[jira] [Assigned] (CLOUDSTACK-4855) Throttle based on the # of outstanding requests to the directly managed HV host (direct agents)

["Koushik Das (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/CLOUDSTACK-4855?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Koushik Das reassigned CLOUDSTACK-4855:
---------------------------------------

    Assignee: Koushik Das

> Throttle based on the # of outstanding requests to the directly managed HV host (direct agents)
> -----------------------------------------------------------------------------------------------
>
>                 Key: CLOUDSTACK-4855
>                 URL: https://issues.apache.org/jira/browse/CLOUDSTACK-4855
>             Project: CloudStack
>          Issue Type: Bug
>      Security Level: Public(Anyone can view this level - this is the default.) 
>          Components: Management Server
>    Affects Versions: pre-4.0.0, 4.1.0, 4.2.0
>            Reporter: Koushik Das
>            Assignee: Koushik Das
>             Fix For: 4.3.0
>
>
> Currently requests to all direct managed HV hosts (direct agents) are handled by the direct agent thread pool. The size of the pool is determined by global config direct.agent.pool.size defaulted to 500.
> Currently there is no restriction on the number of requests that can be sent to a given HV host. The down side is if a lot commands are getting generated  for some specific hosts (may be there is some issue with the host, the host is slow in responding and there is a pile up of outstanding requests), it may essentially starve the requests going to other hosts due to unavailability of direct agent threads as most of them will be serving a very few hosts.
> The problem being addressed is that a few bad hosts should not affect the entire management server. The solution is to localize the impact of the bad hosts.
> One such way is to throttle based on the # of outstanding requests on per host basis. The outstanding requests will be a % of the direct agent pool size.
> This will ensure that the impacted host will be bound by a upper cap on the number of threads it can use to process request and not the entire pool.



--
This message was sent by Atlassian JIRA
(v6.1#6144)