----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/71269/#review217152 -----------------------------------------------------------
Bad review! Reviews applied: [71269, 71258, 71257, 71255, 71254] Error: 2019-08-10 23:33:52 URL:https://reviews.apache.org/r/71257/diff/raw/ [2246/2246] -> "71257.patch" [1] error: patch failed: src/master/allocator/mesos/sorter/drf/sorter.cpp:15 error: src/master/allocator/mesos/sorter/drf/sorter.cpp: patch does not apply error: patch failed: src/master/allocator/mesos/sorter/random/sorter.cpp:16 error: src/master/allocator/mesos/sorter/random/sorter.cpp: patch does not apply - Mesos Reviewbot On Aug. 10, 2019, 11:18 p.m., Meng Zhu wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/71269/ > ----------------------------------------------------------- > > (Updated Aug. 10, 2019, 11:18 p.m.) > > > Review request for mesos, Andrei Sekretenko and Benjamin Mahler. > > > Bugs: MESOS-9917 > https://issues.apache.org/jira/browse/MESOS-9917 > > > Repository: mesos > > > Description > ------- > > The role concept in Mesos fits into a tree structure naturally. > However, the role state in the allocator are currenstored > in a hashmap. This is less efficient and harder to use and reason. > > This patch introduced a `RoleTree` structure in the allocator > and organizes all the roles in to a tree. This should simplify > the code logic and opens further refactor and optimization > opportunities. > > In addition, the master code also lacks a proper tree structure > for tracking roles. We should leverage the same role tree code > here to simplify that as well. > > > Diffs > ----- > > src/master/allocator/mesos/hierarchical.hpp > 29dfa9f828f452b1aed8162c1e13f96d20907553 > src/master/allocator/mesos/hierarchical.cpp > f1fa838d654816aba14ac032174bf1522b7ae033 > > > Diff: https://reviews.apache.org/r/71269/diff/1/ > > > Testing > ------- > > make check > > Benchmaring using optimized build shows no noticable performance change. > > Benchmark: `BENCHMARK_HierarchicalAllocator_WithQuotaParam.LargeAndSmallQuota` > > ## Before: > > Added 30 agents in 1.252621ms > Added 30 frameworks in 7.747202ms > Benchmark setup: 30 agents, 30 roles, 30 frameworks, with drf sorter > Made 36 allocations in 12.791427ms > > Added 300 agents in 9.765295ms > Added 300 frameworks in 273.978325ms > Benchmark setup: 300 agents, 300 roles, 300 frameworks, with drf sorter > Made 350 allocations in 424.603736ms > > Added 3000 agents in 79.646516ms > Added 3000 frameworks in 19.17449514secs > Benchmark setup: 3000 agents, 3000 roles, 3000 frameworks, with drf sorter > Made 3500 allocations in 31.687207066secs > > ## After: > > Added 30 agents in 1.376619ms > Added 30 frameworks in 7.647487ms > Benchmark setup: 30 agents, 30 roles, 30 frameworks, with drf sorter > Made 36 allocations in 11.638116ms > > Added 300 agents in 9.506101ms > Added 300 frameworks in 263.008198ms > Benchmark setup: 300 agents, 300 roles, 300 frameworks, with drf sorter > Made 350 allocations in 418.962396ms > > Added 3000 agents in 81.553527ms > Added 3000 frameworks in 19.201708305secs > Benchmark setup: 3000 agents, 3000 roles, 3000 frameworks, with drf sorter > Made 3500 allocations in 31.583417136secs > > > Thanks, > > Meng Zhu > >
