[
https://issues.apache.org/jira/browse/LUCENE-2324?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12983720#action_12983720
]
Michael McCandless commented on LUCENE-2324:
--------------------------------------------
I like the grocery store analogy! Yes, just like that :)
bq. Every DWPT has its own private FieldInfos. When a segment is flushed the
DWPT uses its private FI and then it updates the original DW.fieldInfos (from
IW), which is a synchronized call.
OK that sounds good.
bq. Hmm, given that IW.flush() isn't synchronized anymore I assume this can
lead into a problem? E.g. the SegmentMerger gets a FieldInfos that's "newer"
than the list of segments it's trying to flush?
Yes... or, the FieldInfos changes (due to flush) while SegmentMerger is still
merging. Probably SR should make a deep copy of the FieldInfos when it starts?
bq. DW has a SegmentDeletes (pendingDeletes) which gets pushed to the last
segment. We only add delTerms to DW.pendingDeletes if we couldn't push it to
any DWPT. Btw. I think the whole pushDeletes business isn't working correctly
yet, I'm looking into it. I need to understand the code that coalesces the
deletes better.
OK :)
bq. How can it figure out to which docs the deletes to apply to? _1 and _2 are
probably gone a long time ago. If we apply the deletes to all of _3 this would
be a mistake too.
Hmmm... I think you're right (this is a problem).
I think we also have problems w/ updateDocument? That call is
supposed be atomic (ie the del & addDoc can never be separately
committed), but, I think if one DWPT (holding the del term) gets
flushed but another (holding the del term and the added doc) aborts
and then you commit, you could see the del "make it" but not the
addDocument?
Finally, we are wanting to allow out-of-order-merges soon... so
that eg BSMP becomes much simpler to implement & bring to core,
but, these buffered deletes also make that more complicated.
Gonna have to mull on this...
> Per thread DocumentsWriters that write their own private segments
> -----------------------------------------------------------------
>
> Key: LUCENE-2324
> URL: https://issues.apache.org/jira/browse/LUCENE-2324
> Project: Lucene - Java
> Issue Type: Improvement
> Components: Index
> Reporter: Michael Busch
> Assignee: Michael Busch
> Priority: Minor
> Fix For: Realtime Branch
>
> Attachments: LUCENE-2324-SMALL.patch, LUCENE-2324-SMALL.patch,
> LUCENE-2324-SMALL.patch, LUCENE-2324-SMALL.patch, LUCENE-2324-SMALL.patch,
> LUCENE-2324.patch, LUCENE-2324.patch, LUCENE-2324.patch, lucene-2324.patch,
> lucene-2324.patch, LUCENE-2324.patch, test.out, test.out, test.out, test.out
>
>
> See LUCENE-2293 for motivation and more details.
> I'm copying here Mike's summary he posted on 2293:
> Change the approach for how we buffer in RAM to a more isolated
> approach, whereby IW has N fully independent RAM segments
> in-process and when a doc needs to be indexed it's added to one of
> them. Each segment would also write its own doc stores and
> "normal" segment merging (not the inefficient merge we now do on
> flush) would merge them. This should be a good simplification in
> the chain (eg maybe we can remove the *PerThread classes). The
> segments can flush independently, letting us make much better
> concurrent use of IO & CPU.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
