> On March 8, 2016, 12:37 a.m., Ian Downes wrote: > > src/master/master.cpp, line 6461 > > <https://reviews.apache.org/r/44474/diff/1/?file=1283503#file1283503line6461> > > > > I'm not familiar with this code but it appears to be changing the > > behavior substantially. > > > > The original code sums across all slaves that are registered at the > > time of the metric call. The new code tracks counters for the various > > states when a task on a registered slave changes. IIUC, this is a > > fundamental change in how the metrics are determined because the slave > > registration state is considered at different times. For example, a task > > may change state on a registered slave and it gets counted, then the slave > > becomes unregistered, then the metric endpoint is queried. Old and new code > > will give different numbers? > > > > If my understanding is correct, then perhaps (somewhat) more performant > > code could be achieved by maintaining counters at the slave level, and then > > aggregating the counters from *registered* slaves?
So are you saying I miss the slave unregistration event? But when we unregister a slave, we remove all the tasks from that slave too, and the counter is updated in removeTask(). Or I misunderstand your point? As the slaves/tasks states are complicated, I am not surprised at all I miss some case, please point it out explicitly so that I can fix it. ;) > On March 8, 2016, 12:37 a.m., Ian Downes wrote: > > src/master/metrics.cpp, lines 38-46 > > <https://reviews.apache.org/r/44474/diff/1/?file=1283505#file1283505line38> > > > > Please keep the original order which more closely matches the > > progression of states, i.e., staging before starting. Nope, we use TASK_* as an index for this array, therefore it has to be in this order. - Cong ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/44474/#review122423 ----------------------------------------------------------- On March 7, 2016, 10:51 p.m., Cong Wang wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/44474/ > ----------------------------------------------------------- > > (Updated March 7, 2016, 10:51 p.m.) > > > Review request for mesos, Ian Downes and Vinod Kone. > > > Bugs: MESOS-4740 > https://issues.apache.org/jira/browse/MESOS-4740 > > > Repository: mesos > > > Description > ------- > > Avoid iterate the list of slaves, instead just maintain some counters. > > > Diffs > ----- > > src/master/master.hpp ea26670e6c6c67314406fded510e8fdd46053dc8 > src/master/master.cpp 57ff4a39039f573b8586bc03f873f97826b97f6f > src/master/metrics.hpp 9d201fcce1c46a890c86a889ab31029f9a061561 > src/master/metrics.cpp 30c091198a8fdd6d6a957a351dc37d3dae7788e4 > > Diff: https://reviews.apache.org/r/44474/diff/ > > > Testing > ------- > > Manual check > > > Thanks, > > Cong Wang > >