fix: self-check gateway probe misreports connection-refused as pass

[gitricko]

Summary

One-line fix to docker/self-check.sh — the Hermes gateway health check was incorrectly reporting ✅ Gateway HTTP 000000 when the dashboard/gateway port (9119) was unreachable, because of a bash stdout-concatenation bug in the curl fallback pattern.

The Bug

# Before (broken):
has_gateway=$(curl ... || echo "000")

When curl to an unreachable port returns 000 (from -w %{http_code}) AND the || echo "000" fallback fires, stdout gets concatenated: 000 + 000 = 000000 (6 chars).

The downstream check:

[ "$has_gateway" != "000" ]

Then evaluates "000000" != "000" as true (bash string comparison), so the gateway is reported as ✅ even though it is actually down.

The Fix

# After (fixed):
has_gateway=$(curl ... || true)
has_gateway="${has_gateway:-000}"

• || true prevents the non-zero exit from killing the set -e script, but adds nothing to stdout
• "${has_gateway:-000}" only substitutes 000 when the result is truly empty
• No concatenation → 000 stays 000 → [ "000" != "000" ] is false → gateway correctly reported as ❌

How to Test Locally

1. Ensure dashboard/gateway on port 9119 is not running
2. Run the script: HERMES_GATEWAY_URL=http://localhost:9119 ./docker/self-check.sh
3. Before fix: Shows ✅ Gateway HTTP 000000 at http://localhost:9119
4. After fix: Shows ❌ Gateway no response from http://localhost:9119

Checklist

• One surgical change — just the fallback pattern
• No other behaviour affected
• Service port polling (line 196) uses a different pattern and is unaffected