[jira] [Comment Edited] (LUCENE-7976) Make TieredMergePolicy respect maxSegmentSizeMB and allow singleton merges of very large segments

["Erick Erickson (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/LUCENE-7976?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16429433#comment-16429433 ] 

Erick Erickson edited comment on LUCENE-7976 at 4/7/18 4:12 PM:
----------------------------------------------------------------

Thanks for taking the time Mike! Going in reverse order:

bq: I wonder if we could just relax TMP to allow it to consider merges with fewer than maxMergeAtOnce, and then "improve" the scoring function to give a good score to cases that would reclaim > X% deletions?

Interesting. At this point in the change cycle I've got my head around most of what's going on and can think about how to do it better. We'd want to tweak things I should think so scoring could return "don't do this merge at all"? I'm thinking of the case where we have, say, 1 max-sized (or bigger) segment as a candidate with a few deletions, we wouldn't want to merge that at all, right? I'm thinking some threshold score above which we score it as "don't bother"....

The number of changes I'm introducing here does make me nervous, I wonder if taking a fresh look at it with an eye toward just doing the above would lead to less surgery.... I mean this has been working fine for years, I do worry that I'm introducing bugs... I don't mind throwing away a bunch of work if smaller changes can cure my problem.

bq: I think what you mean is you want to change the forceMerge and forceMergeDeletes APIs in IndexWriter

Right, that would have been the consequence. But I changed my mind on this yesterday, I don't think any Lucene API change is needed after all. What I did instead (not in the current patch) is default the Solr "update" command to pass Integer.MAX_VALUE for the max number of segments to forceMerge. That just flows into the TMP code without changing the API and lets maxMergedSegmentBytes control how many segments are created. Anyone who wants the old behavior needs to pass 1 like the default is now.

bq:  I think it's a bug that findForceMergeDeletes doesn't do the same thing....

OK, let me look this over again. Yesterday I started to see the differences between forceMerge and forceMergeDeletes and thought they should stay separate, but you seem to be saying the idea of combining them is worth exploring. I'll revisit this again this weekend. Wouldn't making that work require changing the findForceMergeDeletes interface? I'm perfectly willing but didn't want to do that without discussion. And it seems that then findForcedDeletesMerges and findForcedMerges would be very thin wrappers around the same code for both.... Or were you thinking of handling this differently?

Thanks again...


was (Author: erickerickson):
Thanks for taking the time Mike! Going in reverse order:

bq: I wonder if we could just relax TMP to allow it to consider merges with fewer than maxMergeAtOnce, and then "improve" the scoring function to give a good score to cases that would reclaim > X% deletions?

Interesting. At this point in the change cycle I've got my head around most of what's going on and can think about how to do it better. We'd want to tweak things I should think so scoring could return "don't do this merge at all"? I'm thinking of the case where we have, say, 1 max-sized (or bigger) segment as a candidate with a few deletions, we wouldn't want to merge that at all, right? I'm thinking some threshold score above which we score it as "don't bother"....

The number of changes I'm introducing here does make me nervous, I wonder if taking a fresh look at it with an eye toward just doing the above would lead to less surgery.... I mean this has been working fine for years, I do worry that I'm introducing bugs... I don't mind throwing away a bunch of work if smaller changes can cure my problem.

bq: I think what you mean is you want to change the forceMerge and forceMergeDeletes APIs in IndexWriter

Right, that would have been the consequence. But I changed my mind on this yesterday, I don't think any Lucene API change is needed after all. What I did instead (not in the current patch) is default the Solr "update" command to Integer.MAX_VALUE for forceMerge. That just flows into the TMP code without changing the API and lets maxMergedSegmentBytes control how many segments are created. Anyone who wants the old behavior needs to pass 1, which is a change in behavior.

bq:  I think it's a bug that findForceMergeDeletes doesn't do the same thing....

OK, let me look this over again. Yesterday I started to see the differences between forceMerge and forceMergeDeletes and thought they should stay separate, but you seem to be saying the idea of combining them is worth exploring. I'll revisit this again this weekend. Wouldn't making that work require changing the findForceMergeDeletes interface? I'm perfectly willing but didn't want to do that without discussion. And it seems that then findForcedDeletesMerges and findForcedMerges would be very thin wrappers around the same code for both.... Or were you thinking of handling this differently?

Thanks again...

> Make TieredMergePolicy respect maxSegmentSizeMB and allow singleton merges of very large segments
> -------------------------------------------------------------------------------------------------
>
>                 Key: LUCENE-7976
>                 URL: https://issues.apache.org/jira/browse/LUCENE-7976
>             Project: Lucene - Core
>          Issue Type: Improvement
>            Reporter: Erick Erickson
>            Assignee: Erick Erickson
>            Priority: Major
>         Attachments: LUCENE-7976.patch, LUCENE-7976.patch
>
>
> We're seeing situations "in the wild" where there are very large indexes (on disk) handled quite easily in a single Lucene index. This is particularly true as features like docValues move data into MMapDirectory space. The current TMP algorithm allows on the order of 50% deleted documents as per a dev list conversation with Mike McCandless (and his blog here:  https://www.elastic.co/blog/lucenes-handling-of-deleted-documents).
> Especially in the current era of very large indexes in aggregate, (think many TB) solutions like "you need to distribute your collection over more shards" become very costly. Additionally, the tempting "optimize" button exacerbates the issue since once you form, say, a 100G segment (by optimizing/forceMerging) it is not eligible for merging until 97.5G of the docs in it are deleted (current default 5G max segment size).
> The proposal here would be to add a new parameter to TMP, something like <maxAllowedPctDeletedInBigSegments> (no, that's not serious name, suggestions welcome) which would default to 100 (or the same behavior we have now).
> So if I set this parameter to, say, 20%, and the max segment size stays at 5G, the following would happen when segments were selected for merging:
> > any segment with > 20% deleted documents would be merged or rewritten NO MATTER HOW LARGE. There are two cases,
> >> the segment has < 5G "live" docs. In that case it would be merged with smaller segments to bring the resulting segment up to 5G. If no smaller segments exist, it would just be rewritten
> >> The segment has > 5G "live" docs (the result of a forceMerge or optimize). It would be rewritten into a single segment removing all deleted docs no matter how big it is to start. The 100G example above would be rewritten to an 80G segment for instance.
> Of course this would lead to potentially much more I/O which is why the default would be the same behavior we see now. As it stands now, though, there's no way to recover from an optimize/forceMerge except to re-index from scratch. We routinely see 200G-300G Lucene indexes at this point "in the wild" with 10s of  shards replicated 3 or more times. And that doesn't even include having these over HDFS.
> Alternatives welcome! Something like the above seems minimally invasive. A new merge policy is certainly an alternative.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org