Redisearch + Cluster seems unsupported

[rpkak]

Version: 4.3.4

Platform: Python 3.10.6 on Archlinux with 5.19.5-arch1-1 kernel

Description:

Dockerfile

FROM redis/redis-stack-server:latest

RUN mkdir "/configs" \
        \
	&& echo "port 7000" >> "/configs/7000"  \
	&& echo "cluster-enabled yes" >> "/configs/7000"  \
	&& echo "cluster-config-file /configs/7000nodes.conf" >> "/configs/7000"  \
	&& echo "cluster-node-timeout 5000" >> "/configs/7000"  \
	&& echo "appendonly yes" >> "/configs/7000"  \
	&& echo "protected-mode no" >> "/configs/7000"  \
	\
	&& echo "port 7001" >> "/configs/7001"  \
	&& echo "cluster-enabled yes" >> "/configs/7001"  \
	&& echo "cluster-config-file /configs/7001nodes.conf" >> "/configs/7001"  \
	&& echo "cluster-node-timeout 5000" >> "/configs/7001"  \
	&& echo "appendonly yes" >> "/configs/7001"  \
	&& echo "protected-mode no" >> "/configs/7001"  \
	\
	&& echo "port 7002" >> "/configs/7002"  \
	&& echo "cluster-enabled yes" >> "/configs/7002"  \
	&& echo "cluster-config-file /configs/7002nodes.conf" >> "/configs/7002"  \
	&& echo "cluster-node-timeout 5000" >> "/configs/7002"  \
	&& echo "appendonly yes" >> "/configs/7002"  \
	&& echo "protected-mode no" >> "/configs/7002"  \
	\
	&& echo "port 7003" >> "/configs/7003"  \
	&& echo "cluster-enabled yes" >> "/configs/7003"  \
	&& echo "cluster-config-file /configs/7003nodes.conf" >> "/configs/7003"  \
	&& echo "cluster-node-timeout 5000" >> "/configs/7003"  \
	&& echo "appendonly yes" >> "/configs/7003"  \
	&& echo "protected-mode no" >> "/configs/7003"  \
	\
	&& echo "port 7004" >> "/configs/7004"  \
	&& echo "cluster-enabled yes" >> "/configs/7004"  \
	&& echo "cluster-config-file /configs/7004nodes.conf" >> "/configs/7004"  \
	&& echo "cluster-node-timeout 5000" >> "/configs/7004"  \
	&& echo "appendonly yes" >> "/configs/7004"  \
	&& echo "protected-mode no" >> "/configs/7004"  \
	\
	&& echo "port 7005" >> "/configs/7005"  \
	&& echo "cluster-enabled yes" >> "/configs/7005"  \
	&& echo "cluster-config-file /configs/7005nodes.conf" >> "/configs/7005"  \
	&& echo "cluster-node-timeout 5000" >> "/configs/7005"  \
	&& echo "appendonly yes" >> "/configs/7005"  \
	&& echo "protected-mode no" >> "/configs/7005"  \
	\
	&& echo "#!/usr/bin/env bash" >> "/entrypoint"  \
	&& echo "redis-server /configs/7000 --loadmodule /opt/redis-stack/lib/redisearch.so &" >> "/entrypoint"  \
	&& echo 'R7000=$!' >> "/entrypoint"  \
	&& echo "redis-server /configs/7001 --loadmodule /opt/redis-stack/lib/redisearch.so &" >> "/entrypoint"  \
	&& echo 'R7001=$!' >> "/entrypoint"  \
	&& echo "redis-server /configs/7002 --loadmodule /opt/redis-stack/lib/redisearch.so &" >> "/entrypoint"  \
	&& echo 'R7002=$!' >> "/entrypoint"  \
	&& echo "redis-server /configs/7003 --loadmodule /opt/redis-stack/lib/redisearch.so &" >> "/entrypoint"  \
	&& echo 'R7003=$!' >> "/entrypoint"  \
	&& echo "redis-server /configs/7004 --loadmodule /opt/redis-stack/lib/redisearch.so &" >> "/entrypoint"  \
	&& echo 'R7004=$!' >> "/entrypoint"  \
	&& echo "redis-server /configs/7005 --loadmodule /opt/redis-stack/lib/redisearch.so &" >> "/entrypoint"  \
	&& echo 'R7005=$!' >> "/entrypoint"  \
	&& echo "_term() {" >> "/entrypoint" \
	&& echo "  echo 'Stopping redisearch cluster'" >> "/entrypoint" \
	&& echo '  kill -TERM "$R7000"' >> "/entrypoint" \
	&& echo '  kill -TERM "$R7001"' >> "/entrypoint" \
	&& echo '  kill -TERM "$R7002"' >> "/entrypoint" \
	&& echo '  kill -TERM "$R7003"' >> "/entrypoint" \
	&& echo '  kill -TERM "$R7004"' >> "/entrypoint" \
	&& echo '  kill -TERM "$R7005"' >> "/entrypoint" \
	&& echo "}" >> "/entrypoint" \
	&& echo "sleep 1" >> "/entrypoint"  \
	&& echo "redis-cli --cluster create 127.0.0.1:7000 127.0.0.1:7001 127.0.0.1:7002 127.0.0.1:7003 127.0.0.1:7004 127.0.0.1:7005 --cluster-replicas 1 --cluster-yes" >> "/entrypoint"  \
	&& echo "trap '_term' TERM" >> "/entrypoint" \
	&& echo "trap '_term' INT" >> "/entrypoint" \
	&& echo "wait -n" >> "/entrypoint" \
	&& echo 'exit $?' >> "/entrypoint" \
	\
	&& chmod 755 "/entrypoint"

