[ https://issues.apache.org/jira/browse/STRATOS-716?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14169244#comment-14169244 ]
Nirmal Fernando commented on STRATOS-716: ----------------------------------------- I've implemented this for Kubernetes - Docker scenario. Implementation Details: Autoscaler provides an API to update a cluster monitor created for a particular cluster, by accepting set of properties. Autoscaler API implementation delegates such a request to relevant cluster monitor via AbstractClusterMonitor. It's up to that particular Cluster Monitor to decide, what properties to read and what properties it'll update. REST API curl -X PUT -H "Content-Type: application/json" -d @manual-scaling.json -k -v -u admin:admin https://localhost:9443/stratos/admin/subscriptions/{subscriptionAlias}/properties Sample manual-scaling file: { "property": [ { "name": "KUBERNETES_REPLICAS_MIN", "value": "4" } ] } CLI command: stratos> help update-subscription Update a previously made subscription. usage: update-subscription [cartridge-subscription-alias] [-p <resource path>] -p,--resource-path <resource path> Update subscription properties resource path stratos> update-subscription myphp --resource-path /Volumes/wso2/stratos-4.1.0/M1/manual-scaling.json Successfully updated subscription alias: myphp We can improve this feature further. > Support for manual scaling > -------------------------- > > Key: STRATOS-716 > URL: https://issues.apache.org/jira/browse/STRATOS-716 > Project: Stratos > Issue Type: New Feature > Components: Autoscaler > Reporter: Nirmal Fernando > Assignee: Nirmal Fernando > Fix For: 4.1.0 M2 > > > It's important that we have an option to scale nodes manually, so that as a > user you could prepare for high load before hand. While implementing this > feature, it's also important to decide whether we should consider these > manually spawned instances for the auto-scaling decision. IMO we shouldn't do > that, otherwise even if we manually spin up instances, they'll be terminated > by the auto-scaler, if there's no actual load. One other important is whether > we should consider stats of these manually scaled nodes. IMO again we > shouldn't do that, otherwise it'll lead the auto-scaler to take incorrect > decisions. We need to discuss these points and also other important points in > the mailing list and decide. -- This message was sent by Atlassian JIRA (v6.3.4#6332)