----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/65109/#review197066 -----------------------------------------------------------
Ship it! src/slave/slave.cpp Lines 3361-3362 (patched) <https://reviews.apache.org/r/65109/#comment277177> // TODO(mzhu): Consider shutting down the executor here // if all of its initial tasks are killed rather than // waiting for it to register. - Benjamin Mahler On Feb. 6, 2018, 5:45 p.m., Meng Zhu wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/65109/ > ----------------------------------------------------------- > > (Updated Feb. 6, 2018, 5:45 p.m.) > > > Review request for mesos, Benjamin Mahler and Vinod Kone. > > > Bugs: MESOS-8411 > https://issues.apache.org/jira/browse/MESOS-8411 > > > Repository: mesos > > > Description > ------- > > An executor should be shutdown if all of its tasks are > killed while the executor is launching. > > This patch fixes and issue where the executor is left > running when the task(s) get killed between the executor > registration/subscription and `Slave::___run()`. See > MESOS-8411 for more details. There is an additional race > in the agent failover case that is addressed in this patch. > > The fix here is to fix the race by checking an executor's various > tasks queues during task kill and executor (re-)registration, > and shutting down executors that had never received any tasks. > > > Diffs > ----- > > src/slave/constants.hpp e6cb7cc0ccdaaf981eb66defa21b38720f4e1de9 > src/slave/slave.hpp 09c01ebd1b5e8008ba9e7d412042f1db76a2c5a5 > src/slave/slave.cpp a6a5c93ab2d541c870cb52587495de20ed5ac1f4 > > > Diff: https://reviews.apache.org/r/65109/diff/7/ > > > Testing > ------- > > make check > new tests in #65111 > > > Thanks, > > Meng Zhu > >