[ 
https://issues.apache.org/jira/browse/MESOS-4007?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Yong Qiao Wang updated MESOS-4007:
----------------------------------
    Description: 
To consider the Mesos master recovery and failover case, Mesos master needs to 
persist the roles and weights information in registry: 
- In the first boot, the first leading master initialize the replicated log 
with the roles/weights specified by command-line flags(--roles and --weights). 
The flags values are only useful to bootstrap the cluster, after which point 
the registry becomes the source of truth.

- At runtime, the replicated log can only be updated to add/remove/update 
entries by the operator REST API.

- For Mesos master restart/failover case, if the replicated log for 
roles/weights has exist, and then it prefers to use the registry values and 
ignore the flags (--roles/--weights), and also log a warning in Mesos master 
that the flags values are being ignored.

- For the future works, we can educate end users to create the replicated log 
to initialize the supported roles/weights before Mesos cluster bootstrap, and 
reset roles/weights configurations by update the replicated log.


  was:Persist role information to registry across master recovery/failover.


> Persist role information to registry
> ------------------------------------
>
>                 Key: MESOS-4007
>                 URL: https://issues.apache.org/jira/browse/MESOS-4007
>             Project: Mesos
>          Issue Type: Task
>            Reporter: Yong Qiao Wang
>            Assignee: Yong Qiao Wang
>
> To consider the Mesos master recovery and failover case, Mesos master needs 
> to persist the roles and weights information in registry: 
> - In the first boot, the first leading master initialize the replicated log 
> with the roles/weights specified by command-line flags(--roles and 
> --weights). The flags values are only useful to bootstrap the cluster, after 
> which point the registry becomes the source of truth.
> - At runtime, the replicated log can only be updated to add/remove/update 
> entries by the operator REST API.
> - For Mesos master restart/failover case, if the replicated log for 
> roles/weights has exist, and then it prefers to use the registry values and 
> ignore the flags (--roles/--weights), and also log a warning in Mesos master 
> that the flags values are being ignored.
> - For the future works, we can educate end users to create the replicated log 
> to initialize the supported roles/weights before Mesos cluster bootstrap, and 
> reset roles/weights configurations by update the replicated log.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to