fix(vexa-bot): browser-session entrypoint exits when node exits (#258)

[DmitriyG228]

Summary

• Fixes browser-session pod stays Running after node process exits — entrypoint keeps container alive forever for VNC #258: wait in the browser-session entrypoint blocks forever on x11vnc -forever and websockify, so the container never exits after node dist/docker.js returns. K8s sees the pod as Running indefinitely and "stop" from the dashboard appears to do nothing.
• New env var VEXA_BROWSER_SESSION_VNC_KEEPALIVE_SECONDS (default 0) controls a bounded post-exit debug window. Default behavior: container exits as soon as node does.
• Preserves the original exit code so K8s sees 0 for clean exits and ≠0 for crashes.

Diff

services/vexa-bot/core/entrypoint.sh (browser-session branch only — meeting-mode branch unchanged):

node dist/docker.js
EXIT_CODE=$?
KEEPALIVE="${VEXA_BROWSER_SESSION_VNC_KEEPALIVE_SECONDS:-0}"
if [[ "$KEEPALIVE" =~ ^[0-9]+$ ]] && [ "$KEEPALIVE" -gt 0 ]; then
  echo "[entrypoint] node dist/docker.js exited with code $EXIT_CODE — keeping container alive ${KEEPALIVE}s for VNC access"
  sleep "$KEEPALIVE"
else
  echo "[entrypoint] node dist/docker.js exited with code $EXIT_CODE — exiting (set VEXA_BROWSER_SESSION_VNC_KEEPALIVE_SECONDS>0 to debug)"
fi
exit "$EXIT_CODE"

Implementation note

The issue body sketched timeout $N wait — that doesn't actually work because wait is a shell builtin and timeout only runs binaries. sleep $N achieves the equivalent bounded keep-alive: the parent shell sleeps, then exits, and the kernel cleans up the still-running x11vnc / websockify / sshd children.

Non-numeric values fall through to the default branch ([[ ... =~ ^[0-9]+$ ]] guard) so a typo can't crash the script.

How to opt back into the post-exit debug window

In services/runtime-api/runtime_api/profiles.py profile YAML:

browser-session:
  image: ...
  env:
    VEXA_BROWSER_SESSION_VNC_KEEPALIVE_SECONDS: "3600"

profiles.py already passes the env block straight into the container — no code change needed there.

Test plan

• bash -n services/vexa-bot/core/entrypoint.sh (syntax)
• Smoke-test the three branches (default-0, valid N, non-numeric) — verified the if/else picks the right branch and EXIT_CODE is preserved
• Build the runtime image and exercise a real browser session: confirm pod terminates within seconds of bot stop on staging
• Set VEXA_BROWSER_SESSION_VNC_KEEPALIVE_SECONDS=60 in a profile, exercise stop, confirm container stays for ~60s then exits with the original exit code

🤖 Generated with Claude Code

[DmitriyG228]

Planned for merge in the current hardening release (cycle 260424); closes #258.