[ https://issues.apache.org/jira/browse/HBASE-4684?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13141311#comment-13141311 ]
Jean-Daniel Cryans commented on HBASE-4684: ------------------------------------------- On the patch: - Instead of creating and closing HBAs, could you stick one instance in RESTServlet instead? - In its current form, the patch will need some modifications to work with 0.90 because HBA doesn't have a close method. bq. Still not thrilled with scan/rescan of table attributes for transformation directives, in retrospect. Yeah it's kinda odd :) Not being a REST user I don't really have special fondness for one solution or another. The current one "works". > REST server is leaking ZK connections in 0.90 > --------------------------------------------- > > Key: HBASE-4684 > URL: https://issues.apache.org/jira/browse/HBASE-4684 > Project: HBase > Issue Type: Bug > Affects Versions: 0.90.4 > Reporter: Jean-Daniel Cryans > Assignee: Andrew Purtell > Priority: Critical > Fix For: 0.90.5 > > Attachments: HBASE-4684-0.90.patch > > > As reported a month ago, http://search-hadoop.com/m/FD6gmKzrxY1, the REST > server is leak ZK connections. Upon investigation I see that > TableResource.scanTransformAttrs creates a new HBA per minute per table (when > the server is getting requests) but never deletes the connection created in > there. > There are a bunch of other places where HBAs are created but not cleaned > after like SchemaResource, StorageClusterStatusResource, > StorageClusterVersionResource, ExistsResource, etc. Those places shouldn't be > as leaky under normal circumstances tho. > Thanks to Jack Levin for bringing up this issue again when he tried to > upgrade. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa For more information on JIRA, see: http://www.atlassian.com/software/jira