ambari-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Zsombor Gegesy (JIRA)" <>
Subject [jira] [Created] (AMBARI-18930) Ambari thinks a component is installed, even if the install is partially done
Date Fri, 18 Nov 2016 13:19:58 GMT
Zsombor Gegesy created AMBARI-18930:

             Summary: Ambari thinks a component is installed, even if the install is partially
                 Key: AMBARI-18930
             Project: Ambari
          Issue Type: Bug
          Components: ambari-server
    Affects Versions: 2.4.2
            Reporter: Zsombor Gegesy

We have lot's of deployment problems which ends up in a state, where Ambari thinks a component
is installed, but when it tries to start it, the component fails to start up, because missing
kerberos keytabs.
The investigation found the following scenario :
* blueprint install
* For the datanode to install it's needs the following steps: 
** install the datanode package on host A
** generate keytab(s) on server
** set up datanode-keytab on host A
* In the "Install components on host ...." step the following steps are scheduled :
** package install
** (other package install)
** keytab setup
* If there was an error in the 'other package install', then the kerberos key generation,
and host setup is skipped.
* But Ambari thinks, this component is installed - because the package install step succeeded.
* However, when/if someone tries to start the component, it fails with the missing keytabs
* The only way to fix this is to manually "Delete" and "Reinstall" the component, which doesn't
make the best impression on the user.

Proposed ideas to fix :
* re-order the install steps - first generate the keytabs on the server, distribute to the
host, and later install the package. This would ensure, when the package is installed on a
host, it's keytabs are there as well.
* mark the partially installed component as 'install failed' on the UI - so the user could
try to re-install, when they see it.

This message was sent by Atlassian JIRA

View raw message