> On Dec. 11, 2017, 9:37 p.m., Vinod Kone wrote: > > src/master/master.cpp > > Lines 6433-6434 (patched) > > <https://reviews.apache.org/r/64506/diff/1/?file=1912671#file1912671line6433> > > > > Instead of this, I would propose that we have a proper state machine > > and transition the agent through that. That would avoid issues like this.
I think its a matter of time allocation: This would probably take 1-2 days, a complete refactoring to use a state machine more like 1-2 weeks, and both would improve on the current situation. So I wouldn't want to rule out doing this, even if I agree that it would be even better to use a state machine. - Benno ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/64506/#review193449 ----------------------------------------------------------- On Dec. 12, 2017, 9:57 a.m., Benno Evers wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/64506/ > ----------------------------------------------------------- > > (Updated Dec. 12, 2017, 9:57 a.m.) > > > Review request for mesos and Benjamin Mahler. > > > Repository: mesos > > > Description > ------- > > Previously, the slave was not erased from the `registering` > and `reregistering` sets in the master for some code paths > that would result in a failed (re-)registration attempt. > > This could lead to a state where the reason of the unsuccessful > (re-)registration attempt is fixed on the agent, but the master > ignores subsequent attempts because it assumes the previous > operation is still in progress. > > > Diffs > ----- > > src/master/master.cpp b3e074cfe86600793310deb87932fa145e95055d > src/tests/master_tests.cpp 7b087675200fe0ca69319b715997700a91d9f6b2 > > > Diff: https://reviews.apache.org/r/64506/diff/1/ > > > Testing > ------- > > `make check` > > > Thanks, > > Benno Evers > >
