[ https://issues.apache.org/jira/browse/SOLR-6734?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16503760#comment-16503760 ]
Shawn Heisey commented on SOLR-6734: ------------------------------------ bq. If the agent is capable of reconfiguring and restarting a node, it needs to be secured. That's easy enough. Only make it accessible via loopback/localhost. I had not imagined it as something that would be accessed remotely. If there's another viable method of interprocess communication available, we wouldn't even need a network socket. The only caveat is that I'd very much prefer a cross-platform implementation. If it *is* opened up beyond local access, then you're absolutely right -- authentication would be required, and probably should NOT have a default setting. Authentication would not be a bad plan even with local-only access. Jetty's stop mechanism requires it and is not remotely accessible. bq. How would the agent start the real Solr and monitor it? That's an excellent question. I have general thoughts, but they're mostly untested. If process monitoring/control is too braindead in Java, which would not surprise me at all, I have some other ideas. I will write a separate comment to discuss the various pros and cons of those ideas. bq. Again, I think we try to reinvent what should perhaps instead be an effort to improve official ansible/puppet modules for Solr and ease production-grade container-based installs. One of the principles driving me on this issue and its parent is to make things easier for someone to get a reasonable production installation going, using only the Solr download and nothing else. I'm not saying that we shouldn't be involved in designing or writing modules for deployment tools. But I don't think we should expect the casual novice to use those tools. Also, writing those modules will be easier if we have bulletproof installation, startup, and configuration. bq. Multicast is so 1990 and not supported at all in many environments, so you'd need to build separate discovery protocols to choose from (like ES) - and at the end of the day you'll probably end up with recommending stable hard-wired ZK_HOSTs for production. Another case where you might be completely right. And I would never want to take away anyone's ability to hard-wire their ZK hosts and/or use a completely external ensemble. The idea that had me think of multicast is automated setup in a LAN environment. Users install Solr on three nodes, type some *simple* commands, and the machines find each other. Solr detects details for each server like hostnames, IP addresses, ports (for both ZK and Solr), and possibly other things, all automatically. The user can override what's detected if they want to, and then one more simple command executed on any node finalizes the config to create a fully operational cluster. It is true that some environments will not allow multicast communication to work. Anything involving multiple IP subnets would be unable to use it without some significant network infrastructure configuration. But I think *most* environments where the Solr servers are all on the same subnet would work without issues, and I don't think the code to do it would be supremely difficult. One of the complaints I hear about most with mutlicast-based technologies is an abundance of traffic on the network, to the point where network links are so overloaded that other traffic gets dropped. The traffic levels here should be quite low, and with proper usage, will not bleed to unexpected places in the network. The cluster-building stuff would need a way to manually specify the information that the automatic detection would gather, in case the automatic detection doesn't work or the user doesn't want to use it. A mandatory key/password with no default would have to be part of starting the automatic detection on each node. > Standalone solr as *two* applications -- Solr and a controlling agent > --------------------------------------------------------------------- > > Key: SOLR-6734 > URL: https://issues.apache.org/jira/browse/SOLR-6734 > Project: Solr > Issue Type: Sub-task > Reporter: Shawn Heisey > Priority: Major > > In a message to the dev list outlining reasons to switch from a webapp to a > standalone app, Mark Miller included the idea of making Solr into two > applications, rather than just one. There would be Solr itself, and an agent > to control Solr. > http://mail-archives.apache.org/mod_mbox/lucene-dev/201305.mbox/%3C807476C6-E4C3-4E7E-9F67-2BECB63990DE%40gmail.com%3E -- This message was sent by Atlassian JIRA (v7.6.3#76005) --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org