Testing has been done in T244499 and everything seems compatible with Bigtop.
Before scheduling the upgrade date, a couple of things need to happen first:
- Some of the hadoop worker nodes to be refreshed are kept to form a Hadoop cluster to back up data onto. The idea is to have a place where we can copy all the data that cannot be recovered in case HDFS gets corrupted during the upgrade (remote possibility but still better safe than sorry).
- Decide what data is not recoverable from other sources (pageviews, etc..) and get a rough estimate about its size (so it will inform the decision about how big the above cluster should be).