Posted by nadia.codes · Jul 28, 2026
The upstream issue is 200+ comments of theories and I can't tell what's been verified vs guessed, so trying to sort it. Please correct anything wrong here.
Seems confirmed (multiple people, independently):
Tried, didn't help (lots of people):
Helped some people, not everyone:
My read is this is upstream capacity showing up as an unbounded client-side wait, and the only bit you control is how long you're willing to sit there. If anyone has a repro that's deterministic at a given hour, that's the thing the upstream issue is missing.
Strong evidence gets promoted into the record above.
timeout 300 gemini ..., retry once, then fall back to something else for that task. only thing that's held up for us. doesn't fix anything but it turns an unbounded hang into a bounded one and that's the difference between annoying and can't-use-it-in-CI.
not seeing this at all on 0.53.1, europe, mostly mornings. which if the time of day thing is right is exactly what you'd expect I guess.
Sign in to join the discussion, vote, and verify fixes.
Related pages
the silent retry thing explains weeks of confusion for me. I'd ctrl-c at ten minutes assuming it was wedged. it sometimes finishes at fourteen. there's no progress output at all so "stuck" and "still working" look completely identical from outside.
honestly that's the bug. the latency is annoying, the not knowing is what makes it unusable.