Graceful reload with streaming bouncer causes admin api to hang, error=shutting down admin server: context deadline exceeded

[sholdee]

I am using caddy-docker-proxy with the bouncer in this image, which generates a config from Docker labels and loads it via the admin api on localhost:2019. Everything works well until Docker metadata changes and triggers a graceful reload, which causes the admin api to go unresponsive and generate these errors:

ERR | ts=1732909868.6840918 logger=admin msg=stopping current admin endpoint error=shutting down admin server: context deadline exceeded

This issue is not present when using the live bouncer. Graceful reloads complete successfully and the admin api remains responsive.

[hslatman]

Hey @sholdee,

Do you have more log output to debug indicating the bouncer shutting down (or similar), and are there other modules in your custom Caddy build? I'm not able to reproduce this with a locally running instance of Caddy and then reloading its configuration, basically mimicking what the caddy-docker-proxy would do.

[sholdee]

Hey @sholdee,

Do you have more log output to debug indicating the bouncer shutting down (or similar), and are there other modules in your custom Caddy build? I'm not able to reproduce this with a locally running instance of Caddy and then reloading its configuration, basically mimicking what the caddy-docker-proxy would do.

Sure, here is the streaming bouncer log and the live bouncer log. In each example I added/removed label "test: testing" on an upstream container to trigger a reload.

I am using this Dockerfile with caddy-docker-proxy, Cloudflare DNS, and Cloudflare IP modules.

[tenpaiyomi]

Just adding in, I am experiencing this behavior as well. I am using the caddy-docker-proxy in a slightly different manner from @sholdee, but to the same effect.