delve-debugger

name	delve-debugger
description	Practical playbook for debugging the talos-cluster provider with Delve using the background terminal and pdebug to capture panics.

Overview

• Goal: attach to the remote Delve server, trigger pdebug.sh <command> to reproduce a Pulumi run, and collect backtraces/locals without dropping the session.
• Use the background terminal for dlv attachment; orchestrator can trigger pdebug.sh via MCP from the target program directory.

Environment

• Use the local dlv client; run commands from repo root. Always create an init file under /tmp before dlv connect to avoid --init missing errors.
• Delve server: pulumi-talos-cluster-runner-delve.pulumi-talos-cluster-workbench:2345.
• If prompted “Would you like to kill the headless instance?”, answer n to keep it running.
• Breakpoints: do not use break; add runtime.Breakpoint() in code and rebuild if a breakpoint is required.

Quick Attach (background terminal)

1. Prepare init file (rebuild + continue):

printf 'rebuild\ncontinue\n' > /tmp/dlv_cmds_full

2. Attach in a TTY background terminal:

dlv connect pulumi-talos-cluster-runner-delve.pulumi-talos-cluster-workbench:2345 --init=/tmp/dlv_cmds_full

3. Stay attached; do not quit. If Unknown process id appears, recreate the init file and reconnect.

Catching a Panic

1. Attach and leave Delve waiting (do not quit) using the quick-attach steps.
2. From the target program directory, trigger the provider via MCP:

/spigell-reforge-ai/deploy/workbench/pdebug.sh pre

Example directory: integration-tests/testdata/programs/<program>. 3) When Delve stops on panic, collect:

bt
goroutines
goroutine <id> bt   # crashing goroutine
goroutine <id> locals
goroutine <id> args

4. Optional inspections:

print <expr>
whatis <expr>

5. Resume after inspection:

continue

Debugging an Active State (background terminal)

• If the process is already running and you need to inspect its current state:
  1. Create an empty init file: printf '' > /tmp/dlv_cmds_capture.
  2. Attach in a TTY background terminal:
     dlv connect <server> --init=/tmp/dlv_cmds_capture
  3. Inspect the state without restarting:
     • threads (or goroutines) to find the interesting goroutine.
     • bt or goroutine <id> bt for stack.
     • goroutine <id> locals / args or print <expr> for values.
  4. Use continue only when you’re ready to let the process run again.
  5. If the session drops (Unknown process id), recreate the init file and reconnect.

Common Symptoms & Actions

• plugin ... did not begin responding to RPC connections: ensure the dlv session issued rebuild/continue, then rerun pdebug.sh.
• EOF / resource monitor shut down: reattach with the init file and rerun pdebug.sh.
• --init file missing: create the file under /tmp before connecting.