[ https://issues.apache.org/jira/browse/TS-2490?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14102866#comment-14102866 ]
Leif Hedstrom commented on TS-2490: ----------------------------------- As we discussed a few times, it feels that with the new life cycles in traffic_server, some of this could move over there. Such as, traffic_server knows when it's ready to serve requests (proxy), and maybe that should trigger the start of external health checks and monitoring. A good question was brought up, in that you don't want a life cycle to be allowed to take arbitrarily long time, but that can probably be configured and managed in traffic_server itself, such that it won't allow a certain life cycle to take too long. Basically, simplify traffic_cop (maybe even eliminate all / most of it), in favor of using Linux tools, and more intelligence in traffic_server itself > traffic_cop kills TS process before it completes starting > --------------------------------------------------------- > > Key: TS-2490 > URL: https://issues.apache.org/jira/browse/TS-2490 > Project: Traffic Server > Issue Type: Bug > Components: Cop, Core > Reporter: David Carlin > Assignee: James Peach > Labels: review, yahoo > Fix For: 5.2.0 > > Attachments: TS-2490.diff > > > Several things can slow traffic_server startup > - Long remap.config file > - Long list of certs in ssl_multicert.config (see TS-2058) > - Plugins that take too long to initialize > You end up with a race condition where traffic_server never starts because > traffic_cop kills it before it completes. > There was a hack at Yahoo to address this - a YTS setting > proxy.config.watch_sleep_time to change the heartbeat interval as a > workaround. Doesn't seem like a practical solution for issues like TS-2058 > where the heartbeat would need to be in minutes. -- This message was sent by Atlassian JIRA (v6.2#6252)