[ https://issues.apache.org/jira/browse/CASSANDRA-2176?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12999451#comment-12999451 ]
Jonathan Ellis commented on CASSANDRA-2176: ------------------------------------------- There's three options for capping the number of connections: - close the listening socket - leave the socket open but stop accept()ing connections - accept new connections, send an error message, then close them The third option is generally seen as "best" (some discussion: http://fixunix.com/unix/379049-server-socket-how-limit-number-connections.html) but Thrift doesn't give us a way to send an error w/o first processing a request. So I'm wondering if we should go with option 2 instead -- "timed out trying to connect" seems more straightforward than option 1 (indistinguishable from cassandra not running) or option 3 (which makes it easy to do the wrong thing on the client side -- i.e. a naive client may immediately try to reconnect). Thoughts? > Add configuration setting to cap the number of Thrift connections > ----------------------------------------------------------------- > > Key: CASSANDRA-2176 > URL: https://issues.apache.org/jira/browse/CASSANDRA-2176 > Project: Cassandra > Issue Type: Improvement > Components: API > Reporter: Jonathan Ellis > Assignee: T Jake Luciani > Priority: Minor > Fix For: 0.7.3 > > Attachments: > 0001-CASSANDRA-2176-limit-connected-clients-and-config.txt > > Original Estimate: 4h > Remaining Estimate: 4h > > At least until CASSANDRA-1405 is done, it's useful to have a connection cap > to prevent misbehaving clients from DOSing the server. -- This message is automatically generated by JIRA. - For more information on JIRA, see: http://www.atlassian.com/software/jira