> On Feb. 3, 2018, 12:05 a.m., Greg Mann wrote: > > src/master/master.cpp > > Lines 7596 (patched) > > <https://reviews.apache.org/r/65482/diff/1/?file=1952241#file1952241line7596> > > > > Does this mean we will not correctly account for resources when a > > framework has performed operations on an agent, but has not launched tasks > > there?
Yes, that is what would happen currently. To work around that we would need to add a channel to update the slave about framework infos before accepting operations. - Benjamin ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/65482/#review196753 ----------------------------------------------------------- On Feb. 2, 2018, 3:07 p.m., Benjamin Bannier wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/65482/ > ----------------------------------------------------------- > > (Updated Feb. 2, 2018, 3:07 p.m.) > > > Review request for mesos, Greg Mann, Jie Yu, and Jan Schlicht. > > > Bugs: MESOS-8536 > https://issues.apache.org/jira/browse/MESOS-8536 > > > Repository: mesos > > > Description > ------- > > This patch fixes a bug where pending operations on a resource provider > resources where not properly accounted for in the allocator. This lead > to assertion failures when the operation became terminal and we > attempted to recover the used resources. > > Since framework information is only remembered on agents if the > framework launched a task, there exists the possibility that a master > learns about an allocation to a framework unknown to it, yet. To > accommodate that do not bookkeep allocations to unknown frameworks in > the allocator and update code handling of terminal operation updates > accordingly. > > > Diffs > ----- > > src/master/allocator/mesos/hierarchical.cpp > f33ff767dcb93556beb696c96f8cfc17baccb05e > src/master/master.cpp cc2685a6bc14103c639ce776cf1c912361e93381 > > > Diff: https://reviews.apache.org/r/65482/diff/1/ > > > Testing > ------- > > `make check`, also tested with a version of the test added in r/65045 which > triggered this issue. > > > Thanks, > > Benjamin Bannier > >