> On Янв. 27, 2015, 3:02 п.п., Jonathan Hurley wrote: > > ambari-server/src/main/java/org/apache/ambari/server/agent/HeartBeatHandler.java, > > lines 491-494 > > <https://reviews.apache.org/r/30306/diff/1/?file=836148#file836148line491> > > > > Manual database edits are not something that we should support because > > they mess with JPA's caching. This work seems dangerous to do on a > > heartbeat as it involves a lot of work and queries. I'd rather you use an > > event to do this work asynchronously. > > Alejandro Fernandez wrote: > I wasn't expecting manual edits while the server was running, but if the > value in the DB does change and it happens to equal the same value that the > agent reports, which could certainly happen with support, then we still need > a way for the changes to the hostcomponentstate to bubble up through the > host_version and then the cluster_version tables. So this ensures consistency > for very little additional overhead.
This code is called only when components advertise their version and it doesn't happen often. I'd keep the safer variant until we get to the point when RU always succeeds without manual intervention. - Yurii ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/30306/#review69800 ----------------------------------------------------------- On Янв. 27, 2015, 4:10 д.п., Alejandro Fernandez wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/30306/ > ----------------------------------------------------------- > > (Updated Янв. 27, 2015, 4:10 д.п.) > > > Review request for Ambari, Jonathan Hurley, Nate Cole, and Yurii Shylov. > > > Bugs: AMBARI-9340 > https://issues.apache.org/jira/browse/AMBARI-9340 > > > Repository: ambari > > > Description > ------- > > New cluster install through the Install Wizard correctly transitioned all of > the host_versions into CURRENT, but did not propagate the changes to the > cluster_version, which stayed stuck on UPGRADING, because one host only had > AMS, so it could not create a host_version record. > > > Diffs > ----- > > > ambari-server/src/main/java/org/apache/ambari/server/agent/HeartBeatHandler.java > 092f9d54db36686412d7ccaa3318330688cf67cd > ambari-server/src/main/java/org/apache/ambari/server/state/Cluster.java > b344a052549bb393028a3d60ea9fdb0c2f70fde0 > > ambari-server/src/main/java/org/apache/ambari/server/state/cluster/ClusterImpl.java > 220a5af9923444703eab4cff1875f19aeee08742 > > ambari-server/src/main/java/org/apache/ambari/server/state/svccomphost/ServiceComponentHostImpl.java > 372325df8b1e1191b87441304dc9a994f8b56f72 > > Diff: https://reviews.apache.org/r/30306/diff/ > > > Testing > ------- > > Installed a 3-node cluster with only HDFS, ZK, and AMS. One of the hosts only > had the AMS component. When the install finishes, the other 2 hosts correctly > had host_version records in CURRENT, and the cluster_version was in CURRENT. > After the install finishes, added a 4th node with all 3 components, and it > correctly created a host_version record in the CURRENT state. > > This means that the UI needs to be robust enough to handle the case where > some hosts don't have host_version records. > > Unit tests are in progress. > > > Thanks, > > Alejandro Fernandez > >
