[
https://issues.apache.org/jira/browse/CASSANDRA-19367?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17814491#comment-17814491
]
David Capwell commented on CASSANDRA-19367:
-------------------------------------------
Here are the places isVector is called
* org.apache.cassandra.index.sai.memory.MemtableIndexManager#update - I don't
need "update", but not sure why we even special case vector here? removing
"index" in favor of "update" seems fine to me. I also see we don't update
"memtableIndexWriteLatency" in the vectors case... so cleaning that up would
fix this metric?
*
org.apache.cassandra.index.sai.plan.StorageAttachedIndexQueryPlan#StorageAttachedIndexQueryPlan
- this is checking if any index is a factor, if so we are "top-key"... Thats
super specific but mostly ignored in my POC as I query SAI lower level than CQL
so I avoid post filtering and loading the partition/row
*
org.apache.cassandra.index.sai.disk.v1.V1OnDiskFormat#perColumnIndexComponents
- Just saw that my POC adds accord here, but I missed refactoring this to
Strat... still need to do that for this patch
* org.apache.cassandra.index.sai.disk.v1.IndexWriterConfig#fromOptions - would
be nice to leverage Strat here, but don't need for my use case as its a
internal table with an internal index... I validate w/e you try to construct
the index
* Several cases in StorageAttachedIndex and IndexTermType
I need to fix the v1 format as that does impact my POC, but open to other
places depending on feedback
> Refactor SAI so the selection of the index type is not scattered to multiple
> places
> -----------------------------------------------------------------------------------
>
> Key: CASSANDRA-19367
> URL: https://issues.apache.org/jira/browse/CASSANDRA-19367
> Project: Cassandra
> Issue Type: Improvement
> Components: Feature/2i Index
> Reporter: David Capwell
> Assignee: David Capwell
> Priority: Normal
> Fix For: 5.x
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> For Accord we want to write an internal index and finding plugging into SAI
> is a bit more channeling than it could be… we need to find multiple places
> where the SAI code “infer” the index type so it can delegate… this logic
> should be done once and made pluggable so custom SAI indexes can be defined
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
