[
https://issues.apache.org/jira/browse/MESOS-8513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16658435#comment-16658435
]
Benjamin Mahler commented on MESOS-8513:
----------------------------------------
[~abudnik] sorry I didn't see your comment, to be clear, the message will still
be logged, we just silenced the case that the shutdown was unnecessarily called.
However, we need to shore up our logging to ensure that all socket errors in
the embedded http server are logged at the warning level. Looks like we have
the following:
(1) recv error: VLOG(1) and nothing
(2) send error: nothing
(3) shutdown error: LOG(INFO)
(4) connect error: LOG(WARNING)
(5) accept error: LOG(WARNING)
(6) listen error: LOG(FATAL)
> Noisy "transport endpoint is not connected" logs on closing sockets.
> --------------------------------------------------------------------
>
> Key: MESOS-8513
> URL: https://issues.apache.org/jira/browse/MESOS-8513
> Project: Mesos
> Issue Type: Bug
> Affects Versions: 1.4.1, 1.5.0, 1.6.0
> Environment: https://reviews.apache.org/r/65474/
> Reporter: Till Toenshoff
> Assignee: Till Toenshoff
> Priority: Minor
> Labels: libprocess, logging, socket
> Fix For: 1.6.0
>
>
> When within libprocess a socket is closing, we try to shut it down. That
> shutdown fails as the socket is not connected. This is intended behavior. The
> error code returned {{ENOTCONN}} tells us that there is nothing to see here
> for such common scenario.
> The problem appears to be the logging of such event - that might appear as
> not useful - no matter which log-level is used.
> {noformat}
> E1214 08:15:18.017247 20752 process.cpp:2401] Failed to shutdown socket with
> fd 288: Transport endpoint is not connected
> {noformat}
> We should try to prevent this specific, non actionable logging entirely while
> making sure we do not hinder debugging scenarios.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
