Actor not found
We're using Rivet deployed on Cloudflare Workers and at around 15:40 UTC today (2025-11-20) we started getting "Actor not found" for 25% of our calls when calling our actors via
getForId. The issue seems to have resolve itself after a new workers deploy 5 hours later; wondering if y'all have encountered this before? I'm assuming it's some transient Durable Objects issue.
15 Replies
hey! can you send what rivetkit version you're using
2.0.21 (I work with @maxbittker 🙂 )
was there a trace on that error by chance? older versions didn't always log the trace unfortunately
Here's an example
give me one moment, i'm digging
is this the same id or is this where the id is frequently getitng fetched after creation?
tbh it's hard to reason about 2.0.21 since it used to rely on workers kv, which is not strongly consistent. the bleeding edge version is purely DO and should be strongly consistent, it no longer relies on KV.
the above code is the only place that could throw ActorNotFound. something is clearly wrong with the eventual consistency.

however, i'm not sure why a fresh deploy would affect this unless that somehow purges the cache
workers kv has been a nightmare every time i try to use it
this is basically what we're doing:
hmm okay so maybe upgrading to the bleeding edge version should resolve this problem
that smells like a worker kv consistency issue
i don't see any other way
i like it
(moving off worker kv)
i would recommend going back to bleeding edge which uses pure DO. i can do a quick sanity check and send over the package name if you'd like
haha yeah
do you have a version to recommend?
yeah give me one moment
it's building gonna take a sec
fixing a bug related to https://discord.com/channels/822914074136018994/1437924726327738398 hold on
give this a go:
this has the inline client + does not use workers kv
@s̸̘̩̓̍ẻ̸͕ͅå̶̡͖ṅ̴̹̊͜ if you didn't see https://discord.com/channels/822914074136018994/1437924726327738398/1441195119859470561
this was released in 2.0.24
lmk if that works for yall
works great, thanks @Nathan !
FYI we ran into some hiccups with DO migrations since ActorDO was previously deployed without sqlite and the new version needed sqlite
instead of dealing with playing out the right sequence of migrations to turn down the old namespace and turn up the new one we just opted to have some downtime aka delete our worker entirely, and redeploy it
thx for the heads up, we moved over from do kv to sqlite and i suspect there was a glitch in the migration there. will take a look.