cloudstack-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF subversion and git services (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CLOUDSTACK-9112) deployVM thread is holding the global lock on network longer and cause delays and some improvements in the planner
Date Tue, 16 May 2017 05:51:04 GMT

    [ https://issues.apache.org/jira/browse/CLOUDSTACK-9112?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16011781#comment-16011781
] 

ASF subversion and git services commented on CLOUDSTACK-9112:
-------------------------------------------------------------

Commit f34469a41b5eedade04790e7f2b3306294b3b4f5 in cloudstack's branch refs/heads/master from
[~harikrishna.patnala]
[ https://gitbox.apache.org/repos/asf?p=cloudstack.git;h=f34469a ]

CLOUDSTACK-9112: deployVM thread is holding the global lock on network longer and cause delays
and some improvements in the planner

There are some VM deployment failures happening when multiple VMs are deployed at a time,
failures mainly due to NetworkModel code that iterates over all the vlans in the pod. This
causes each deployVM thread to hold the global lock on Network longer and cause delays. This
delay in turn causes more threads to choose same host and fail since capacity is not available
on that host.

Following are some changes required to be done to reduce delays during VM deployments which
in turn causes some vm deployment failures when multiple VMs are launched at a time.

    In Planner, remove the clusters that do not contain a host with matching service offering
tag. This will save some iterations over clusters that dont have matching tagged host
    In NetworkModel, do not query the vlans for the pod within the loop. Also optimized the
logic to query the ip/ipv6
    In DeploymentPlanningManagerImpl, do not process the affinity group if the plan has hostId
provided.


> deployVM thread is holding the global lock on network longer and cause delays and some
improvements in the planner
> ------------------------------------------------------------------------------------------------------------------
>
>                 Key: CLOUDSTACK-9112
>                 URL: https://issues.apache.org/jira/browse/CLOUDSTACK-9112
>             Project: CloudStack
>          Issue Type: Bug
>      Security Level: Public(Anyone can view this level - this is the default.) 
>          Components: Management Server
>    Affects Versions: 4.6.0
>            Reporter: Harikrishna Patnala
>            Assignee: Harikrishna Patnala
>             Fix For: Future
>
>
> There are some VM deployment failures happening when multiple VMs are deployed at a time,
failures mainly due to NetworkModel code that iterates over all the vlans in the pod. This
causes each deployVM thread to hold the global lock on Network longer and cause delays. This
delay in turn causes more threads to choose same host and fail since capacity is not available
on that host.
> Following are some changes required to be done to reduce delays during VM deployments
which in turn causes some vm deployment failures when multiple VMs are launched at a time.
> - In Planner, remove the clusters that do not contain a host with matching service offering
tag. This will save some iterations over clusters that dont have matching tagged host 
> - In NetworkModel, do not query the vlans for the pod within the loop. Also optimized
the logic to query the ip/ipv6 
> - In DeploymentPlanningManagerImpl, do not process the affinity group if the plan has
hostId provided.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

Mime
View raw message