Alex Stanfield f5bae45ec9 fix: admin-tools container ignores SIGTERM until kill deadline (#10187) ...

## Summary
- The admin-tools container CMD runs `sleep infinity` in the foreground,
which blocks the shell from processing signal traps
- SIGTERM is never handled, so the container hangs until the kubelet
termination deadline before being force-killed
- Background the sleep and use `wait` as the foreground command instead
-- `wait` is a shell builtin that gets interrupted by signals, allowing
the trap handler to exit immediately

## Test plan

Run the following to compare signal handling before and after:

```bash
#!/usr/bin/env bash
set -euo pipefail

STOP_TIMEOUT=5

echo "=== OLD: sleep infinity in foreground (should hang for ${STOP_TIMEOUT}s) ==="
docker run -d --name admin-tools-old alpine:latest \
  sh -c "trap exit INT HUP TERM; sleep infinity" >/dev/null
sleep 1
echo "Container running. Sending SIGTERM..."
start=$(date +%s)
docker stop --timeout "$STOP_TIMEOUT" admin-tools-old >/dev/null
elapsed=$(( $(date +%s) - start ))
echo "Stopped in ${elapsed}s (expected: ${STOP_TIMEOUT}s -- signal was ignored)"
docker rm admin-tools-old >/dev/null

echo ""
echo "=== NEW: sleep infinity & wait (should exit immediately) ==="
docker run -d --name admin-tools-new alpine:latest \
  sh -c "trap exit INT HUP TERM; sleep infinity & wait" >/dev/null
sleep 1
echo "Container running. Sending SIGTERM..."
start=$(date +%s)
docker stop --timeout "$STOP_TIMEOUT" admin-tools-new >/dev/null
elapsed=$(( $(date +%s) - start ))
echo "Stopped in ${elapsed}s (expected: 0s -- signal was handled)"
docker rm admin-tools-new >/dev/null
```

Expected output:
```
=== OLD: sleep infinity in foreground (should hang for 5s) ===
Container running. Sending SIGTERM...
Stopped in 5s (expected: 5s -- signal was ignored)

=== NEW: sleep infinity & wait (should exit immediately) ===
Container running. Sending SIGTERM...
Stopped in 0s (expected: 0s -- signal was handled)
```

2026-05-07 14:52:10 -05:00

1.6 KiB

Raw Blame History

View Raw