[ https://issues.apache.org/jira/browse/IGNITE-2643?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15272413#comment-15272413 ]
ASF GitHub Bot commented on IGNITE-2643: ---------------------------------------- Github user asfgit closed the pull request at: https://github.com/apache/ignite/pull/487 > ODBC: Potential memory leak during client disconnect. > ----------------------------------------------------- > > Key: IGNITE-2643 > URL: https://issues.apache.org/jira/browse/IGNITE-2643 > Project: Ignite > Issue Type: Sub-task > Components: odbc > Affects Versions: 1.5.0.final > Reporter: Vladimir Ozerov > Assignee: Vladimir Ozerov > Priority: Critical > Fix For: 1.6 > > > *Problem* > When client executes a query, we preserve the cursor in concurrent > collection. This could lead to two potential problems: > 1) If problem disconnected abruptly, cursor gets stuck forever => memory leak. > 2) Malicious client could flood us with requests which are never closed until > node is out-of-memory. > *Proposed solution* > 1) When onDisconnect() callback is triggered, all pending client queries must > be released. To achieve this it is better to move "OdbcNioListener.qryCurs" > to session meta. > 2) When new request is to be created, we must ensure that concurrent > disconnect is not in progress. Otherwise, we might end up in a leak again. To > achieve this, lets guard "add" logic with read-write lock. > 3) Lets think about "max-concurrent-cursors-per-connection" property. It > could be available from OdbcConfiguration. Once number of opened cursors is > exceeded, we must throw an error to the client. -- This message was sent by Atlassian JIRA (v6.3.4#6332)