IndexError: list index out of range about django-redis HOT 5 CLOSED

The aforementioned change makes tests fail so more investigation is needed.

I'd really like to look into this but any hints in how to reproduce would be very helpful.

The pattern that emerges so far is that when the bug occurs, the idx value in the following line (hash_ring.py) is always number_of_redis_backends*128:

return (self.ring[self.sorted_keys[idx]], idx)

This problem occurs when using the sharding client.

from django-redis.

Hi!

Thanks for report it!

You can give me a number of redis backends are you using? Is just for try reproduce this bug an make a test case for it.

from django-redis.

Nice! Thanks a lot for the fix :)

from django-redis.

Nice to know that it's fixed.

I was recently looking into it a bit but hadn't had the time too create a testcase for it.

What I found out was that in all cases where the error occured, the idx resulting from the bisect was replicas*nodes: 256 for 2 shards set-up, 512 for a system where we have 4 redis shards.
As to be expected the hash generated from the keys was always something in the 'fff...rest_of_hash' range.

What puzzles me though is why the min() statement doesn't correct the index to become (replicas*nodes)-1. I'm probably overlooking something but debug efforts until now showed this should have worked as expected (meaning: prevent indexError by correcting the bisect result to the highest available index).

Anyway, the fix has the nice side-effect that hashes starting like '000....' will result in idx -1, which turns the sorted hashes in a true ring without the last ranges being merged to the last node.

Thx again!

from django-redis.

Thank you for reporting errors and help make this library more robust :D

from django-redis.