Edit history

[Self-hosted] GIF picker intermittently returns shard_unavailable on first GetFeatured request has not been edited, so there are no earlier versions.

Current version | Original by penguin
Show

[Self-hosted] GIF picker intermittently returns shard_unavailable on first GetFeatured request

Reproduction steps

  1. Open the self-hosted Fluxer web client and sign in.
  2. Open browser DevTools → Network.
  3. Open the GIF picker and locate the request to /v1/gifs/featured?locale=en-US.
  4. Edit and resend the request using a previously unused locale that is accepted by the endpoint, for example sv-SE or pt-BR.
  5. Observe that the first request can return 503 shard_unavailable after almost exactly 5 seconds.
  6. Send the exact same request with the same locale again.
  7. Observe that it now returns 200 almost immediately.
Examples from my instance:
sv-SE:
first request  -> 503 shard_unavailable - 5019 ms
retry          -> 200                   - 20 ms

pt-BR:
first request  -> 503 shard_unavailable - 5009 ms
later request  -> 200                   - 13 ms
Cached en-US requests return immediately. The same issue also occurs naturally when users open the GIF picker, without modifying requests in DevTools.

Observed behaviour

On my self-hosted instance, opening the GIF picker sometimes fails on the first attempt with "GIFs could not load". image The API returns 503 Service Unavailable with shard_unavailable for GET /v1/gifs/featured. This has happened during normal use for multiple users on the instance. Retrying shortly afterwards succeeds immediately. image Observed examples:
  • 503 after 5005 ms
  • 503 after 5159 ms
  • 503 after 5134 ms
Successful retries typically complete in only a few milliseconds (e.g. 12–20 ms). I have not been able to reproduce the same issue on the official web.canary.fluxer.app instance. A normal Docker Compose restart does not reliably reproduce the issue. The first GIF request after a restart can also succeed normally. Expected behaviour: The GIF picker should load successfully on the first attempt.

Build information

Stable Web 2026.1001.233209, Linux (x64), Firefox 157.0, Locale en-US

Platform

Self-hosting

Details

Reproducible sv-SE case

GET /v1/gifs/featured
status: 503
durationMs: 5019
error: ServiceUnavailableError
message: shard_unavailable
op: GetFeatured
Immediate retry:
GET /v1/gifs/featured
status: 200
durationMs: 20

Reproducible pt-BR case

GET /v1/gifs/featured
status: 503
durationMs: 5009
error: ServiceUnavailableError
message: shard_unavailable
op: GetFeatured
Same locale afterwards:
GET /v1/gifs/featured
status: 200
durationMs: 13

Example from normal use

GET /v1/gifs/featured
status: 503
durationMs: 5134
error: ServiceUnavailableError
message: shard_unavailable
op: GetFeatured
A retry approximately four seconds later:
GET /v1/gifs/featured
status: 200
durationMs: 18
The failures consistently occurring at approximately 5 seconds seem notable. The service router currently appears to use a 5-second shard request timeout. I have not confirmed what causes gifs-shard to take longer than that on the initial request. FLUXER_GIF_SERVICE_TIMEOUT_MS is not explicitly configured on this instance; the stock self-hosting Compose configuration is used.