[
https://issues.apache.org/jira/browse/IGNITE-10307?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16745175#comment-16745175
]
Vladimir Ozerov commented on IGNITE-10307:
------------------------------------------
Both failures are false-positive.
https://ci.ignite.apache.org/viewQueued.html?itemId=2825852
https://ci.ignite.apache.org/viewQueued.html?itemId=2825850
> SQL: Extract partition info from JOINs
> --------------------------------------
>
> Key: IGNITE-10307
> URL: https://issues.apache.org/jira/browse/IGNITE-10307
> Project: Ignite
> Issue Type: Task
> Components: sql
> Reporter: Vladimir Ozerov
> Assignee: Vladimir Ozerov
> Priority: Major
> Labels: iep-24
> Fix For: 2.8
>
> Time Spent: 0.5h
> Remaining Estimate: 0h
>
> Currently we do not extract partitions when JOINs are involved. Let's
> implement it. We may start with relatively simple rules:
> # No subqueries
> # No GROUP BY
> Then walk through JOINed tables and extract partitions from AND clauses.
> There are some tricky things to consider:
> # Resulting model (tree) must be craefted carefully so that we can reuse it
> later in thin clients for efficient co-location.
> # Resulting model may affect how we group tables during push-down phase.
> Probably this would be huuuge thing, so may be it is better to implement it
> in separate ticket
> # When JOIN is performed partition info might be "transferred" between
> tables. E.g.:
> {code}
> a INNER JOIN b ON a.id = b.affinity_id WHERE a.id = :1
> {code}
> In this case if tables are co-located (we may infer it automatically in some
> cases), then {{a.id=:1}} partition rule can be "transferred" to
> {{b.affinity_id=:1}}.
> Very good test coverage would be needed here.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
