> On Nov. 11, 2015, 9:28 p.m., Vinod Kone wrote: > > src/slave/slave.cpp, lines 4244-4247 > > <https://reviews.apache.org/r/40177/diff/1/?file=1122973#file1122973line4244> > > > > why do it here instead of in recoverFramework() #4363? that feels more > > consistent with #1345. > > James Peach wrote: > I did this after recovery because the original code did not write > framework checkpoints if the slave was in RECOVERING state. I did not see a > reason for that, but decided to preserve the behavior as much as possible > just in case. > > Vinod Kone wrote: > originally, the slave didn't checkpoint framework during recovery stage > because it was not needed. if it is creating a framework object during > recovery, it is because it read the checkpointed data. so no need to > checkpoint again. > > but due to the compatibility issue you found, the slave can re-checkpoint > framework info during recovery because the framework info is *updated*. so i > would recommend moving this down to #1345 and do re-checkpoint if necessary.
Fixed and re-tested. We now upgrade the checkpoint during recovery but only if we need to. - James ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/40177/#review106142 ----------------------------------------------------------- On Nov. 12, 2015, 5:41 a.m., James Peach wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/40177/ > ----------------------------------------------------------- > > (Updated Nov. 12, 2015, 5:41 a.m.) > > > Review request for mesos, Kapil Arya and Vinod Kone. > > > Bugs: MESOS-3834 > https://issues.apache.org/jira/browse/MESOS-3834 > > > Repository: mesos > > > Description > ------- > > When performing an upgrade cycle, it is possible for a 0.24 and > later agent to recover from a framework checkpoint written by 0.22 > or earlier. In this case, we need to compatibly accept a missing > FrameworkID, and then rewrite the framework checkpoint so that > subsequent upgrades don't hit the same problem. > > > Diffs > ----- > > src/slave/slave.hpp ec2dfa99e6b553e2bcd82d12db915ae8625075a1 > src/slave/slave.cpp ac2d0e0153721a66495cd6539b25f5b3cee9d979 > > Diff: https://reviews.apache.org/r/40177/diff/ > > > Testing > ------- > > make check on CentOS 6.7. > Manual testing with a rolling upgrade from 0.22 > > > Thanks, > > James Peach > >
