----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/37958/#review97311 -----------------------------------------------------------
Ship it! Just some minor updates and we can get this committed. Thanks James! Let's be clear in the description that this is motivated by some performance bugs in which event-based allocation will enqueue too many allocation rounds, and the allocator's performance is degraded due to the allocator's queue getting backed up. "Unpredictable performance" might not be clear to those who encounter this later :) docs/configuration.md (lines 291 - 295) <https://reviews.apache.org/r/37958/#comment153150> Can you have this match the help message from flags.cpp? In particular, let's not tell people that this gives more predictable performance, as that is alluding to the fact that there is are performance bug(s) in the allocator, and we should fix those! include/mesos/master/allocator.hpp (lines 43 - 56) <https://reviews.apache.org/r/37958/#comment153151> How about a 'struct' with no default constructor? Putting in these default values is a bit odd, since these fields should be coming from the master flags. include/mesos/master/allocator.hpp (line 81) <https://reviews.apache.org/r/37958/#comment153152> How about s/allocationOptions/options/ here and everywhere else in this patch? src/master/allocator/mesos/hierarchical.hpp (line 282) <https://reviews.apache.org/r/37958/#comment153153> Can you move this back up to below 'initialized' for now? - Ben Mahler On Sept. 1, 2015, 4:34 p.m., James Peach wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/37958/ > ----------------------------------------------------------- > > (Updated Sept. 1, 2015, 4:34 p.m.) > > > Review request for mesos and Ben Mahler. > > > Bugs: MESOS-3157 > https://issues.apache.org/jira/browse/MESOS-3157 > > > Repository: mesos > > > Description > ------- > > Normally, the mesos-master performs allocation cycles in response > to protocol messages and cluster events. This ties the cost of > allocation to the amount of churn in the environment, which can led > to unpredictable performance. Add an option to disable this, meaning > that allocations are only every performed by the batch interval > task. This makes resource allocation a fixed cost, independent of > the amount of environmental churn. > > > Diffs > ----- > > docs/configuration.md 2b23d48c8841e43e6e2776dfe6bfa7c022a941a7 > include/mesos/master/allocator.hpp 659f37b3f9d9fa02da9bdb6c85cd3c180a24b73a > src/master/allocator/mesos/allocator.hpp > aa55755a9c3250579e9366bdbc17a2449e95d659 > src/master/allocator/mesos/hierarchical.hpp > 38f8fd2c84314bb3731684d0e9795cb4f50a227e > src/master/flags.hpp e4b1df3f5a33049defff4688463274067f1f1ebf > src/master/flags.cpp 230c1dcedfbbb410b89defd90159d964c001a615 > src/master/master.cpp 56bcbcc08fa0f98416c5048080adb25efc588019 > src/tests/hierarchical_allocator_tests.cpp > 9748ca0b38888fee25dcec51c64d8ba84dbd4aaf > src/tests/mesos.hpp 906948d459b5a88a4ad7952801eb8c540b58c569 > > Diff: https://reviews.apache.org/r/37958/diff/ > > > Testing > ------- > > "make check" on CentOS 6 w/ devtoolset-3. Running in production for multiple > weeks. > > > Thanks, > > James Peach > >
