[
https://issues.apache.org/jira/browse/YUNIKORN-165?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17503972#comment-17503972
]
Craig Condit commented on YUNIKORN-165:
---------------------------------------
I agree completely on the configuration side. I think if we default to M =
10^6 as the default units for memory and no unit = 1000 millicore for vcore we
should be good. We can consider all other resource types as untyped and use no
units by default, but still allow 1M = 1,000,000 for the untyped units. As for
int64 limits, a 1 PB disk size per node is pretty massive today which still
allows for 9000 nodes at that size, so I think we’re ok for a while. This would
also be around $150MM USD for disk storage with no redundancy so asking users
to spin up a second cluster if they hit this is probably ok. :)
Also agree the REST API should use untyped units exclusively and consider the
case u it if vcore to be a millicpu.
> API response fields can be returned as proper data structure rather than free
> text
> ----------------------------------------------------------------------------------
>
> Key: YUNIKORN-165
> URL: https://issues.apache.org/jira/browse/YUNIKORN-165
> Project: Apache YuniKorn
> Issue Type: Improvement
> Components: core - common, webapp
> Reporter: Ayub Pathan
> Priority: Critical
>
> For most of the APIs, some of the fields are recorded as free text, this can
> be improved by sending them as proper data structure like map etc. This helps
> in easy consumption of the REST aPI response.
> For example:
> * Some fields from Apps API like resource,usedResource, partition can be
> returned as maps..
> {noformat}
> {
> "allocations": [
> {
> "allocationKey": "1e5225f7-4d60-400f-8b9b-5b7027b4c394",
> "allocationTags": null,
> "applicationId": "abcd",
> "nodeId": "docker-desktop",
> "partition": "default",
> "priority": "<nil>",
> "queueName": "root.development",
> "resource": "[memory:100 vcore:200]",
> "uuid": "b1b89eaf-81fb-4b2b-b20a-bf7bd96d3430"
> }
> ],
> "applicationID": "abcd",
> "applicationState": "Running",
> "partition": "[mycluster]default",
> "queueName": "root.development",
> "submissionTime": 1589740021170274400,
> "usedResource": "[memory:100 vcore:200]"
> }, {noformat}
>
> * same for nodes API response(fields like available, capacity, occupied,
> allocated) as well.
> {noformat}
> [
> {
> "nodesInfo": [
> {
> "RackName": "/rack-default",
> "allocated": "[memory:0 vcore:0]",
> "allocations": null,
> "available": "[ephemeral-storage:56450230179 hugepages-1Gi:0
> hugepages-2Mi:0 memory:1824 pods:110 vcore:3250]",
> "capacity": "[ephemeral-storage:56450230179 hugepages-1Gi:0
> hugepages-2Mi:0 memory:1982 pods:110 vcore:4000]",
> "hostName": "docker-desktop",
> "nodeID": "docker-desktop",
> "occupied": "[memory:158 vcore:750]",
> "schedulable": true
> }
> ],
> "partitionName": "[mycluster]default"
> }
> ] {noformat}
>
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
