[ 
https://issues.apache.org/jira/browse/CASSANDRA-17032?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17427898#comment-17427898
 ] 

Josh McKenzie commented on CASSANDRA-17032:
-------------------------------------------

{quote}I wonder if we should send quota exceeded warnings at the 65%, 75%, 85% 
marks (configurable)
{quote}
Was thinking along the same lines this morning. So if we add:
 * A soft cap (configurable per Keyspace)
 * A hard cap (configurable per Keyspace) (effectively this is all this first 
iteration does)
 * More visible behavior when at various caps (log warning, diag event(s), 
client warnings)
 * Maybe a couple JMX hooks to both set quota sizes and to reload quotas from 
CQL (break glass for operators)

That'd make it a much more flexible feature, and also give us the best of both 
worlds (lazy reactive eval on the way up so no read-before-write, then ability 
to deterministically change it if needed for operators when things are back in 
line). wdyt [~paulo]?

> Allow configuring max allowable disk usage quota by keyspace
> ------------------------------------------------------------
>
>                 Key: CASSANDRA-17032
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-17032
>             Project: Cassandra
>          Issue Type: Improvement
>          Components: Local/Other
>            Reporter: Josh McKenzie
>            Assignee: Josh McKenzie
>            Priority: Normal
>
> This is a similar idea (basically identical idea; didn't want to steamroll 
> that old ticket w/an update to description etc) to what was raised and closed 
> in CASSANDRA-5394. In multi-tenant situations it can be very helpful to 
> provide disk quotas per Keyspaces for users.
> We can do this via a low priority background job that periodically checks 
> disk usage and flips a bit on the Keyspaces when it's over as well as reverts 
> it when things settle back down to prevent needing the read-before-write a 
> hard realtime requirement would introduce. While this would of course allow 
> users to "burst" their usage past that limit in the time window, it would 
> still be sufficient for many use-cases.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to