----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/62847/#review187488 -----------------------------------------------------------
src/slave/slave.cpp Lines 1275 (patched) <https://reviews.apache.org/r/62847/#comment264514> Noticed some other issues: 1) What about 'reregistered' path? 2) What if LRP does exist, but its total has been changed to zero? 3) What if UpdateSlaveMessage is sent due to LRP registration? Will you have the same race condition? In other words, while the master is sending `CheckpointResourcesMessage`, an LRP registered with the agent and result in a `UpdateSlaveMessage` being sent out with the current total of the agent default resources. - Jie Yu On Oct. 9, 2017, 4:35 p.m., Benjamin Bannier wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/62847/ > ----------------------------------------------------------- > > (Updated Oct. 9, 2017, 4:35 p.m.) > > > Review request for mesos, Jie Yu and Jan Schlicht. > > > Bugs: MESOS-8058 > https://issues.apache.org/jira/browse/MESOS-8058 > > > Repository: mesos > > > Description > ------- > > An agent sending an 'UpdateSlaveMessage' including a total can race > with a master sending a 'CheckpointResourcesMessage'. Since > the total in 'UpdateSlaveMessage' is intended to inform the master > about changes to resource provider resources (the resources of the > agent itself are static), and 'CheckpointResourcesMessage' is used by > master not yet capable of offer operation feedback, we here explicitly > disallow combining an agent with resource providers and a master not > supporting offer operation feedback. > > > Diffs > ----- > > src/slave/slave.cpp 2e05637ed10a39eb3f4ce953506b1cb5d50caf3c > src/tests/oversubscription_tests.cpp > 09b4a423a8a9fc543f5456010e7c54919500f78b > > > Diff: https://reviews.apache.org/r/62847/diff/1/ > > > Testing > ------- > > `make check` > > > Thanks, > > Benjamin Bannier > >
