Files
immich-magic-proxy/docs/troubleshooting.md
T

5.3 KiB

Troubleshooting

Video playback

If you're using Cloudflare and having issues with videos not playing well, make sure your /share/video/ paths are set to bypass cache. I ran into this issue myself, and found some helpful advice here.

Cloudflare cache bypass rule for video paths

"Download all" fails partway through

If you use Cloudflare with a "cache everything" rule and large zip downloads stop with a browser error after a few seconds, add a cache rule that bypasses cache for /share/*/download and /s/*/download (the slug-link form), the same as for /share/video/* above. From v3.3.0 the zip is streamed as it is fetched from Immich, and Cloudflare's cache cuts those downloads short.

Versions after v3.3.0 send the zip with Cache-Control: no-store, which is enough on its own if your cache rule's Edge TTL is set to respect the origin. If the rule overrides the Edge TTL, Cloudflare ignores that header and you need the bypass rule.

The gallery itself uses relative URLs, so it works behind any reverse proxy. The one place IPP needs a fully qualified URL is the og:image tag that messaging apps use for link previews. Without PUBLIC_BASE_URL, IPP builds it from the incoming request, and behind a TLS-terminating reverse proxy that request arrives as plain http:// on whatever hostname or IP the proxy used.

Set PUBLIC_BASE_URL in your docker-compose.yml to the public address of IPP, without a trailing slash:

environment:
  PUBLIC_BASE_URL: https://your-proxy-url.com

If you serve IPP from several domains, leave it unset and make sure your reverse proxy forwards the original Host header. See Running on a single domain.

Container won't start after mounting config.json

Docker refuses to start the container, with an error that ends:

Are you trying to mount a directory onto a file (or vice-versa)? Check if the specified host path exists and is the expected type

The file on the left of the volume line doesn't exist. When the source of a bind mount is missing, Docker creates an empty directory in its place, and a directory can't be mounted over the config.json file inside the image.

Check that your config.json is at that path. A relative path such as ./config.json resolves from the folder that holds docker-compose.yml. A Portainer stack resolves it inside Portainer's own data folder instead, so use an absolute path there:

    volumes:
      - /opt/immich-public-proxy/config.json:/app/config.json:ro

Delete the empty directory Docker created, then recreate the container. See #83.

Can't reach Immich using localhost:2283

This is a normal Docker thing, nothing to do with IPP.

From inside a Docker container, you can't reach another container using localhost. You need to use a Docker network IP or your server's IP address.

Here's a guide on connecting Docker containers.

IPP logs "Unable to reach Immich", but curl inside the container works

If IPP can't connect to Immich even though the container clearly can, you'll see this in the logs:

Unable to reach Immich on http://immich_server:2283
From the server IPP is running on, see if you can curl to http://immich_server:2283/api/server/ping and receive a JSON result.

yet running that same curl from a shell inside the container succeeds:

/app $ curl http://immich_server:2283/api/server/ping
{"res":"pong"}

This happens with both Docker Compose service names (like immich_server) and hostnames from a local DNS resolver (for example an AdGuard Home or dnsmasq CNAME rewrite). It is a quirk of the musl libc used by the node:alpine base image, not IPP itself.

Node's fetch asks for the IPv4 (A) and IPv6 (AAAA) records at once. If your resolver answers the AAAA query with NXDOMAIN instead of the correct empty NOERROR, musl fails the whole lookup - even though the IPv4 address is valid

  • and IPP reports the connection as failed. curl and glibc (the non-alpine node image) tolerate this, which is why curl works and this only shows up here. The underlying error, hidden behind the log message above, is getaddrinfo ENOTFOUND.

The reliable fix is to disable IPv6 for the container so only the IPv4 lookup happens. Add the sysctls block to your service in docker-compose.yml:

services:
  immich-public-proxy:
    image: alangrainger/immich-public-proxy:latest
    # ...your existing config...
    sysctls:
      - net.ipv6.conf.all.disable_ipv6=1
      - net.ipv6.conf.default.disable_ipv6=1

Alternatively, point IMMICH_URL at Immich's IP address (a numeric address skips DNS resolution entirely), or fix your resolver to return an empty NOERROR (NODATA) rather than NXDOMAIN for the missing AAAA record.

See issues #203 and #263 for the full investigation.


I use Linux/Android, so this project is tested with BrowserStack for Apple/Windows devices.