> On Oct. 18, 2016, 6:36 p.m., Sid Wagle wrote: > > ambari-server/src/main/java/org/apache/ambari/server/controller/AmbariCustomCommandExecutionHelper.java, > > line 612 > > <https://reviews.apache.org/r/52964/diff/2/?file=1540779#file1540779line612> > > > > This can be really simplified with 1 Query rerun first host with > > nothing running on it that matches cretieria of component, randomness does > > not dd anything to the equation of runnign service check on a candidate > > host. > > > > No point maintaining local list of what hosts have running commands and > > throwing them away. > > Jonathan Hurley wrote: > I thought about this as well, but: > > - Didn't want the choice being made in the DB > - What happens if there are tasks in progress on all hosts?
Agree with that point. Also I talked to Alejandro about value of randomizing and to some extent I buy the point of not always picking same host, thereby droping the issue. - Sid ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/52964/#review153135 ----------------------------------------------------------- On Oct. 18, 2016, 6:25 p.m., Alejandro Fernandez wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/52964/ > ----------------------------------------------------------- > > (Updated Oct. 18, 2016, 6:25 p.m.) > > > Review request for Ambari, Dmytro Grinenko, Di Li, Dmitro Lisnichenko, > Jonathan Hurley, Nate Cole, Sumit Mohanty, and Sid Wagle. > > > Bugs: AMBARI-18619 > https://issues.apache.org/jira/browse/AMBARI-18619 > > > Repository: ambari > > > Description > ------- > > STR: > > * Deploy a 3-node cluster with Ambari 2.4 and HDP 2.5 with clients on every > host. > * Run multiple service checks in parallel, but notice that they typically run > on the same 1 or 2 hosts. > > Currently, Ambari relies on getting the list of candidate hosts from the DB > and excludes all hosts that are in maintenance mode. From that list, it picks > the first host that is healthy (i.e., heartbeating). This means that the > logic does not pick a random host. > > Instead, Ambari should always pick a random host and prefer to schedule on > hosts that have 0 in-progress commands. > > > Diffs > ----- > > > ambari-server/src/main/java/org/apache/ambari/server/controller/AmbariCustomCommandExecutionHelper.java > a671f0c > > ambari-server/src/main/java/org/apache/ambari/server/controller/AmbariManagementController.java > 746bca4 > > ambari-server/src/main/java/org/apache/ambari/server/controller/AmbariManagementControllerImpl.java > 5d8f279 > > ambari-server/src/main/java/org/apache/ambari/server/orm/dao/HostRoleCommandDAO.java > 17760c7 > > ambari-server/src/test/java/org/apache/ambari/server/controller/AmbariCustomCommandExecutionHelperTest.java > e82900d > > Diff: https://reviews.apache.org/r/52964/diff/ > > > Testing > ------- > > Waiting for unit test results. > > > Thanks, > > Alejandro Fernandez > >