core-dump-node-lldb-llnode.md

Node postmortem debugging (OSX)

1. Taking core dump

To get core dump of running node process:

• Take PID of that process:

  pgrep -lf node

• Run lldb debugger with attached process:

  lldb -p <PID>

• Take a dump of process data:

  (lldb) process save-core <FILE_NAME>

Alternative: If you have signed (on OSX only) gdb with gcore you can run simply gcore <PID>

2. Extend lldb with v8 debugger support

• Fetch the repository with lldb extension. (It will be used later).

  git clone git@github.com:nodejs/llnode.git

• Install llnode

  brew install llnode

• Install python with pip if you don't have it already.

  brew install python

• Install python six module

  pip install six

• Link llnode with lldb

  mkdir -p ~/Library/Application\ Support/LLDB/PlugIns
  ln -sf /usr/local/opt/llnode/llnode.dylib ~/Library/Application\ Support/LLDB/PlugIns/

3. Debug

In order to have full functionallity we need to have generated memory ranges file for each core dump that we want to investigate.

• To do so we need to run from previously fetched llnode repo(this is not needed for lldb version 3.9 - nodejs/llnode#91):

  ./scripts/otool2segments.py <GENERATED_CORE_DUMP> > <RANGES_FILE>

• Then just export that file in environment variable (this is not needed for lldb version 3.9).

  export LLNODE_RANGESFILE=<RANGES_FILE>

• Run lldb with our node dump - we need to use the exact node binary as the dump was created

  lldb node -c <GENERATED_CORE_DUMP>

• Check if everything works:

  (lldb) help v8

Some docs for v8 commands are available here: https://developer.ibm.com/node/2016/09/27/advances-in-core-dump-debugging-for-node-js/