You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
rvl index listall fails on Redis Cluster: FT._LIST is sent without a target node #847
Against a Redis Cluster, rvl index listall exits with status 1 and prints:
No way to dispatch this command to Redis Cluster. Missing key.
You can execute the command by specifying target nodes.
Command: ('FT._LIST',)
Other rvl index subcommands work against the same cluster. rvl index info -i <name>, for example, goes through SearchIndex. listall is the only one that bypasses it.
Reproduction
Start a Redis Cluster with the Search module (Redis 8.x).
Run:
rvl index listall --url "redis://127.0.0.1:7201?cluster=true"
The --json variant fails the same way, with nothing on stdout.
The CLI effectively does this:
from redis.cluster import RedisCluster
client = RedisCluster.from_url("redis://127.0.0.1:7201")
client.execute_command("FT._LIST") # RedisClusterException: Missing key
# Sending it to one node works and returns the full index list:
client.execute_command("FT._LIST", target_nodes=client.get_default_node())
# [b'my_index']
Expected behaviour: the command prints the indexes in the cluster, as it does against a standalone Redis.
Actual behaviour: it exits with status 1 and prints the RedisClusterException above to stderr.
Cause
Index.listall() in redisvl/cli/index.py opens its own connection and sends the command directly:
conn = RedisConnectionFactory.get_redis_connection(redis_url=redis_url)
indices = convert_bytes(conn.execute_command("FT._LIST"))
FT._LIST takes no key. redis-py 7.x and 8.x route keyless search commands using a static policy table (redis/commands/policies.py), and _list isn’t in that table’s ft section. So redis-py falls back to slot routing by key, finds no key, and raises. Sync SearchIndex.listall() behaves the same way on main: it worked on redis-py 6.4 and failed on 7.4 and 8.1.
Related: #795 fixes SearchIndex.listall() and AsyncSearchIndex.listall() on a cluster by sending FT._LIST to client.get_default_node(). The CLI doesn’t call either method, so it stays broken after #795 merges.
Suggested fix
Have the CLI list indexes through SearchIndex.listall(), so it uses the same cluster routing. There’s already a helper that does this: redisvl.migration.utils.list_indexes() builds a placeholder SearchIndex and calls listall(). The CLI can call it, or the helper can move to a neutral module such as redisvl/redis/utils.py so cli/index.py doesn’t import from migration/.
Ordering: this should land after #795. The change doesn’t touch #795’s files, but the CLI only works on a cluster once #795’s listall() routing is on main.
Tests: add a unit test that runs rvl index listall (with and without --json) and checks that it calls SearchIndex.listall() and prints the result.
Activity