> On Feb. 27, 2017, 9:52 p.m., David McLaughlin wrote: > > src/main/java/org/apache/aurora/scheduler/app/SchedulerMain.java, lines > > 96-104 > > <https://reviews.apache.org/r/57061/diff/5/?file=1650839#file1650839line96> > > > > Feels like a good place to add some context from the design doc here. > > E.g. why do we have multiple drivers? What are the implications of using > > each one?
Done. > On Feb. 27, 2017, 9:52 p.m., David McLaughlin wrote: > > src/main/java/org/apache/aurora/scheduler/mesos/LibMesosLoadingModule.java, > > line 25 > > <https://reviews.apache.org/r/57061/diff/5/?file=1650840#file1650840line25> > > > > Update this doc to reflect the new behavior? Technically the new behaviour is the same. `DriverFactoryImpl` is a wrapper around the constructor for the scheduler driver and we wrap around the `V0Mesos` and `V1Mesos` construtors here, just using lambdas instead. We need this because all of those classes have a hard dependency on libmesos and load in native code in the constructor. > On Feb. 27, 2017, 9:52 p.m., David McLaughlin wrote: > > src/main/java/org/apache/aurora/scheduler/mesos/VersionedSchedulerDriverService.java, > > lines 114-115 > > <https://reviews.apache.org/r/57061/diff/5/?file=1650845#file1650845line114> > > > > Can we avoid repeating this in each method? I tried but not really. This is why: 1. `ensureRunning` ensures the Guava service is running before we do anything. It is a good sanity check. 2. `blockUntilRegistered` forces callers to block until we have sucessfully persisted the framework id we get from mesos. Every method that does a call needs this as a safety/sanity check. Removing some of these requires changing the interface and some of the callers. > On Feb. 27, 2017, 9:52 p.m., David McLaughlin wrote: > > src/main/java/org/apache/aurora/scheduler/mesos/VersionedSchedulerDriverService.java, > > line 48 > > <https://reviews.apache.org/r/57061/diff/5/?file=1650845#file1650845line48> > > > > Missing Javadoc. Added. > On Feb. 27, 2017, 9:52 p.m., David McLaughlin wrote: > > src/main/java/org/apache/aurora/scheduler/mesos/MesosSchedulerImpl.java, > > lines 414-426 > > <https://reviews.apache.org/r/57061/diff/5/?file=1650841#file1650841line414> > > > > Seems like a good place to use a CacheBuilder (to avoid the String > > creation per event received). > > > > Also, we don't normally initialize counters to 0. Any particular reason > > why it's necessary here? Added a cache for the metric name. We normally don't but I find that aggravating for dynamic names in a fixed space. It's much easier on metric systems and for operators who just curl `/vars` to have all of them initialized to 0 instead of having them missing. - Zameer ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/57061/#review167006 ----------------------------------------------------------- On Feb. 27, 2017, 4:43 p.m., Zameer Manji wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/57061/ > ----------------------------------------------------------- > > (Updated Feb. 27, 2017, 4:43 p.m.) > > > Review request for Aurora, David McLaughlin, Mehrdad Nurolahzade, and Stephan > Erb. > > > Bugs: AURORA-1887 and AURORA-1888 > https://issues.apache.org/jira/browse/AURORA-1887 > https://issues.apache.org/jira/browse/AURORA-1888 > > > Repository: aurora > > > Description > ------- > > This patch completes the design doc[1] and enables operators to choose between > two V1 Mesos API implementations. The first is `V0Mesos` which offers the V1 > API > backed by the scheduler driver and the second is `V1Mesos` which offers the V1 > API backed by a new HTTP API implementation. > > There are three sets of changes in this patch. > > First, the V1 Mesos code requires a Scheduler callback with a different API. > To > maximize code reuse, `MesosSchedulerImpl` was extended to implement the new > callback as well as the old callback. The code and tests were reshufled to > maxmize logic reuse. > > Second, a new driver implementation using the new API was created. All of the > logic for the new driver is encapsulated in the > `VersionedSchedulerDriverService` class. > > Third, some wiring changes were done to allow for Guice to do it's work and > allow for operators to select between the different driver implementations. > > [1] > https://docs.google.com/document/d/1bWK8ldaQSsRXvdKwTh8tyR_0qMxAlnMW70eOKoU3myo > > > Diffs > ----- > > examples/vagrant/upstart/aurora-scheduler.conf > 49fdcbd8b7406a59ae7882473b9eddbfce3ece7c > src/main/java/org/apache/aurora/scheduler/app/AppModule.java > e2ef9c30720698263106f22e3e24db5d0468b155 > src/main/java/org/apache/aurora/scheduler/app/SchedulerMain.java > 805e9de9bc45396cb8fc6e33ddb3d7428312c608 > src/main/java/org/apache/aurora/scheduler/mesos/LibMesosLoadingModule.java > e1a23590c795a489e753b77b0835d30d3be174b5 > src/main/java/org/apache/aurora/scheduler/mesos/MesosSchedulerImpl.java > eb210962c54cd0d33e3760b32f5b0ca1a7079204 > src/main/java/org/apache/aurora/scheduler/mesos/ProtosConversion.java > bc9e23b7410c00b7d5ffa4f23db93a51e9d0f405 > src/main/java/org/apache/aurora/scheduler/mesos/SchedulerDriverModule.java > 5519323079b2c957a23e093dcc77929148b4a59a > src/main/java/org/apache/aurora/scheduler/mesos/VersionedDriverFactory.java > PRE-CREATION > > src/main/java/org/apache/aurora/scheduler/mesos/VersionedSchedulerDriverService.java > PRE-CREATION > src/test/java/org/apache/aurora/scheduler/app/SchedulerIT.java > 05518048ca5518a007281269aa402a7d0710eb62 > src/test/java/org/apache/aurora/scheduler/mesos/MesosSchedulerImplTest.java > c599fe30bc903b3a3fb178df70a46d2421b6c45e > > src/test/java/org/apache/aurora/scheduler/mesos/VersionedSchedulerDriverServiceTest.java > PRE-CREATION > src/test/java/org/apache/aurora/scheduler/thrift/ThriftIT.java > f2275c757ebfa52179e31b95bf0c02b6753fb7e3 > > Diff: https://reviews.apache.org/r/57061/diff/ > > > Testing > ------- > > The e2e test has been run three times, each time with a different driver > option. > > > Thanks, > > Zameer Manji > >
