----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/49915/#review141766 -----------------------------------------------------------
Ship it! Ship It! - Alejandro Fernandez On July 11, 2016, 7:17 p.m., Jonathan Hurley wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/49915/ > ----------------------------------------------------------- > > (Updated July 11, 2016, 7:17 p.m.) > > > Review request for Ambari, Alejandro Fernandez, Dmitro Lisnichenko, and Nate > Cole. > > > Bugs: AMBARI-17660 > https://issues.apache.org/jira/browse/AMBARI-17660 > > > Repository: ambari > > > Description > ------- > > When downgrading an express upgrade, the orchestration of the downgrade does > not perform the same steps as the upgrade did. Namely, the stopping of high- > and low-level services is not present. This leads to a scenario like this: > > - Upgrade HDP 2.x to 2.y > -- Stop Storm > -- Stop ZK > -- Update Stack to 2.y > -- Restart ZK on 2.y > -- Restart Storm on 2.y > > - Downgrade HDP 2.y to 2.x > -- Update Stack to 2.x > -- Restart ZK on 2.x > -- Restart Storm on 2.x > > Notice that we didn't stop the running services. This leads to a problem > where actions which must take place while services are down can't complete > successfully. > > The case in point is Storm. Between HDP 2.4 and HDP 2.5, Storm changed the > name of a serialized class. Part of the Storm upgrade/downgrade is to always > delete local storm data. However, during an EU, if Nimbus and Supervisor are > co-located on the same host, Supervisor will write out 2.5 data since it > wasn't shut down. Consider: > > - Nimbus deletes local data and restarts on the downgrade version > - A running 2.5 Supervisor on the same host then re-creates that directory > and puts 2.5 data back in > - When the 2.5 Supervisor goes to downgrade and restart, it can't delete that > data again since Nimbus is already running and would stop. > > For this reason, we should always ensure that services are stopped on the > downgrade for an EU. > > > Diffs > ----- > > > ambari-server/src/main/resources/stacks/HDP/2.1/upgrades/nonrolling-upgrade-2.3.xml > 4c44afe > > ambari-server/src/main/resources/stacks/HDP/2.2/upgrades/nonrolling-upgrade-2.2.xml > 6495dda > > ambari-server/src/main/resources/stacks/HDP/2.2/upgrades/nonrolling-upgrade-2.3.xml > a665723 > > ambari-server/src/main/resources/stacks/HDP/2.2/upgrades/nonrolling-upgrade-2.4.xml > 90b2f2a > > ambari-server/src/main/resources/stacks/HDP/2.3/upgrades/nonrolling-upgrade-2.3.xml > 96ea82d > > ambari-server/src/main/resources/stacks/HDP/2.3/upgrades/nonrolling-upgrade-2.4.xml > 948c0e5 > > ambari-server/src/main/resources/stacks/HDP/2.3/upgrades/nonrolling-upgrade-2.5.xml > a7383a2 > > ambari-server/src/main/resources/stacks/HDP/2.4/upgrades/nonrolling-upgrade-2.4.xml > d89a38a > > ambari-server/src/main/resources/stacks/HDP/2.4/upgrades/nonrolling-upgrade-2.5.xml > 00dbaa7 > > ambari-server/src/main/resources/stacks/HDP/2.5/upgrades/nonrolling-upgrade-2.5.xml > 072f4b0 > > Diff: https://reviews.apache.org/r/49915/diff/ > > > Testing > ------- > > Verified that on HDP 2.4 to 2.5, Storm supervisors are no longer writing to > local data since they were stopped on downgrade. > > > Thanks, > > Jonathan Hurley > >