Closed marcoamorales closed 7 years ago
Hi @marcoamorales this sounds like a bug, thanks for your report. As I have time I will try and reproduce this.
I have a rather similar issue, only in this case consul-snapshot is not running on one of the master nodes. Massive number of connections and corresponding memory usage. We have several consul client nodes with a cluster of three masters. Consul-snapshot is running on one of the client hosts.
I did some more digging. This seems healtcheck related. We were polling the /health endpoint with marathon. Lowering the HC interval decreases the memory/connection growth. Looks like there's an issue with the connection recycling/closing then I guess.
same issue here, any update on this?
I think this may be part of the key to what is happening: https://github.com/hashicorp/consul/blob/master/api/api.go#L261
Can you build from master and see if the connection issues go away? If you could help me test this change that would be great as I don't have a long running cluster anymore.
Dropped in the new 2.3.0 release. Going to close this but if its still happening please reopen.
Rolling out 2.3 to production today, thanks for looking into it 👍
Hello,
I've found a problem when running consul-snapshot on the same server that is running a consul agent with -ui option.
From what I can tell, if I'm using the -ui option, consul-snapshot doesn't close the connections it creates with the local agent. The connection count will keep growing until the server is no longer able to create more connections.
Shortend output: