Dmitry Lysnichenko created AMBARI-12125:
-------------------------------------------
Summary: If agent is not restarted after upgrade, it reads new
version file and registers as a new version
Key: AMBARI-12125
URL: https://issues.apache.org/jira/browse/AMBARI-12125
Project: Ambari
Issue Type: Bug
Components: ambari-server
Reporter: Dmitry Lysnichenko
Assignee: Dmitry Lysnichenko
Fix For: 2.1.0
I think now I understand how issue occurs.
I had consideration for a while that traceback does not match lines in
CustomServiceOrchestrator.py source. Now the picture is complete: line numbers
match exactly CustomServiceOrchestrator.py as it appears in 2.0.maint branch.
That means that agent on affected host was upgraded, but not restarted after
package upgrade.
The bug here is in workflow:
- server is upgraded
- agents on hosts are upgraded. Linux systems allow overwriting files in use
(e.g. running binaries)
- agent on all hosts but one are restarted
- server is restarted
- agent, that was not upgraded, registers on server. During registration, it
provides version that appears at /var/lib/ambari-agent/data/version file. But
agent version file is already upgraded! So *old* agent executable reads *new*
version file, and successfully registers on server.
That is an actual bug.
The fix probably would be to substitute agent version directly into agent
script, and not into separate files. Python would not re-import sources after
upgrade.
The same applies to the server upgrade.
Another option is to check if agent is running from package upgrade scripts.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
