The meter read 1.1M commands and 133,421 connections, against 17,000 writes.
The queue cannot produce that: an idle worker asks once every fifteen seconds
and a run publishes a progress line at a time.
It was CrewAI's lock. With REDIS_URL set, CrewAI uses `portalocker.RedisLock` -
a pub/sub lock: `pubsub_numsub`, `client_setname`, `subscribe`, a second
`pubsub_numsub` and an `unsubscribe`, on a pubsub connection opened and closed
around them, with a thread polling every 100ms while it is held. One is taken
per `Crew(...)` built and another per task output stored, and this platform
builds a crew per model call.
What it guards is a SQLite file in the container's own storage directory, so a
distributed lock has nothing to coordinate. The file backend existed but was
installed only when a probe found Redis unreachable - covering the laptop and
never the server. It is installed always now.
Also: the event stream wrapped a `get_message()` that defaults to timeout=0 in
`wait_for(..., 2.0)`…
Find sources with the key that works, and show the whole day
Four Arabic bags between 26 and 30 August shipped with no source at all: no
in-text citation anywhere, and a references page carrying the word المراجع and
nothing under it. Nothing failed. Every step reported success.
The first Serper key had run out of credit. A spent key does not fail the way
a broken key fails - it answers 400 with a well-formed body that simply has no
`organic` field - so the caller read an empty list and could not tell it from
"nothing matched". The second key, working the whole time, was never tried:
`_serper_key()` returned `keys[0]` and there was no rotation behind it, in
either bag. Keys are walked now, a refused key is remembered as spent for the
process rather than asked again on each of several hundred queries, and a pool
with nothing left in it raises. A bag that found no source for any cover stops
there and says why, instead of writing itself from the model's own memory.
The deck planner was asked for a day and returned its first session. Every
bag, eve…
Decide who administers the platform, from the platform
The planner no longer receives the evidence it never reads, the quality
pass is off on its own measurements, the outline is held to the number of
مواضيع the page count pays for, a محور is written in one call instead of
one per موضوع, and the writer's model comes from settings.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
ContiAI platform: API and queue worker in one container