On 10/18/2019 9:28 AM, Drew Kidder wrote:
I'm beginning to think that ZK is not setup correctly. I haven't uploaded any configuration files to ZK yet; my understanding was that I could start up a solr cloud node with no collections and upload the configuration from there. I was under the impression that it would try to connect to ZK and if it couldn't get config files from there it would use local config files.
SolrCloud will always read index configs from ZooKeeper. It will not use local config files. I believe that the only config that will be read locally is solr.xml, but that can also be placed in ZK.
Solr will run with no collections in the cloud. When the first SolrCloud node in a cluster is started, that is the state it will be in. All of the nodes can run with no collections.
Do I need to upload the solr cloud configuration files to ZK before starting up the cluster? The netstat output makes it look like the solr container is indeed connected to the ZK containers, but there's no indication as to why it cannot connect to Zookeeper that I can see.
If Solr finds no information at all in ZK when it starts, then it will create the required structures within ZK for the cluster. Index configs will not normally be uploaded just by starting Solr. Some methods of creating collections will also upload the config. Some will require that you upload the configuration first, or use one that is already there.
The entries on the netstat output show TIME_WAIT. If there were active connections, they would show ESTABLISHED. When a ZK client is running, it maintains continuous connections to all of the servers that it is given.
All of the work related to making ZK connections is handled by the ZK client, not Solr itself. I'm not sure what options are available for getting that client to provide more information about what went wrong. Based on the information available, there seems to be some kind of network problem. I do not know whether it is something in Java, Docker, or somewhere else.
Have you tried your "ruok" test as the user that is running Solr, or was that test done as root?
Thanks, Shawn
