----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/65368/#review196404 -----------------------------------------------------------
It would be helpful for posterity if the commit message is tailored towards the behavioral change we made, e.g.: ``` Only rescind offers from affected agents in maintenance API. ``` I held off on a ship it because I was left confused about when agents' machines are stored in `machines` src/master/http.cpp Line 4460 (original), 4473 (patched) <https://reviews.apache.org/r/65368/#comment276024> Not yours, but seems odd to copy the whole map rather than just the keys? src/master/http.cpp Lines 4502-4504 (original), 4481-4483 (patched) <https://reviews.apache.org/r/65368/#comment276025> Not yours, but the mutation of master variables directly from the http handler seems a little odd, I would have expected the master to have a function for replacing the maintenance schedule, so that the http handler can just call it. src/master/master.cpp Lines 7749-7750 (patched) <https://reviews.apache.org/r/65368/#comment276026> Not yours, but the `machines` variable and registry field could use some documentation, the invariant here seems to be that the master only stores non-UP machines? I'm still left confused about this after reading the code. src/master/master.cpp Lines 7761-7763 (patched) <https://reviews.apache.org/r/65368/#comment276027> Hm.. the invariant is that each agent will always have a corresponding machine stored within `machines`? Is that the same invariant as the registry `machines`? A little puzzling src/master/master.cpp Lines 7769 (patched) <https://reviews.apache.org/r/65368/#comment276028> Could you use .at where you're not intending to insert? - Benjamin Mahler On Jan. 27, 2018, 1:40 a.m., Joseph Wu wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/65368/ > ----------------------------------------------------------- > > (Updated Jan. 27, 2018, 1:40 a.m.) > > > Review request for mesos and Benjamin Mahler. > > > Bugs: MESOS-7882 > https://issues.apache.org/jira/browse/MESOS-7882 > > > Repository: mesos > > > Description > ------- > > This moves most of the conditional logic of maintenance schedules > into the master's `updateUnavailability` method, while the HTTP > handler will only loop through all MachineIDs once. > > The `updateUnavailability` method will now check if the given > Unavailability has changed before it updates the allocator. This will > prevent a maintenance schedule from rescinding offers unrelated to the > maintenance schedule. > > > Diffs > ----- > > src/master/http.cpp c489b6f525f157811549b2cc84addd9d85e87990 > src/master/master.cpp b97ebae6ebfd8ae0f73e617d0c55e140b9c3fce7 > > > Diff: https://reviews.apache.org/r/65368/diff/1/ > > > Testing > ------- > > make check > > > Thanks, > > Joseph Wu > >
