It is often the case that we need to get a heap snapshot from running zero-cache processes.
Online guides and documentation suggest using the chrome-remote-interface package and then calling v8.writeHeapSnapshot().
Doing this for zero-cache yields much smaller snapshots than those you get if you connect to the node process in Chrome's devtools and get a snapshot that way. These smaller snapshots lack many core classes of zero-cache such as ViewSyncer.
I do not understand why but ChatGPT suggests that the implementation of v8.writeHeapSnapshot() is different and simpler than the HeapProfiler class that Chrome itself uses, and misses some "uncommon" allocation patterns 😬.
This tool automates using the lower-level HeapProfiler class to create a dump that contains all of zero-cache's objects.
kill -USR1 <pid-to-snapshot>- The targetted process prints out a message like:
Debugger listening on ws://127.0.0.1:9229/a8712980-d35d-434f-824c-db212b277b5e npx @rocicorp/heap-snapshot <that-url>