KeyError in asyncio cluster get_node_from_slot

[sman591]

Version: 4.3.4

Platform: Python 3.8.10, AWS Elasticache, Redis 6

Description:

When calling await cluster.delete(key), some of our machines will throw a KeyError (e.g. KeyError: 14097) from this line:

redis-py/redis/asyncio/cluster.py

Line 1041 in bea00b1

return self.slots_cache[slot][0]

This isn't consistent, though. We load balance across 3 servers, and only 2 out of the 3 serve this error when it happens.

[sman591]

I think I have a reproducible example:

cluster = RedisCluster(
  host=config.REDIS_HOST,
  port=config.REDIS_PORT,
  decode_responses=True,
  read_from_replicas=read_from_replicas,
  skip_full_coverage_check=True,
)

await cluster.delete("foo")  # -> KeyError

However, if I run cluster.ping() first, it works:

cluster = RedisCluster(
  host=config.REDIS_HOST,
  port=config.REDIS_PORT,
  decode_responses=True,
  read_from_replicas=read_from_replicas,
  skip_full_coverage_check=True,
)

await cluster.ping()
await cluster.delete("foo")  # -> Success

Similarly, if I run await cluster.initialize(), this also fixes it

[sman591]

Seems like there might be an auto-initialization piece missing in some of the asyncio cluster code, or .initialize() needs to be enforced as a required call

[dvora-h]

@sman591 Thanks for reporting this bug! All the details you provided helped us find exactly what is the problem.
I opened this PR to fix it: #2439

[sman591]

Sweet! That PR looks great. I'll keep an eye out for the next release.

Thank also for porting all of the asyncio + cluster support over to redis-py -- I'm migrating over two apps from aredis, and this has been invaluable