EXPOSE 7000
EXPOSE 7001
EXPOSE 7002
EXPOSE 7003
EXPOSE 7004
EXPOSE 7005

CMD [ "/entrypoint" ]

For the redis cluster I use a docker image build from the Dockerfile and start it with:

docker run -p 7000-7005:7000-7005 [imagename]

Setup:

from redis.cluster import RedisCluster, ClusterNode
from redis.commands.search.indexDefinition import IndexDefinition, IndexType
from redis.commands.search.field import TextField

r = RedisCluster(startup_nodes=[
    ClusterNode('localhost', '7000'),
    ClusterNode('localhost', '7001'),
    ClusterNode('localhost', '7002'),
    ClusterNode('localhost', '7003'),
    ClusterNode('localhost', '7004'),
    ClusterNode('localhost', '7005')
])

When I try to create a index the response is b'OK', but you can see that there was an error. Here my IPython logs:

In [2]: r.ft('foo').create_index([TextField('bar')])
MovedError
Traceback (most recent call last):
  File "path/venv/lib/python3.10/site-packages/redis/cluster.py", line 1074, in _execute_command
    response = redis_node.parse_response(connection, command, **kwargs)
  File "path/venv/lib/python3.10/site-packages/redis/client.py", line 1254, in parse_response
    response = connection.read_response()
  File "path/venv/lib/python3.10/site-packages/redis/connection.py", line 839, in read_response
    raise response
redis.exceptions.MovedError: 12182 127.0.0.1:7002
Out[2]: b'OK'

When I run r.ft('foo').create_index([TextField('bar')]) again I get:

...

ResponseError: Index already exists

So probably the first try worked as expected, but only prints an error.

By looking into the code I think that these two lines of RedisCluster._determine_nodes are causeing the issue because they always use the default node and not the node of slot of the index name:

redis-py/redis/cluster.py

Lines 836 to 837 in e6cd4fd

	elif command in self.__class__.SEARCH_COMMANDS[0]:
	return [self.nodes_manager.default_node]

When I removed these lines the error was gone, but I think the lines have to be there for a reason.

I while looking into the code I also saw, that Search.pipeline always (for clusters and single servers) returns a redis.commands.search.Pipeline object, which does not inharit ClusterPipeline.

Because of this r.ft('foo').pipeline() fails. Because Search.sugadd uses this function, r.ft('foo').sugadd('bar') fails as well.

[uglide]

Related to #2470

[github-actions]

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