heap-snapshot-analysis

Parse the .heapsnapshot file with a Buffer-based parser, group nodes by constructor, and compare against a baseline snapshot. Then trace retainer paths from suspicious instances to GC roots using non-weak reverse edges to find what prevents collection.

Load both snapshots and group objects by type and constructor name, then compute count and size deltas per group. The comparison reports top groups by size increase, count increase, and object groups that only exist in the after snapshot.

Heap snapshots often exceed 500MB, surpassing V8's maximum string length, so JSON.parse throws. Buffer-based section extraction reads the meta, nodes, edges, and strings arrays directly from the file buffer without creating one giant string.

WeakMap keys show up as weak edges from key to backing array in the snapshot graph, but weak references do not prevent garbage collection. Retainer analysis must skip weak edges, otherwise it reports false retainers that are red herrings.

Run analysis scripts with node --max-old-space-size=16384 to give the process enough heap for multi-hundred-megabyte snapshots. Typical analysis takes 30 to 120 seconds per snapshot depending on file size.