[
https://issues.apache.org/jira/browse/AMBARI-9106?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Dmitry Lysnichenko updated AMBARI-9106:
---------------------------------------
Description:
STR:
1) Install multinode cluster
2) Imagine that cluster deployment fails on node host1 (strictly at the same
time when some package like hbase_2_2_* is installed on any other node host2)
3) Now ambari-server sends CANCEL commands to all hosts, and agent on host2
kills yum process during installation. As a result, some hbase_2_2_* packages
are installed and some are not.
3) Retry cluster installation once more.
4) On host2, full package set for hbase_2_2_* is never installed:
{code}
[root@112 data]# grep -ri hbase_2_2 output-*
output-25.txt:2014-12-22 17:49:04,598 - Package['hbase_2_2_*'] {}
output-25.txt:2014-12-22 17:49:05,137 - Skipping installing existent package
hbase_2_2_*
output-26.txt:2014-12-22 17:49:09,322 - Package['hbase_2_2_*'] {}
output-26.txt:2014-12-22 17:49:09,816 - Skipping installing existent package
hbase_2_2_*
output-98.txt:2014-12-22 17:56:07,910 - Package['hbase_2_2_*'] {}
output-98.txt:2014-12-22 17:56:08,419 - Skipping installing existent package
hbase_2_2_*
[root@112 data]# rpm -qa hbase_2_2_*
hbase_2_2_0_0_2041-doc-0.98.4.2.2.0.0-2041.el6.noarch
{code}
while full package set for hbase_2_2_* is
{code}
[root@111 ~]# rpm -qa |grep hbase
ranger_2_2_0_0_2041-hbase-plugin-0.4.0.2.2.0.0-2041.el6.x86_64
hbase_2_2_0_0_2041-regionserver-0.98.4.2.2.0.0-2041.el6.noarch
hbase_2_2_0_0_2041-0.98.4.2.2.0.0-2041.el6.noarch
hbase_2_2_0_0_2041-thrift-0.98.4.2.2.0.0-2041.el6.noarch
hbase_2_2_0_0_2041-rest-0.98.4.2.2.0.0-2041.el6.noarch
hbase_2_2_0_0_2041-doc-0.98.4.2.2.0.0-2041.el6.noarch
hbase_2_2_0_0_2041-thrift2-0.98.4.2.2.0.0-2041.el6.noarch
hbase_2_2_0_0_2041-master-0.98.4.2.2.0.0-2041.el6.noarch
{code}
The issue here is that RMF implementation of package existence checks does
not work correctly with globs like "hbase_2_2_*". If at least one package that
matches glob is installed, it believes that all packages are installed.
Probably the same issue with Ubuntu (did not check)
I've assigned this jira to myself since my patch will simplify fixing this
issue.
was:
STR:
1) Install multinode cluster
2) Imagine that cluster deployment fails on node host1 (strictly at the same
time when some package like hbase_2_2_* is installed on any other node host2)
3) Now ambari-server sends CANCEL commands to all hosts, and agent on host2
kills yum process during installation. As a result, some hbase_2_2_* packages
are installed and some are not.
3) Retry cluster installation once more.
4) On host2, full package set for hbase_2_2_* is never installed:
{code}
[root@112 data]# grep -ri hbase_2_2 output-*
output-25.txt:2014-12-22 17:49:04,598 - Package['hbase_2_2_*'] {}
output-25.txt:2014-12-22 17:49:05,137 - Skipping installing existent package
hbase_2_2_*
output-26.txt:2014-12-22 17:49:09,322 - Package['hbase_2_2_*'] {}
output-26.txt:2014-12-22 17:49:09,816 - Skipping installing existent package
hbase_2_2_*
output-98.txt:2014-12-22 17:56:07,910 - Package['hbase_2_2_*'] {}
output-98.txt:2014-12-22 17:56:08,419 - Skipping installing existent package
hbase_2_2_*
[root@112 data]# rpm -qa hbase_2_2_*
hbase_2_2_0_0_2041-doc-0.98.4.2.2.0.0-2041.el6.noarch
{code}
while full package set for hbase_2_2_* is
{code}
[root@111 ~]# rpm -qa |grep hbase
ranger_2_2_0_0_2041-hbase-plugin-0.4.0.2.2.0.0-2041.el6.x86_64
hbase_2_2_0_0_2041-regionserver-0.98.4.2.2.0.0-2041.el6.noarch
hbase_2_2_0_0_2041-0.98.4.2.2.0.0-2041.el6.noarch
hbase_2_2_0_0_2041-thrift-0.98.4.2.2.0.0-2041.el6.noarch
hbase_2_2_0_0_2041-rest-0.98.4.2.2.0.0-2041.el6.noarch
hbase_2_2_0_0_2041-doc-0.98.4.2.2.0.0-2041.el6.noarch
hbase_2_2_0_0_2041-thrift2-0.98.4.2.2.0.0-2041.el6.noarch
hbase_2_2_0_0_2041-master-0.98.4.2.2.0.0-2041.el6.noarch
{code}
The issue here is that RMF implementation of package existence checks does
not work correctly with globs like "hbase_2_2_*". If at least one package that
matches glob is installed, it believes that all packages are installed.
Probably the same issue with Ubuntu (did not check)
I've assigned this jira to myself since my patch for BUG-27701 will simplify
fixing this issue.
> RMF checks existence of hbase_2_2_* packages incorrectly
> --------------------------------------------------------
>
> Key: AMBARI-9106
> URL: https://issues.apache.org/jira/browse/AMBARI-9106
> Project: Ambari
> Issue Type: Task
> Components: ambari-server
> Affects Versions: 2.0.0
> Reporter: Dmitry Lysnichenko
> Assignee: Dmitry Lysnichenko
> Fix For: 2.0.0
>
> Attachments: AMBARI-9106.patch
>
>
> STR:
> 1) Install multinode cluster
> 2) Imagine that cluster deployment fails on node host1 (strictly at the same
> time when some package like hbase_2_2_* is installed on any other node host2)
> 3) Now ambari-server sends CANCEL commands to all hosts, and agent on host2
> kills yum process during installation. As a result, some hbase_2_2_* packages
> are installed and some are not.
> 3) Retry cluster installation once more.
> 4) On host2, full package set for hbase_2_2_* is never installed:
> {code}
> [root@112 data]# grep -ri hbase_2_2 output-*
> output-25.txt:2014-12-22 17:49:04,598 - Package['hbase_2_2_*'] {}
> output-25.txt:2014-12-22 17:49:05,137 - Skipping installing existent package
> hbase_2_2_*
> output-26.txt:2014-12-22 17:49:09,322 - Package['hbase_2_2_*'] {}
> output-26.txt:2014-12-22 17:49:09,816 - Skipping installing existent package
> hbase_2_2_*
> output-98.txt:2014-12-22 17:56:07,910 - Package['hbase_2_2_*'] {}
> output-98.txt:2014-12-22 17:56:08,419 - Skipping installing existent package
> hbase_2_2_*
> [root@112 data]# rpm -qa hbase_2_2_*
> hbase_2_2_0_0_2041-doc-0.98.4.2.2.0.0-2041.el6.noarch
> {code}
> while full package set for hbase_2_2_* is
> {code}
> [root@111 ~]# rpm -qa |grep hbase
> ranger_2_2_0_0_2041-hbase-plugin-0.4.0.2.2.0.0-2041.el6.x86_64
> hbase_2_2_0_0_2041-regionserver-0.98.4.2.2.0.0-2041.el6.noarch
> hbase_2_2_0_0_2041-0.98.4.2.2.0.0-2041.el6.noarch
> hbase_2_2_0_0_2041-thrift-0.98.4.2.2.0.0-2041.el6.noarch
> hbase_2_2_0_0_2041-rest-0.98.4.2.2.0.0-2041.el6.noarch
> hbase_2_2_0_0_2041-doc-0.98.4.2.2.0.0-2041.el6.noarch
> hbase_2_2_0_0_2041-thrift2-0.98.4.2.2.0.0-2041.el6.noarch
> hbase_2_2_0_0_2041-master-0.98.4.2.2.0.0-2041.el6.noarch
> {code}
> The issue here is that RMF implementation of package existence checks does
> not work correctly with globs like "hbase_2_2_*". If at least one package
> that matches glob is installed, it believes that all packages are installed.
> Probably the same issue with Ubuntu (did not check)
> I've assigned this jira to myself since my patch will simplify fixing this
> issue.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
