issues on the Instagram Android App
summary
Instagram engineers describe a production pipeline that automatically uploads sanitized HPROF dumps via Omura, analyzes them with Eclipse MAT to extract paths‑to‑GC‑root and aggregate leaks, and wires results into dashboards and crashbot tasks; they define empirical memory red thresholds (Java heap, system, address space), run end‑to‑end leak tests that bisect and revert bad commits, and deploy a memory‑red app destroyer. The system catches problems before users see them and cuts the ~44,000 OOMs/day (51% on low‑end devices).
AI-written summary · about
chapters (24)
- 0:00Memory issues at Instagram
- 1:06Memory management philosophy
- 3:11Prioritizing memory crashes
- 6:45Memory red thresholds
- 9:13Memory red zones impact
- 11:41Monitoring memory issues
- 13:48FAQ on memory thresholds
- 15:56Detecting leaks in production
- 18:32Handling leaks at scale
- 20:25Heap dumps with Omura
- 22:39Heap dump triggers and privacy
- 24:29Analyzing heap dumps
- 25:31Finding leaks and large objects
- 27:47Reports for leaks
- 29:16Assigning leak fixes
- 30:44End-to-end leak testing
- 32:31Benefits of leak tests
- 33:56Fixing leaks in practice
- 35:02Leaks and crashes correlation
- 36:11Common Android leak patterns
- 39:01Debugging memory leaks
- 40:44Bitmap handling best practices
- 41:48Why leak fixes matter
- 44:05Q&A on memory zones
related talks
53mMemory on Android, Leaks and OOM
43mMemory Dumps: Going beyond the memory leak
07mAndroid memory tools and best practices
40m640K ought to be enough for anyone
38mOptimize your app performance for the next generation of Android- 38mOptimize your app performance for the next generation of Android
37mRoot causing memory growth with LeakCanary 3
25mMemory leaks on Android: lessons learned and methodology