> On June 28, 2017, 3:51 p.m., Alejandro Fernandez wrote: > > ambari-server/src/main/resources/common-services/HIVE/0.12.0.2.0/configuration/hive-env.xml > > Lines 118 (patched) > > <https://reviews.apache.org/r/60480/diff/3/?file=1766375#file1766375line118> > > > > This is missing <on-ambari-upgrade add="true"/> > > Di Li wrote: > right, I removed it per Jonathan's comment. Or I can add it back but set > to false ( not seeing the point of that comparing to not having the line at > all) > > " This will cause upgrades to change configs and have Hive report > staleness and require a restart. Instead, it seems like we should default to > False if not found and use the upgrade packs to update this. > > In other words; since Stack-A might need this when moving to Stack-B, > then it should provide this as a configuration setting. " > > Di Li wrote: > I think I misundertood Jonathan's comment. It should have been a Py only > change with default value to false and rely on config update during the > migration to tell Py to use mariadb instead of mysqld.
Yes, my point was to track this in trunk. A cluster which has gone through an upgrade with Ambari 2.5 will regress on Ambari 3.0. The problem is that we should really be tracking this via a real config property. Introducing the property is fine, but on-ambari-upgrade=false must be used since it's not critical for Ambari's operation. This means it should have sensible defaults and should be updated using the upgrade packs for stack upgrade. The whole problem centers around downtime. If a business has an outage for an Ambari upgrade, they won't be able to restart Hive when it shows stale configs. - Jonathan ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/60480/#review179157 ----------------------------------------------------------- On June 29, 2017, 11:47 a.m., Di Li wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/60480/ > ----------------------------------------------------------- > > (Updated June 29, 2017, 11:47 a.m.) > > > Review request for Ambari, Dmytro Grinenko, Di Li, Dmitro Lisnichenko, > Jonathan Hurley, Sumit Mohanty, Sid Wagle, and Tim Thorpe. > > > Bugs: AMBARI-21351 > https://issues.apache.org/jira/browse/AMBARI-21351 > > > Repository: ambari > > > Description > ------- > > Some stacks, such as BigInights, support MariaDB as the database for Hive. > When upgrading from BigInights to another stack, such as HDP, this causes the > following exception when starting the Hive service: > ```resource_management.core.exceptions.ExecutionFailed: Execution of 'service > mysqld stop' returned 5. Redirecting to /bin/systemctl stop mysqld.service > Failed to stop mysqld.service: Unit mysqld.service not loaded.``` > This looks to be caused by the hard-coded reliance of mysqld in the Python > scripts. > > > Diffs > ----- > > > ambari-server/src/main/resources/common-services/HIVE/0.12.0.2.0/package/scripts/status_params.py > b7cb148 > > > Diff: https://reviews.apache.org/r/60480/diff/4/ > > > Testing > ------- > > Tim and I patched a cluster with basically the same fix when migrating an IOP > 4.2.5 cluster to HDP 2.6/Ambari 2.5.1. THis patch has one additional change > that's to set StackFeature boolean in case this code gets merged back to > trunk later on. > > If the code only stays in this branch, the StackFeature boolean change can be > omitted. > > > Thanks, > > Di Li > >
