[ 
https://issues.apache.org/jira/browse/CASSANDRA-18796?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17759076#comment-17759076
 ] 

Jeremiah Jordan edited comment on CASSANDRA-18796 at 8/25/23 3:18 PM:
----------------------------------------------------------------------

I really don't think we should fail queries just because someone is using LCS.  
-1 on that hard failure.  You can change your LCS file size and then you will 
not have thousands of sstables.  It all comes down to tuning.

I think it would be more productive to issue a WARN if the number of sstables 
searched is over some threshold, and then suggest to the user that they should 
look into tuning their compaction options differently.


was (Author: jjordan):
I really don't think we should fail queries just because someone is using LCS.  
You can change your LCS file size and then you will not have thousands of 
sstables.  It all comes down to tuning.  I think it would be more productive to 
issue a WARN if the number of sstables searched is over some threshold, and 
then suggest to the user that they should look into tuning their compaction 
options differently.

> Optionally fail when a non-partition-restricted query is issued against a 
> storage-attached index with a backing table using LCS
> -------------------------------------------------------------------------------------------------------------------------------
>
>                 Key: CASSANDRA-18796
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-18796
>             Project: Cassandra
>          Issue Type: Improvement
>          Components: Feature/2i Index, Feature/SAI, Local/Compaction/LCS
>            Reporter: Caleb Rackliffe
>            Assignee: Caleb Rackliffe
>            Priority: Normal
>             Fix For: 5.0.x, 5.x
>
>
> With LCS, we will have potentially thousands of SSTables for a given user 
> table. Storage-attached also means SSTable-attached, and searching thousands 
> of attached indexes is not going to scale well at all locally, due to the 
> sheer number of searches and amount of postings list merging involved. We 
> should have a guardrail to prohibit this by default.
> Partition-restricted queries, the use-case SAI is broadly designed for, 
> should be very efficient.



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

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

Reply via email to