> On June 11, 2014, 11:34 p.m., Ben Mahler wrote: > > 3rdparty/libprocess/3rdparty/stout/include/stout/os.hpp, lines 816-841 > > <https://reviews.apache.org/r/21379/diff/5/?file=605406#file605406line816> > > > > We shouldn't have this LOG(FATAL) when the user cannot be found, it > > will mean the slave crashes before it had a chance to shutdown, which can > > render the shutdown ineffective. > > > > Let's remove the change here and just leave a TODO to follow up with a > > separate change to print the user that signaled the slave.
OK. I will print the user who sent the signal as soon as the refactorization of os::user method will be completed. Here is the review for os::user : https://reviews.apache.org/r/22722/ > On June 11, 2014, 11:34 p.m., Ben Mahler wrote: > > src/slave/slave.cpp, lines 141-159 > > <https://reviews.apache.org/r/21379/diff/5/?file=605408#file605408line141> > > > > I think we can clean this up a bit further: > > > > (1) We can remove signaledWrapper, it's a global variable and it's a > > bit hard to understand it's lifecycle and how it behaves when multiple > > slaves are running. > > > > (2) 'callWrapper' can just be called 'handler'. Instead of invoking a > > function object it can just 'dispatch' to Slave::signaled. > > > > (3) Per (2), Slave::signaled can be a non-static member function that > > directly calls shutdown(). As it stands, we're doing a LOG(WARNING) inside > > the signal handler, which is not safe! The problem is that in the handler (ex callWrapper) I can't do the dispatch because I don't have the PID of the slave. That's why there's the whole thing with the function object which stores the PID of the slave. - Alexandra ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/21379/#review45445 ----------------------------------------------------------- On June 8, 2014, 10:42 a.m., Alexandra Sava wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/21379/ > ----------------------------------------------------------- > > (Updated June 8, 2014, 10:42 a.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 dispatch the 'shutdown' method of 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 > ----- > > 3rdparty/libprocess/3rdparty/stout/include/stout/os.hpp > 72f3e70ccc189d0a7dd613ccaf79537d3587c49c > src/slave/slave.hpp 34687e555e6ba07863c45840aa6d07717388cf62 > src/slave/slave.cpp 643c0882a4bab1b612b3fb6fd1004e09edf5f368 > > Diff: https://reviews.apache.org/r/21379/diff/ > > > Testing > ------- > > > Thanks, > > Alexandra Sava > >
