Heartbeat check fails for ExApp on remote host; Nextcloud keeps targeting 'localhost', ignores BIND_ADDRESS/OVERRIDE_APP_HOST flags

[bensiongleif]

Describe the bug

Set up Nextcloud 30.0.12 on CPU-only Rocky Linux 9 with an additional GPU-supported Rocky Linux 9 server as a processing backend for apps.

All is done in a http-only enviroment.

When doing a test deploy on the DSP, it fails the heartbeat check every time. But accessing the heartbeat via wget or curl either inside the container or inside the gpu server, where the test-deploy is made, works. The Nextcloud instance cannot reach the container. Connection is refused everytime, according to the log. However, looking a tcpdump, we can see that Nextcloud attempts to get the Heartbeat check from the localhost (nextcloud server) instead of the container, which is in the GPU server.

Neither -e OVERRIDE_APP_HOST='true' nor -e APP_API_HOST='0.0.0.0' on the DSP container had any effect. The heartbeat still targets localhost, and the app still binds to 127.0.0.1.

Using Docker's default bridge network (by leaving the Network field in the UI blank) caused the test container to fail the "Container started" step.

Steps/Code to Reproduce

Recommended settings used for docker run:
docker run
-d
--name GPU-DSP
--restart unless-stopped
--net host
-v /var/run/docker.sock:/var/run/docker.sock
-e BIND_ADDRESS=""
-e EX_APPS_NET="ipv4@127.0.0.1" \
-e NC_HAPROXY_PASSWORD='xxx'
ghcr.io/nextcloud/nextcloud-appapi-dsp:v1.5.2

In the Nextcloud Admin UI, configure a new Deploy Daemon with the following settings:
* Deployment method: docker-install
* Daemon host: http://<gpu-server-ip>:2375
* Nextcloud URL: http://<nextcloud-server-ip>
* Network: host
* Compute device: CUDA (NVIDIA)
Click the "Check connection" button (it succeeds).
Click the "Start Deploy test" button.

Expected Results

All checks work fine, app can be deployed.

Actual Results

• The "Image pull" and "Container started" steps pass, but the "Heartbeat" step fails after timing out.
• The Nextcloud log shows a cURL error 7: Failed heartbeat on http://localhost:23001: Connection refused. Nextcloud is targeting its own localhost instead of the remote GPU server.
• The deployed test container log shows the application (Uvicorn) starts successfully but is bound to 127.0.0.1 on the gpu server (docker), but is not reachable externally (by the nextcloud server in this case).
• The deployment ultimately fails with "Container healthcheck failed" after reaching 210 in the heartbeat counter.
• The deployed test container works. It is reachable from within the GPU server and within the container itself via wget http://127.0.0.1:<app-port>.
• Forcing the connection via a two-stage socat tunnel creates an EADDRINUSE (address already in use) error, as the tunnel and the app create a port conflict. This confirms the app is trying to bind to a port that the workaround needs to occupy.

Full line of the relevant nextcloud log:
{"reqId":"vZi5pHotqNXKauq4Ut3G","level":2,"time":"2025-07-07T09:19:57+00:00","remoteAddr":"","user":"--","app":"app_api","method":"","url":"--","message":"Failed heartbeat on http://localhost:23001 for 10 times. Most recent status=0, error: cURL error 7: Failed to connect to localhost port 23001: Connection refused (see https://curl.haxx.se/libcurl/c/libcurl-errors.html) for http://localhost:23001/heartbeat","userAgent":"--","version":"30.0.12.2","data":{"app":"app_api"}}

Setup configuration

Nextcloud Version = 30.0.12.2
Database = mysql/MariaDB 15.1
php = 4.2.13
HostOS = Rocky Linux 9.4 Blue Onyx
GPU and Docker HostOS = Rocky Linux 9.4
Docker Version = 28.3.0

[andrey18106]

Helllo, @bensiongleif !

Thank you for your report!

OVERRIDE_APP_HOST is Deploy daemon additional option (https://docs.nextcloud.com/server/latest/admin_manual/exapps_management/ManagingDeployDaemons.html#additional-options) that is used in AppAPI on Nextcloud side to resolve ExApp URL (https://github.com/nextcloud/app_api/blob/main/lib/DeployActions/DockerActions.php#L1170-L1179). Could you please re-create your Deploy daemon configuration with this option set to correct BIND_ADDRESS (from DSP, and it is empty from your description, it should be the IP on which server with ExApps can accept requests coming from the Nextcloud instance) and try again.

[bensiongleif]

Hello @andrey18106,

Thank you for your response. We managed to get it working with OVERRIDE_APP_HOST set to the remote hosts IP, albeit in host network mode only, which lets us use only the default 2375 port. But we need multiple exApps to run on the remote host. As doing host mode on the second container forces it to port 2375 per default and bridge mode with custom network and a custom port (e.g 2385) doesn’t connect to the container and it fails the heartbeat check and kept seeing a Connection refused error, which suggested the deployed container was binding only to its internal 127.0.0.1. Based on the buildDeployEnvs function in DockerActions.php, we saw an APP_HOST variable was passed to the container. We attempted to use the undocumented APP_API_HOST option in the UI to force this to 0.0.0.0, hoping to make the container reachable, but no luck.

Docker run used for second container:
docker run
-d
--name LLM-Container
--restart unless-stopped
--network="custom net"
-p 2385:2385/tcp
-v /var/run/docker.sock:/var/run/docker.sock
-e HAPROXY_PORT='2385'
-e NC_HAPROXY_PASSWORD='xxx'
ghcr.io/nextcloud/nextcloud-appapi-dsp:v1.5.2

with OVERRIDE_APP_HOST=ip of remote host in deploy daemon config.

Networking of the second container from inside:
Active Internet connections (only servers)
Proto Recv-Q Send-Q Local Address Foreign Address State PID/Program name
tcp 0 0 0.0.0.0:2385 0.0.0.0:* LISTEN 14/haproxy
tcp 0 0 127.0.0.11:43763 0.0.0.0:* LISTEN -
udp 0 0 127.0.0.11:44494 0.0.0.0:* -

I would greatly appreciate your help deploying multltiple exApps on a remote host. Thank you!

[oleksandr-nc]

Thank you for your response. We managed to get it working with OVERRIDE_APP_HOST set to the remote hosts IP, albeit in host network mode only, which lets us use only the default 2375 port. But we need multiple exApps to run on the remote host. As doing host mode on the second container forces it to port 2375 per default and bridge mode with custom network and a custom port (e.g 2385)

Hello. For one machine you need only one DockerSocketProxy.
Did you try to install multiple DockerSocketProxies? If yes, then can you explain the reason for it?

[bensiongleif]

The reason I attempted to use multiple DSPs was to find a working configuration that would let me run more than one ExApp (specifically stt_whisper2 and the LLM exapps) on the same remote host, as deploying a second ExApp on the same DSP in host mode kept failing the heartbeat checks, thus couldn't deploy.

[github-actions]

🔔 Friendly reminder – please triage this issue.

[oleksandr-nc]

Several ExApps should work fine on one DSP.
You don't need to install one DSP on each ExApp.

For now, I can only suggest waiting for the soon-to-be-released Nextcloud 32, and there instead of DSP there will be another module that should be easier to install (I hope that it will be easier).