You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@solr.apache.org by "Jan Høydahl (Jira)" <ji...@apache.org> on 2021/10/26 10:53:00 UTC
[jira] [Comment Edited] (SOLR-15694) Concept of non-data nodes
[ https://issues.apache.org/jira/browse/SOLR-15694?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17434262#comment-17434262 ]
Jan Høydahl edited comment on SOLR-15694 at 10/26/21, 10:52 AM:
----------------------------------------------------------------
Ah, did not know about that. A thousand questions pop up, such as: Should roles be supported in leader/follower mode? What would be the best way to designate roles to a node? A global file in ZK like /roles.json is one way. ENV_VARs or SysProps at each node is another. How dynamic do we need this to be, are there business needs to (re)assign roles at runtime, and in that case do we need a REST API for it? If dynamic, what role(s) would a brand new node (or k8s pod) added to the cluster be assigned? What happens if you have only assigned a role to one node and that node leaves the cluster? Should that be configurable, or would there be some cluster level plugin monitoring events, that makes such decisions? etc etc
Looking forward to seeing a SIP with a thorough design and recommendation on this.
was (Author: janhoy):
Ah, did not know about that. A thousand questions pop up, such as: Should roles be supported in leader/follower mode? What would be the best way to designate roles to a node? A global file in ZK is one way. ENV_VARs or SysProps at each node is another. How dynamic do we need this to be, are there business needs to (re)assign roles at runtime, and in that case do we need a REST API for it? If dynamic, what role(s) would a brand new node (or k8s pod) added to the cluster be assigned? What happens if you have only assigned a role to one node and that node leaves the cluster? Should that be configurable, or would there be some cluster level plugin monitoring events, that makes such decisions?
Looking forward to seeing a SIP with a thorough design and recommendation on this.
> Concept of non-data nodes
> -------------------------
>
> Key: SOLR-15694
> URL: https://issues.apache.org/jira/browse/SOLR-15694
> Project: Solr
> Issue Type: New Feature
> Components: AutoScaling
> Reporter: Ishan Chattopadhyaya
> Priority: Major
>
> I think we should have first class support for starting a Solr node in a mode whereby no cores will be placed on them. These nodes are useful for certain scenarios:
> # Dedicated overseer nodes
> # Nodes where only plugins are installed and used (e.g. cluster/node level plugins)
> # Dedicated nodes for querying (more on this to come later).
> Today, to achieve this effect, one can:
> 1. start a node (which will make it join live_nodes and be immediately available for replica placement).
> 2. Put replica placement rules or autoscaling policies to prevent replicas from being placed there. This is not standardized, 8x has two ways to achieve this (replica placement rules and autoscaling framework), 9x has a new autoscaling framework.
> Proposing a start parameter for starting a node that starts the node in this configuration, and then internally this is handled appropriately (across 8x and 9x). This should be Kubernetes/Docker friendly as well, since it is easy to add an additional parameter for a startup (instead of putting things into autoscaling.json in ZK via init scripts).
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@solr.apache.org
For additional commands, e-mail: issues-help@solr.apache.org