> On May 15, 2014, 1:41 a.m., Ben Mahler wrote: > > src/messages/messages.proto, line 262 > > <https://reviews.apache.org/r/21379/diff/1/?file=579745#file579745line262> > > > > I'm not sure why signals are being introduced here? This message is > > sent from the Master to the Slave, but the master does not need to send a > > 'signal' number to the slave, right? > > Alexandra Sava wrote: > Yes it's true, but I tried removing the signal parameter from > message.proto file (and also from where the protobuf handlers are installed > in slave.cpp) and I receive a compile error because the compiler does not > find a method with "Slave::shutdown(const UPID& from, const string& message)" > signature. > > Ben Mahler wrote: > This is likely because 'shutdown' is being used as both a message handler > and as a function that you're now dispatching to. We definitely don't want > these in the protobuf messages since they have no meaning there.
I created special methods for handling shutdown messages from the master. I think it is more appropriate to have a shutdown method that can be called in any scenario (shutdown issued by the master / SIGUSR1 signal) and handlers for the messages that expose the right parameters to the master. - Alexandra ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/21379/#review43072 ----------------------------------------------------------- On May 22, 2014, 3:37 p.m., Alexandra Sava wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/21379/ > ----------------------------------------------------------- > > (Updated May 22, 2014, 3:37 p.m.) > > > Review request for mesos and Ben Mahler. > > > Bugs: MESOS-544 > https://issues.apache.org/jira/browse/MESOS-544 > > > Repository: mesos-git > > > Description > ------- > > First phase: Mesos-slave support for "node drain" > > > This phase handles the shutdown of the slave, triggered by SIGUSR1 signal: > * implement a signal handler which is invoked when SIGUSR1 signal is issued > * the signal handler will call the 'shutdown' method from the Slave class, > which shuts down all the frameworks and executors that run on the slave. > > In the next phase, the slave will send an unregistration request to the > master in order to overcome the lag of the health check timer (75 sec). > > > Diffs > ----- > > src/slave/slave.hpp 769bd00f1464c9935416f044fd834e2187d83a6e > src/slave/slave.cpp a4b9570c0b4f1b56180888ef975add4c8c3cbee7 > > Diff: https://reviews.apache.org/r/21379/diff/ > > > Testing > ------- > > > Thanks, > > Alexandra Sava > >
