----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/57109/#review167169 -----------------------------------------------------------
src/slave/slave.cpp (lines 5311 - 5315) <https://reviews.apache.org/r/57109/#comment239342> We could clarify here that in order to support of a scheduler upgrading to MULTI_ROLE and then changing its roles we need to do <this>. src/slave/slave.cpp (lines 6948 - 6950) <https://reviews.apache.org/r/57109/#comment239340> It would be nice to avoid checkpointing every time we recover the agent, since we only need to re-checkpoint if any allocation info injection took place. I was also going to suggest a comment here but I think once updated to conditional checkpointing it will be a bit more clear that we're doing this in support of the multi-role upgrade case (if not we probably want to clarify this). - Benjamin Mahler On Feb. 27, 2017, 10:18 p.m., Michael Park wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/57109/ > ----------------------------------------------------------- > > (Updated Feb. 27, 2017, 10:18 p.m.) > > > Review request for mesos and Benjamin Mahler. > > > Bugs: MESOS-7061 > https://issues.apache.org/jira/browse/MESOS-7061 > > > Repository: mesos > > > Description > ------- > > Re-checkpointed the tasks and executors during agent recovery by calling > `checkpointX` to `recoverX` functions for tasks and executors. > > > Diffs > ----- > > src/slave/slave.hpp 3b0aea4e3e9a17501077beccbccaab4abbe11af2 > src/slave/slave.cpp fc480ae23ffa5cdeeb79b3621a08e1f8703bc01a > > Diff: https://reviews.apache.org/r/57109/diff/ > > > Testing > ------- > > > Thanks, > > Michael Park > >
