Conversation
|
@CAFxX is attempting to deploy a commit to the CLERKIEAI Team on Vercel. A member of the Team first needs to authorize it. |
CAFxX
force-pushed
the
cafxx-litellm-deadlocks-2
branch
from
e39fba6 to
5e9c787
Compare
CAFxX
force-pushed
the
cafxx-litellm-deadlocks-2
branch
from
ea2d6b3 to
2cdb95f
Compare
| data={"spend": {"increment": response_cost}}, | ||
| ) | ||
| break | ||
| except DB_CONNECTION_ERROR_TYPES as e: |
There was a problem hiding this comment.
if the request does fail, what would happen ?
There was a problem hiding this comment.
@krrishdholakia the current retry logic seems to be broken... at least this is what our monitoring tells us (one single deadlock in the database immediately causes a spend exception, without any retry -- if the logic was working correctly there should be multiple deadlocks in a row before a single spend exception), and this seems to be confirmed by the definition of DB_CONNECTION_ERROR_TYPES (I actually do not understand at all what those httpx exceptions are, or why they are considered relevant here; I just know that the logic seems to not be working)
FWIW, I am not removing the retry logic; I just moved it to a single place (_handle_db_exception_retriable) - and hopefully fixed it as well to actually handle postgres deadlocks - since it's duplicated with small variations in so many different places
There was a problem hiding this comment.
@krrishdholakia does this sound reasonable?
| if isinstance(e, DB_CONNECTION_ERROR_TYPES): # TODO: is this actually needed? | ||
| return True |
There was a problem hiding this comment.
I am keeping this just because it was the old logic. I have no idea, nor a way to practically test, if this is appropriate or not.
| # Unfortunately there does not seem to be a easy way to properly parse that or otherwise detect the specific | ||
| # issue, so just match using a regular expression. This is definitely not ideal, but not much we can do about | ||
| # it. |
There was a problem hiding this comment.
At least, I could not find a better way. Suggestions are most welcome.
|
This pull request has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. |
2 participants
Title
Relevant issues
Pre-Submission checklist
Please complete all items before asking a LiteLLM maintainer to review your PR
tests/litellm/directory, Adding at least 1 test is a hard requirement - see detailsmake test-unitType
🆕 New Feature
🐛 Bug Fix
🧹 Refactoring
📖 Documentation
🚄 Infrastructure
✅ Test
Changes