Cluster mode KEYS only searches one node by default

[Kyle-sandeman-mrdfood]

redis-py/redis/cluster.py

Line 225 in e39c7ba

"KEYS",

The documentation provides an example: rc.keys(target_nodes=Redis.ALL_NODES). This looks acceptable as a default to me.
The KEYS command does not make sense in a cluster environment without checking at least the Redis.PRIMARIES.

Could KEYS either be moved to PRIMARIES/ALL_NODES or a note be added to the Clustering section of the documentation which explains this behaviour?

In redis-py-cluster, this command was given the flag all-nodes: https://github.com/Grokzen/redis-py-cluster/blob/8a8102a9d758d61a7ec1e2ac9050fcd34029ff3f/rediscluster/client.py#L183

As such rc.keys("...") breaks after switching to redis-py

[jonprindiville]

Oh, I'm not sure about the documentation example.

I can say, though, that generally KEYS is frowned upon in production and SCAN is often the suggested replacement.

That doesn't actually go towards your issue 🙃 but if you can be convinced to use SCAN, then in Grokzen/redis-py-cluster you can use scan_iter to transparently iterate the nodes and also the paged results of the SCAN. Seems like, yeah, it grabs all the primary nodes for that automatically.

In redis-py there's also a scan_iter... I just haven't used it first-hand to tell you that it behaves the same or not. You might have to bring your own list of target nodes.

[Kyle-sandeman-mrdfood]

Hello, a year later!
I believe we've removed usages of KEYS from our system and migrated from redis-py-cluster to redis~=4.4.0.

In any case, the current code in master still exhibits this issue. KEYS should be configured to run against PRIMARIES rather than DEFAULT_NODE (I am not sure whether this requires any other code to change)

[github-actions]

This issue is marked stale. It will be closed in 30 days if it is not updated.

[Kyle-sandeman-mrdfood]

Another year later, the code remains unchanged.