Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

On v2026.4.26, the gateway logs “http server listening” and then “starting channels and sidecars...”, but all internal HTTP routes time out and the container remains unhealthy; v2026.4.23 works in the same setup.

Observed on v2026.4.26:

• The container stays up but remains unhealthy.
• The gateway stages and installs bundled runtime dependencies before startup.
• The gateway logs “http server listening” after 181.3s.
• The last normal startup line is “starting channels and sidecars...”.
• After that, all internal HTTP routes time out:
  • /
  • /healthz
  • /status
  • /openclaw/canvas/
• v2026.4.23 works in the same Docker/Windows setup.

Relevant log lines:

staging bundled runtime deps before gateway startup (16 missing, 16 install specs) installed bundled runtime deps before gateway startup in 31106ms http server listening (9 plugins: acpx, active-memory, bonjour, browser, device-pair, discord, memory-core, phone-control, talk-voice; 181.3s) starting channels and sidecars... OpenRouter pricing fetch failed (timeout 60s) LiteLLM pricing fetch failed (timeout 60s)

Environment:

• Docker Desktop on Windows / WSL2
• Container: openclaw-openclaw-gateway-1
• Image: openclaw:gateway-secure-patched
• Host port: 27889 -> container 18789
• Previous working version: v2026.4.23
• Failing versions observed: v2026.4.24, v2026.4.25, v2026.4.26

Steps to reproduce

Steps to reproduce:

1. Start from a working v2026.4.23 Docker setup on Windows / WSL2.

2. Update the repository to v2026.4.26.

3. Rebuild and recreate the gateway container.

4. Start the gateway container.

5. Wait until the logs show:

   http server listening (...) starting channels and sidecars...

6. Check the container status:

   docker ps --filter "name=openclaw-openclaw-gateway-1"

7. Test the internal gateway routes from inside the container:

   docker exec -it openclaw-openclaw-gateway-1 sh -lc ' for path in / /healthz /status /openclaw/canvas/; do echo "=== $path ===" timeout 6s node -e " fetch("http://127.0.0.1:18789$path\") .then(async r => { console.log(r.status, r.ok); console.log((await r.text()).slice(0,300)); }) .catch(e => { console.error(e); process.exit(1); }) " || echo "TIMEOUT/FAILED" echo done '

Expected result:

The gateway should become healthy, the dashboard should load, and /healthz and /status should respond.

Actual result:

The container remains unhealthy. The dashboard does not become usable. All tested internal routes time out after the gateway logs “starting channels and sidecars...”.

Expected behavior

The gateway should become healthy after startup. The dashboard should load normally on the mapped host port, and the internal routes /healthz and /status should respond within a few seconds.

Actual behavior

On v2026.4.26, the gateway logs “http server listening” and then “starting channels and sidecars...”. After that, the container remains unhealthy and all tested internal HTTP routes time out:

• /
• /healthz
• /status
• /openclaw/canvas/

v2026.4.23 works in the same setup.

OpenClaw version

2026.4.26

Operating system

Windows 11

Install method

Docker Desktop on Windows / WSL2. Repository updated to v2026.4.26, then gateway container rebuilt and recreated. Container: openclaw-openclaw-gateway-1 Image: openclaw:gateway-secure-patched Host port mapping: 27889 -> 18789

Model

openai/gpt-5.4-mini

Provider / routing chain

OpenAI provider via the configured OpenClaw routing chain. The gateway log shows the selected agent model as openai/gpt-5.4-mini.

Additional provider/model setup details

The issue occurs before any user model call is made. The failure appears during gateway startup, after “http server listening” and “starting channels and sidecars...”.

The gateway also logs pricing fetch timeouts for OpenRouter and LiteLLM, but those are not part of the selected agent model in this run:

• OpenRouter pricing fetch failed (timeout 60s)
• LiteLLM pricing fetch failed (timeout 60s)

Logs, screenshots, and evidence

Relevant v2026.4.26 log lines:

[plugins] staging bundled runtime deps before gateway startup (16 missing, 16 install specs)
[plugins] installed bundled runtime deps before gateway startup in 31106ms
[gateway] http server listening (9 plugins: acpx, active-memory, bonjour, browser, device-pair, discord, memory-core, phone-control, talk-voice; 181.3s)
[gateway] starting channels and sidecars...
[gateway/model-pricing] OpenRouter pricing fetch failed (timeout 60s)
[gateway/model-pricing] LiteLLM pricing fetch failed (timeout 60s)

Internal route test result:

=== / ===
TIMEOUT/FAILED

=== /healthz ===
TIMEOUT/FAILED

=== /status ===
TIMEOUT/FAILED

=== /__openclaw__/canvas/ ===
TIMEOUT/FAILED

Impact and severity

High. The gateway container stays up but remains unhealthy, and the dashboard is not usable. This blocks normal use of OpenClaw on v2026.4.26. I had to roll back to v2026.4.23.

Additional information

Failing versions observed in this setup: v2026.4.24, v2026.4.25, v2026.4.26. Previous working version: v2026.4.23.