Oran Agra
Oran Agra
looks like you already addressed them (leaks) in your last two commits. let's let it run again..
look here: https://github.com/redis/redis/blob/unstable/.github/workflows/ci.yml it would basically be: ``` make distclean make SANITIZER=address REDIS_CFLAGS='-Werror' ./runtest ```
it seems like the protocol is broken by the UNSUBSCRIBE command (returning multiple RESP replies). and it seems to have existed in previous versions as well.
i'm not a pubsub expert, but all other commands reply with a RESP array when they have multiple replies (even if they got multiple inputs). maybe it's different when the...
trying to pick this up again, i still think it's a bug in redis (not the CLI). the connection is not in a pubsub mode, so it should reply with...
yes, i think an UNSUBSCUBE that arrives to a connection that's in pubsub mode should speak the "pubsub protocol", but it if arrives on a connection that's not in pubsub...
run it with valgrind, it'll probably help you find it. also, posting the stack trace of the crash here could help.
so it doesn't crash (on a segfault), it fails on an assertion (all that valgrind text is irrelevant since it's part of the crash log code) what's the assertion about?
It also bothered me at some point, but doing that could violate cluster routing since the client won't see these keys. One way out is maybe to add some APIs...
> For example, consider the use-case where we have hashes representing persons and for each person we have a field with the amount of kids the person have. We want...