Connections are not released to a connection pool, when ClusterPipeline faces an exception on get_connection

[zakaf]

Version: What redis-py and what redis version is the issue happening on?
redis-py 4.4.2 / redis 6.0.5

Platform: What platform / version? (For example Python 3.5.1 on Windows 7 / Ubuntu 15.10 / Azure)
Python 3.9 on MacOS Ventura 13.4.1

Description: Description of your issue, stack traces from errors and code that reproduces the issue

Recently, I've faced a situation where connections never get returned to a connection pool, when there is a surge of Redis commands. while investigating the incident, i've found out that the below scenario can cause a connection leak

1. ClusterPipeline is stacked with commands that needs to be directed to at least two nodes
   • let's say node 1 and node 2
2. ClusterPipeline obtains a connection to node 1 successfully
3. ClusterPipeline fails to obtain a connection to node 2 due to ConnectionError
4. ClusterPipeline retries the entire process again, but doesn't release an acquired connection to node 1 before doing that

below code includes a test case to cause a connection leak and a possible solution to the problem

• zakaf@eb7140c

Let me know, if you guys think this is an actual issue :)

[chayim]

Hi @zakaf! First off, thank you for the report, and a potential bug fix.

@dvora-h mind checking if this makes sense in the 4.6, and 5.0 contexts since that's getting much closer to release? Then we can figure out what we may want to do here.

[dvora-h]

@zakaf this is still relevant and the fix makes sense... do you want to open a PR with the fix or wants me to take it?

[zakaf]

@dvora-h sorry for the delayed response. i've created #3133 to fix this issue

[dvora-h]

Thanks @zakaf