> On July 17, 2020, 11:22 p.m., Andrei Budnik wrote: > > src/slave/containerizer/mesos/isolators/volume/csi/isolator.cpp > > Lines 106 (patched) > > <https://reviews.apache.org/r/72690/diff/1/?file=2235927#file2235927line106> > > > > Why don't we return an error here? If an operator makes a mistake, the > > CSI isolator only prints an error and takes the first entry. It may lead to > > subtle and hard-to-detect errors in runtime caused by the misconfiguration > > of CSI plugin.
That is how we handle CNI configurations and resource provider configurations: https://github.com/apache/mesos/blob/1.10.0/src/slave/containerizer/mesos/isolators/network/cni/cni.cpp#L338:L344 https://github.com/apache/mesos/blob/1.10.0/src/resource_provider/daemon.cpp#L383:L387 So I'd like to be consistent with them. > On July 17, 2020, 11:22 p.m., Andrei Budnik wrote: > > src/slave/containerizer/mesos/isolators/volume/csi/isolator.cpp > > Lines 143 (patched) > > <https://reviews.apache.org/r/72690/diff/1/?file=2235927#file2235927line143> > > > > Using the agent's API to launch every container with CSI Volume > > specified may be quite inefficient from both performance and usability > > standpoints. Can we somehow use an object that will provide a simple > > interface to the isolator? Maybe we should create a special class (Facade > > design pattern) that will translate calls from the CSI isolator to > > corresponding calls of underlying objects (SLRP, VolumeManager, etc.)? An > > instance of this class will be passed to the Mesos containerizer. > Using the agent's API to launch every container with CSI Volume specified may > be quite inefficient from both performance and usability standpoints. We are not going to use the agent's API to launch every container with CSI Volume specified (the code flow of launching executor/task containers will not be changed), instead the agentURL here is used to launch the specified CSI plugin as standalone container which I assume should not happen frequently. - Qian ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/72690/#review221251 ----------------------------------------------------------- On July 17, 2020, 4:58 p.m., Qian Zhang wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/72690/ > ----------------------------------------------------------- > > (Updated July 17, 2020, 4:58 p.m.) > > > Review request for mesos, Andrei Budnik and Greg Mann. > > > Bugs: MESOS-10152 > https://issues.apache.org/jira/browse/MESOS-10152 > > > Repository: mesos > > > Description > ------- > > Implemented the framework and `create` method of "volume/csi" isolator. > > > Diffs > ----- > > src/CMakeLists.txt bcb5128f0e61af0d541502e4ed833da0487b7792 > src/Makefile.am a89919dd7d5ccbc4c5fa79d9a83616608f84ef49 > src/slave/containerizer/mesos/isolators/volume/csi/isolator.hpp > PRE-CREATION > src/slave/containerizer/mesos/isolators/volume/csi/isolator.cpp > PRE-CREATION > > > Diff: https://reviews.apache.org/r/72690/diff/1/ > > > Testing > ------- > > > Thanks, > > Qian Zhang > >
