On 5/9/2016 1:11 PM, tedsolr wrote: > I have a development environment that is using an embedded zookeeper, and the > zoo_data folder continues to grow. It's filled with snapshot files that are > not getting purged. zoo.cfg has properties > autopurge.snapRetainCount=10 > autopurge.purgeInterval=1 > Perhaps it's not in the correct location so its not getting read? Or maybe > these props don't apply for embedded instances? > > Anyone know? Thanks! > v5.2.1
Reading the source for the SolrZkServer class, it appears that only a limited set of properties in that config file is parsed by the embedded zookeeper. Only these properties are used to configure the server, all others are ignored: server.* group.* weight.* dataDir dataLogDir clientPort tickTime initLimit syncLimit electionAlg maxClientCnxns The reason that it ignores everything else is that this code is copied from Zookeeper 3.2 -- which is over six years old. Zookeeper did not have snapshot purging functionality back then. Although this is something we can fix by copying some of the code from the latest Zookeeper into Solr and making some changes, the way Solr implements the embedded zookeeper functionality will be susceptible to similar problems in the future unless we upgrade zookeeper to 3.5.x and change the embedded zookeeper implementation. ZK 3.5 is only available as an alpha version, and may not be available in a stable version for a few months. You have mentioned that it's a dev environment. A production environment configured according to recommendations (no embedded zookeeper) would not have this problem. I would recommend scripting something yourself to clean up the zookeeper data directory ... because even if we do fix this problem, the fix won't likely be available in a regular Solr release for several weeks, and will only be available in a new 6.x version, not anything in 4.x or 5.x. Thanks, Shawn
