[ https://issues.apache.org/jira/browse/LUCENE-2881?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13001011#comment-13001011
]
Michael McCandless commented on LUCENE-2881:
--------------------------------------------
This is a big patch! Some comments...:
* Do we even use FieldInfos.add(Document)? Maybe we can remove
it...
* FieldNumberBiMap.addOrGet doesn't need to take the FieldInfoBiMap?
(It's an unused param).
* Why create the FieldInfoBiMap class? Ie, why not "merge" that
back into FieldInfos itself (like it used to be)? (I understand
why we need FieldNumberBiMap -- so we can share a single instance
across FieldInfos).
* We mix up autoboxing then unboxing, eg
FieldInfos.addOrUpdateInternal takes int preferredFieldNumber,
which is boxed when calling localFieldInfos.nextFieldNumber, then
manually unboxed on calling globalFieldNumbers.addOrGet.
* On working with a pre-4.0 index that has non-congruent assignments
across segments... I fear we may not necessarily ever "stabilize" on
a fixed global name/number bimap, because we re-compute this map
on every IW init? Ie, when you first open 4.0 IW on pre-4.0
index, you'll compute a certain global map, and write new segs
with those bindings. But say some fields differed in their
assignment... but then some of those conflicting segs are merged.
Later, when you open the IW again, you'll get a different global
map? And write new segments conflicting with the previous new
segments you had written?
* The fact that SegmentInfo.clearFilesCache is now a public API and
consumer is responsible for knowing when to call this
is... spooky. Previously this cache was a fully private thing (ie
invalidated whenever a change was made to the SegmentInfo). But I
don't see any way around it; since we now embed a FieldInfos and
that FieldInfos (hasVectors) could change... maybe, whenever we
call this method, add a comment explaining why?
* It makes me nervous that the API that's allowed to pick a new
field number (FieldInfo.addInternal) is the same API that used
when reading a FieldInfos from _X.fnm (when we better not pick a
different field number!). In theory of course those
field numbers will never conflict and we'll always get our
preferred field number... but still. Maybe add an assert in
FieldInfos.read() verifying we always get that field number?
* The call to localFieldInfos.setIfNotSet in FieldInfos.addInternal
makes me nervous... is it actually possible for it to already be
set, to a conflicting binding? Shouldn't it always match? (Ie,
above, in addOrUpdateInternal, we just consulted the global map to
get the binding). Can't we assert the global binding is either
not present (and we add it) or if it is present it "matches"?
* Why do we have FieldInfos.clearVectors? Nobody should call
that...?
* It's not great that we open then close the CFS reader inside
SegmentInfo, just to read the FieldInfos. Ie, this means on
opening an SR we will open this CFS reader twice... it also means
that opening a SegmentInfos is quite a bit more costly than it
used to be. EG creating an IW must now go and open/close a CFS reader
per-segment... not sure what we can really do about that
though... maybe, we should store the FieldInfos inside the
segments file? Hmmm....
* Shouldn't IndexWriter.getFieldInfos(SegmentInfo) use the
SegmentInfo's fieldInfos rather than loading it again from the
directory...?
Indentation is also off in various places, for us lonely people who
still use Emacs ;)
> Track FieldInfo per segment instead of per-IW-session
> -----------------------------------------------------
>
> Key: LUCENE-2881
> URL: https://issues.apache.org/jira/browse/LUCENE-2881
> Project: Lucene - Java
> Issue Type: Improvement
> Affects Versions: Realtime Branch, CSF branch, 4.0
> Reporter: Simon Willnauer
> Assignee: Michael Busch
> Fix For: Realtime Branch, CSF branch, 4.0
>
> Attachments: LUCENE-2881.patch, lucene-2881.patch, lucene-2881.patch, lucene-2881.patch,
lucene-2881.patch, lucene-2881.patch
>
>
> Currently FieldInfo is tracked per IW session to guarantee consistent global field-naming
/ ordering. IW carries FI instances over from previous segments which also carries over field
properties like isIndexed etc. While having consistent field ordering per IW session appears
to be important due to bulk merging stored fields etc. carrying over other properties might
become problematic with Lucene's Codec support. Codecs that rely on consistent properties
in FI will fail if FI properties are carried over.
> The DocValuesCodec (DocValuesBranch) for instance writes files per segment and field
(using the field id within the file name). Yet, if a segment has no DocValues indexed in a
particular segment but a previous segment in the same IW session had DocValues, FieldInfo#docValues
will be true since those values are reused from previous segments.
> We already work around this "limitation" in SegmentInfo with properties like hasVectors
or hasProx which is really something we should manage per Codec & Segment. Ideally FieldInfo
would be managed per Segment and Codec such that its properties are valid per segment. It
also seems to be necessary to bind FieldInfoS to SegmentInfo logically since its really just
per segment metadata.
--
This message is automatically generated by JIRA.
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org
|