[jira] [Commented] (CASSANDRA-21179) Introduce a check for minimum time passed before importing / training of a compression dictionary is allowed

["Stefan Miklosovic (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/CASSANDRA-21179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18060491#comment-18060491 ] 

Stefan Miklosovic commented on CASSANDRA-21179:
-----------------------------------------------

https://pre-ci.cassandra.apache.org/job/cassandra/423/#showFailuresLink

> Introduce a check for minimum time passed before importing / training of a compression dictionary is allowed
> ------------------------------------------------------------------------------------------------------------
>
>                 Key: CASSANDRA-21179
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-21179
>             Project: Apache Cassandra
>          Issue Type: Improvement
>          Components: Feature/Compression
>            Reporter: Stefan Miklosovic
>            Assignee: Stefan Miklosovic
>            Priority: Normal
>             Fix For: 5.x
>
>          Time Spent: 10m
>  Remaining Estimate: 0h
>
> After CASSANDRA-21178 where we introduced created_at column to dict table, we can move forward with the check which ensures that a user is not over-training, or they can not train / import at all. 
> 1) by default, there would be a minimum time between training / importing is enabled
> 2) Default would be, for example, 1 hour, and it would be configurable
> 3) The configuration would be done CQL level in compression parameters, that way all nodes would see same values due to TCM schema propagation, so this can not be attacked by "round-robin" training, node by node
> 4) If we want to disable training / importing completely, we would set minimum training frequency to "0" (0h, 0s ...). This means that if we want to allow training again, an operator would need to raise this value, then the last training would still need to occur at least so and so time back to be able to train again. 
> It would be better if "no training allowed" is represented by high value, like 100 years, (36500 days) instead of 0. Setting it to 0 seems like an exception / corner case which is not intuitive to follow, as one would say that 0 means that we can train whenever we want. 0 should be the default. Not 1 hour. 



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@cassandra.apache.org
For additional commands, e-mail: commits-help@cassandra.apache.org