[
https://issues.apache.org/jira/browse/LUCENE-8381?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16532192#comment-16532192
]
Nhat Nguyen commented on LUCENE-8381:
-------------------------------------
[~simonw] Thanks for looking and providing a solution for this.
I continue your patch to optimize the way we calculate numDocs. The calculation
now loops over only the soft-deleted docs, not all docs (please see
LUCENE-8381-inline.patch).
Another option that I see is to pass the hardLiveDocs to the soft-deletes merge
policy. This approach is cleaner but requires other custom MP to handle the
hard-deletes by itself (please see LUCENE-8381-merge.patch).
> PendingSoftDeletes#onNewReader should adjust soft-delete count
> --------------------------------------------------------------
>
> Key: LUCENE-8381
> URL: https://issues.apache.org/jira/browse/LUCENE-8381
> Project: Lucene - Core
> Issue Type: Bug
> Affects Versions: 7.4
> Reporter: Nhat Nguyen
> Assignee: Simon Willnauer
> Priority: Major
> Fix For: master (8.0), 7.5, 7.4.1
>
> Attachments: LUCENE-8381-inline.patch, LUCENE-8381-merge.patch,
> LUCENE-8381.patch, LUCENE-8381.patch, merge-soft-hard-deletes.patch
>
>
> The current PendingSoftDeletes#onNewReader does not handle these two edge
> cases correctly.
> # If the soft-deletes DV was applied to PendingSoftDeletes already,
> newDelCount should be less than the soft delete count recorded in
> SegmentInfo. testSoftDeleteWithTryUpdateDocValue trips the newSoftDelCount
> assertion.
> # If the soft-deleted docs were hard-deleted, we should reduce the
> softDelCount in the segmentInfo on a new reader.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
