Attempting to register and install a repo version was stuck on one of the hosts.
The cluster version was stuck in INSTALLING, and so was one of the hosts; the other hosts were fine
Log from host were it failed
2015-09-03 12:21:59,319 - Attempting to determine actual version with build number.
2015-09-03 12:21:59,319 - Old versions: ['188.8.131.52-2598', '184.108.40.206-2716']
2015-09-03 12:21:59,369 - New versions: ['220.127.116.11-2598', '18.104.22.168-2716']
2015-09-03 12:21:59,369 - Deltas: set()
2015-09-03 12:21:59,370 - Cannot determine a new actual version installed by using the delta method.
2015-09-03 12:21:59,370 - Found actual version 22.214.171.124-210 by parsing file /var/lib/ambari-agent/data/repo_version_history.csv
In fact, the bug (at least at branch-2.1.0) is that we can not overwrite old build version of the same stack. For example,
>>> import repo_version_history
>>> import logging
>>> Logger = logging.getLogger()
>>> global Logger
>>> repo_version_history.write_actual_version_to_history_file('126.96.36.199', '188.8.131.52-210')
>>> repo_version_history.write_actual_version_to_history_file('184.108.40.206', '220.127.116.11-2716')
Originally check for existing repo version was added to avoid appending the same repo version multiple times to file. But later we decided to consider only "normalized" repo version (without build number) and introduced not obvious bug
So the bug STR is to:
- successfully install first repo version on host
- install second repo version (of the same stack) on host, and it should fail during installation.