Retrying not working when executing Redis Pipeline for retry_on_error exceptions

[jfcalcerrada]

Version: 4.3.6 to 5.0.1

Platform: 3.8 to 3.11

Description: redis-py does not attempt to reconnect when executing a pipeline for the errors provided in retry_on_error, instead the exceptions are bubbled up.

This library accepts a retry policy and a list of Errors/Exceptions to attempt a retry.

When executing a single command in the Sync and Async versions, it calls call_with_retry passing _disconnect_raise as a callback in the client.py and asyncio/client.py. In both cases, _disconnect_raise won't bubble up exceptions in the retry_on_error list. See:

• redis-py/redis/client.py

  Line 522 in cc4bc1a

  if (

• redis-py/redis/asyncio/client.py

  Line 589 in cc4bc1a

  if (

As exceptions aren't bubbled up, retry will work as expected when calling that callback:

redis-py/redis/retry.py

Line 49 in cc4bc1a

fail(error)

But this is not the case for Pipeline on both versions (Sync and Async). When executing a pipeline, this time the _disconnect_reset_raise callback is passed in call_with_retry:

redis-py/redis/client.py

Line 1436 in cc4bc1a

def _disconnect_raise_reset(self, conn: Redis, error: Exception) -> None:

This callback DO bubble up all exceptions but TimeError, as the exception is bubbled up, call_with_retry will raise an exception when executing this line of code:

redis-py/redis/retry.py

Line 49 in cc4bc1a

fail(error)

Thus bubbling up the exception all the way up, ignoring the retry_on_error list and the retry policy in place.

[jfcalcerrada]

This is the same issue reported here: #2179

But in this case is affecting client.Pipeline and asyncio.client.Pipeline

[cknv]

Unless I am mistaken the same is also true for the PubSub clients, both sync and async.

Bringing the different versions in sync is probably not going to be too difficult, if there is interest from the maintainers.