[
https://issues.apache.org/jira/browse/MESOS-5999?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Greg Mann updated MESOS-5999:
-----------------------------
Summary: Re-evaluate libevent SSL socket EOF semantics in libprocess (was:
Re-evaluate socket EOF semantics in libprocess)
> Re-evaluate libevent SSL socket EOF semantics in libprocess
> -----------------------------------------------------------
>
> Key: MESOS-5999
> URL: https://issues.apache.org/jira/browse/MESOS-5999
> Project: Mesos
> Issue Type: Bug
> Components: libprocess
> Reporter: Greg Mann
> Labels: mesosphere
>
> While debugging some issues related to libprocess
> finalization/reinitialization, [~bmahler] pointed out that libprocess doesn't
> strictly adhere to the expected behavior of Unix sockets after an EOF is
> received. If a socket receives EOF, this means only that the writer on the
> other end has closed the write end of its socket. However, the other end may
> still be interested in reading. Libprocess currently treats a received EOF as
> if {{shutdown()}} has been called on the socket, and both ends have been
> closed for both reading and writing (see
> [here|https://github.com/apache/mesos/blob/1.0.0/3rdparty/libprocess/src/libevent_ssl_socket.cpp#L349-L360]
> and
> [here|https://github.com/apache/mesos/blob/1.0.0/3rdparty/libprocess/src/process.cpp#L692-L697]).
> We should consider changing the EOF semantics of the {{Socket}} object to
> more closely match those of Unix sockets.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
