Skip to content

fix(bfl): raise the polling timeout to 10 minutes for 4k FLUX 3 jobs - #565

Merged
Kamilbenkirane merged 1 commit into
mainfrom
fix/bfl-polling-timeout
Nov 9, 2026
Merged

Kamilbenkirane merged 1 commit into
mainfrom
fix/bfl-polling-timeout

Conversation

@Kamilbenkirane

Copy link
Copy Markdown
Member

BFL image polling gave up after 120 s (POLLING_TIMEOUT in providers/bfl/images/config.py), so a FLUX 3 Image job at 4k resolution could be billed and still end in TimeoutError. BFL's Technical Parameters guide says of 4k: "The largest output. Generation can take several minutes". Its MCP guide adds that "4k output can take several minutes" and "Large sets of images or complex edits also take longer than smaller generations". BFL documents no upper bound on job time.

Fix:

  • src/celeste/providers/bfl/images/config.py: POLLING_TIMEOUT goes from 120.0 to 600.0 (10 minutes).
    • This keeps the existing pattern: each polling wire package has one constant in its config.py (byteplus/videos, topazlabs/image, fal/queue, xai/videos MAX_POLLS * POLL_INTERVAL). The other long-running packages use 300 s. That would leave a "several minutes" 4k job right at the limit, so this uses 600 s.
    • The loop still returns as soon as a task is Ready. Only a task that never finishes waits longer, 10 minutes instead of 2, before TimeoutError.

Overlap:

Validation:

BFL documents that FLUX 3 Image at 4k "can take several minutes", so the
120 s polling ceiling turned paid jobs into TimeoutError. The loop still
returns as soon as a task is Ready; only a task that never finishes waits
longer.
@Kamilbenkirane
Kamilbenkirane merged commit 3f29674 into main Nov 9, 2026
10 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant