Re: SortingMergePolicy for already sorted segments
Shai, This is the code snippet I use inside my class... public class MySorter extends Sorter { @Override public DocMap sort(AtomicReader reader) throws IOException { final MapInteger, BytesRef docVsId = loadSortTerm(reader); final Sorter.DocComparator comparator = new Sorter.DocComparator() { @Override public int compare(int docID1, int docID2) { BytesRef v1 = docVsId.get(docID1); BytesRef v2 = docVsId.get(docID2); return v1.compareTo(v2); } }; return sort(reader.maxDoc(), comparator); } } My Problem is, the AtomicReader passed to Sorter.sort method is actually a SlowCompositeReader, composed of a list of AtomicReaders each of which is already sorted. I find this loadSortTerm(compositeReader) to be a bit heavy where it tries to all load the doc-to-term mappings eagerly... Are there some alternatives for this? -- Ravi On Tue, Jun 17, 2014 at 10:58 AM, Shai Erera ser...@gmail.com wrote: I'm not sure that I follow ... where do you see DocMap being loaded up front? Specifically, Sorter.sort may return null of the readers are already sorted ... I think we already optimized for the case where the readers are sorted. Shai On Tue, Jun 17, 2014 at 4:04 AM, Ravikumar Govindarajan ravikumar.govindara...@gmail.com wrote: I am planning to use SortingMergePolicy where all the merge-participating segments are already sorted... I understand that I need to define a DocMap with old-new doc-id mappings. Is it possible to optimize the eager loading of DocMap and make it kind of lazy load on-demand? Ex: Pass ListAtomicReader to the caller and ask for next new-old doc mapping.. Since my segments are already sorted, I could save on memory a little-bit this way, instead of loading the full DocMap upfront -- Ravi
RE: Lucene Upgrade from 2.9.x to 4.7.x
Hi, Thanks Uwe. I tried this path and I do not find any .cfs files. Lucene 3 and Lucene 4 indexes do not necessarily always contain CFS files, especially not if they are optimized. This depends on the merge policy. The index upgrader uses the default one, which creates no CFS files for the largest segment of an index. As there is only one after the upgrade, it is not in compound format. All that I see in my index directory after running upgrader is following files. -rw--- 1 root root 245 Jun 16 22:38 _1.fdt -rw--- 1 root root 45 Jun 16 22:38 _1.fdx -rw--- 1 root root 2809 Jun 16 22:38 _1.fnm -rw--- 1 root root 487 Jun 16 22:38 _1_Lucene41_0.doc -rw--- 1 root root 34 Jun 16 22:38 _1_Lucene41_0.pay -rw--- 1 root root 3999 Jun 16 22:38 _1_Lucene41_0.pos -rw--- 1 root root 5575 Jun 16 22:38 _1_Lucene41_0.tim -rw--- 1 root root 834 Jun 16 22:38 _1_Lucene41_0.tip -rw--- 1 root root 110 Jun 16 22:38 _1.nvd -rw--- 1 root root 343 Jun 16 22:38 _1.nvm -rw--- 1 root root 419 Jun 16 22:38 _1.si That looks perfectly fine, although the index is very small. This is already the 4.x index - how did the Lucene 3.6 index look like? The size of the index should be in the same magnitude like before the upgrade. My search query returns zero object. Can someone help me here. The reason for this can be changes in the analysis. Lucene searches only work, if the index and query analysis are compatible, which is not guaranteed with such a gap in Lucene versions. Please make sure that you use same analyzers before and after the upgrade with same matchVersion parameter (in your case you would need to pass Version.LUCENE_2_9 parameter to your analyzer, which is no longer available in Lucene 4). It depends on the behavior anaylyzer that was used before, if it is possible to easily upgrade without reindexing all the data. E.g., StandardAnalyzer changed its behavior to be Unicode conform in Lucene 3.x. This makes it incompatible for some queries, but simple ones still work. Uwe - To unsubscribe, e-mail: java-user-unsubscr...@lucene.apache.org For additional commands, e-mail: java-user-h...@lucene.apache.org
Search degradation on Windows when upgrading from lucene 3.6 to lucene 4.7.2
Hi, We are in the process of upgrading from lucene 3.6.0 to lucene 4.7.2, and our tests show a significant search degradation on Windows platform. Trying to figure this out, here are a couple of points we noticed. Any suggestions/thoughts will be greatly appreciated. Thanks! 1) Running search on an optimized collection. Our first run on Windows machine showed the following results: Lucene 3.6: 115 queries / sec Lucene 4.7.2: 74 queries / sec Looking at the collections themselves, we got the following characterization: Lucene 3.6 General Index Information: == Num docs: 10485760 Num deleted docs: 0 Deletion rate: 0% Number of files in FOLDER: 116 Total size of files in FOLDER: 81558862032 bytes (75.96 GB) Commit Point Information: = Version: 1399567203042 Timestamp: 1399593668185 Generation: 6018 Segments file name: segments_4n6 Number of segments: 32 Committed size: 81216915273 bytes (75.64 GB) Number of files in COMMIT POINT: 89 Total size of files in COMMIT POINT: 81216923390 bytes (75.64 GB) Lucene 4.7.2: General Index Information: == Num docs: 10485760 Num deleted docs: 0 Deletion rate: 0% Number of files in FOLDER: 301 Total size of files in FOLDER: 71019073768 bytes (66.14 GB) Commit Point Information: = Generation: 4518 Segments file name: segments_3hi Number of segments: 38 Committed size: 70635339707 bytes (65.78 GB) Number of files in COMMIT POINT: 115 Total size of files in COMMIT POINT: 70635341223 bytes (65.78 GB) We saw that the collection created by lucene 4.7.2 was10GB smaller but it had a more segments. We thought that more segments might account to the search degradation, and so we decided to run optimization on the 4.7.2 index before rerunning the search test. The index was more compact: Lucene 4.7.2 General Index Information: == Num docs: 10485760 Num deleted docs: 0 Deletion rate: 0% Number of files in FOLDER: 38 Total size of files in FOLDER: 70488334388 bytes (65.65 GB) Commit Point Information: = Generation: 4519 Segments file name: segments_3hj Number of segments: 12 Committed size: 70488333864 bytes (65.65 GB) Number of files in COMMIT POINT: 37 Total size of files in COMMIT POINT: 70488334368 bytes (65.65 GB) And as expected, the search results were much better: 4.7.2. 118 queries / sec We thought that this might be a good direction, so our next step was to simulate a more compact index as part of our indexing session without running a full optimize at the end. To do that we changed maxMergeMB from 4 GB to 6 GB. The collection was indeed more compact: Win64 4.7.2 merge=6000 commitPoints: General Index Information: == Num docs: 10485760 Num deleted docs: 0 Deletion rate: 0% Number of files in FOLDER: 213 Total size of files in FOLDER: 83038952682 bytes (77.34 GB) Commit Point Information: = Generation: 4406 Segments file name: segments_3ee Number of segments: 14 Committed size: 70324985193 bytes (65.50 GB) Number of files in COMMIT POINT: 91 Total size of files in COMMIT POINT: 70324985781 bytes (65.50 GB) But search results were not good at all: 4.7.2: 72 queries / sec Does this make sense? We thought of Optimize as mainly decreasing the number of segments in the collection, and removing deletions. In this scenario, we had no deletions, and we saw that the number of segments did in fact decrease substantially, So why are we not seeing this reflect in search performance? Is there any other optimize influence/hidden-operation that we are missing here? (Note that we are using LogByteSizeMergePolicy. We know that TieredMergePolicy is suppose to be better in this aspect, but it is important to us To keep the order of the documents the same between commit points... ) 2) Search Directory On Lucene 3.6, we did comprehensive testing and saw that the best search performance is reached when using an Mmap directory. (for Indexing we are using SimpleFSDirectory). We tried different directories again with lucene 4.7.2, and while the differences were not big, it still seems that Mmap is no longer the best option: Lucene 4.7.2 with MMap: 72 queries / sec Lucene 4.7.2 with SimpleFS: 84 queries / sec Was there any changes around the MMap directory that might account for this difference? If so, do you think that those changes might account for the overall performance we are seeing? 3) Java 6 / Java 7 We are currently running on Java 6 (that is also the reason we stopped at lucene 4.7.2 and not 4.8). Is there a reason to believe that the degradation might be connected to this? Thanks again in advance!
Re: SortingMergePolicy for already sorted segments
I am afraid the DocMap still maintains doc-id mappings till merge and I am trying to avoid it... I think lucene itself has a MergeIterator in o.a.l.util package. A MergePolicy can wrap a simple MergeIterator for iterating docs across different AtomicReaders in correct sort-order for a given field/term That should be fine right? -- Ravi -- Ravi On Tue, Jun 17, 2014 at 1:24 PM, Shai Erera ser...@gmail.com wrote: loadSortTerm is your method right? In the current Sorter.sort implementation, I see this code: boolean sorted = true; for (int i = 1; i maxDoc; ++i) { if (comparator.compare(i-1, i) 0) { sorted = false; break; } } if (sorted) { return null; } Perhaps you can write similar code? Also note that the sorting interface has changed, I think in 4.8, and now you don't really need to implement a Sorter, but rather pass a SortField, if that works for you. Shai On Tue, Jun 17, 2014 at 9:41 AM, Ravikumar Govindarajan ravikumar.govindara...@gmail.com wrote: Shai, This is the code snippet I use inside my class... public class MySorter extends Sorter { @Override public DocMap sort(AtomicReader reader) throws IOException { final MapInteger, BytesRef docVsId = loadSortTerm(reader); final Sorter.DocComparator comparator = new Sorter.DocComparator() { @Override public int compare(int docID1, int docID2) { BytesRef v1 = docVsId.get(docID1); BytesRef v2 = docVsId.get(docID2); return v1.compareTo(v2); } }; return sort(reader.maxDoc(), comparator); } } My Problem is, the AtomicReader passed to Sorter.sort method is actually a SlowCompositeReader, composed of a list of AtomicReaders each of which is already sorted. I find this loadSortTerm(compositeReader) to be a bit heavy where it tries to all load the doc-to-term mappings eagerly... Are there some alternatives for this? -- Ravi On Tue, Jun 17, 2014 at 10:58 AM, Shai Erera ser...@gmail.com wrote: I'm not sure that I follow ... where do you see DocMap being loaded up front? Specifically, Sorter.sort may return null of the readers are already sorted ... I think we already optimized for the case where the readers are sorted. Shai On Tue, Jun 17, 2014 at 4:04 AM, Ravikumar Govindarajan ravikumar.govindara...@gmail.com wrote: I am planning to use SortingMergePolicy where all the merge-participating segments are already sorted... I understand that I need to define a DocMap with old-new doc-id mappings. Is it possible to optimize the eager loading of DocMap and make it kind of lazy load on-demand? Ex: Pass ListAtomicReader to the caller and ask for next new-old doc mapping.. Since my segments are already sorted, I could save on memory a little-bit this way, instead of loading the full DocMap upfront -- Ravi
Re: SortingMergePolicy for already sorted segments
I am afraid the DocMap still maintains doc-id mappings till merge and I am trying to avoid it... What do you mean 'till merge'? The method OneMerge.getMergeReaders() is called only when the merge is executed, not when the MergePolicy decided to merge those segments. Therefore the DocMap is initialized only when the merge actually executes ... what is there more to postpone? And besides, if the segments are already sorted, you should return a null DocMap, like Lucene code does ... If I miss your point, I'd appreciate if you can point me to a code example, preferably in Lucene source, which demonstrates the problem. Shai On Tue, Jun 17, 2014 at 3:03 PM, Ravikumar Govindarajan ravikumar.govindara...@gmail.com wrote: I am afraid the DocMap still maintains doc-id mappings till merge and I am trying to avoid it... I think lucene itself has a MergeIterator in o.a.l.util package. A MergePolicy can wrap a simple MergeIterator for iterating docs across different AtomicReaders in correct sort-order for a given field/term That should be fine right? -- Ravi -- Ravi On Tue, Jun 17, 2014 at 1:24 PM, Shai Erera ser...@gmail.com wrote: loadSortTerm is your method right? In the current Sorter.sort implementation, I see this code: boolean sorted = true; for (int i = 1; i maxDoc; ++i) { if (comparator.compare(i-1, i) 0) { sorted = false; break; } } if (sorted) { return null; } Perhaps you can write similar code? Also note that the sorting interface has changed, I think in 4.8, and now you don't really need to implement a Sorter, but rather pass a SortField, if that works for you. Shai On Tue, Jun 17, 2014 at 9:41 AM, Ravikumar Govindarajan ravikumar.govindara...@gmail.com wrote: Shai, This is the code snippet I use inside my class... public class MySorter extends Sorter { @Override public DocMap sort(AtomicReader reader) throws IOException { final MapInteger, BytesRef docVsId = loadSortTerm(reader); final Sorter.DocComparator comparator = new Sorter.DocComparator() { @Override public int compare(int docID1, int docID2) { BytesRef v1 = docVsId.get(docID1); BytesRef v2 = docVsId.get(docID2); return v1.compareTo(v2); } }; return sort(reader.maxDoc(), comparator); } } My Problem is, the AtomicReader passed to Sorter.sort method is actually a SlowCompositeReader, composed of a list of AtomicReaders each of which is already sorted. I find this loadSortTerm(compositeReader) to be a bit heavy where it tries to all load the doc-to-term mappings eagerly... Are there some alternatives for this? -- Ravi On Tue, Jun 17, 2014 at 10:58 AM, Shai Erera ser...@gmail.com wrote: I'm not sure that I follow ... where do you see DocMap being loaded up front? Specifically, Sorter.sort may return null of the readers are already sorted ... I think we already optimized for the case where the readers are sorted. Shai On Tue, Jun 17, 2014 at 4:04 AM, Ravikumar Govindarajan ravikumar.govindara...@gmail.com wrote: I am planning to use SortingMergePolicy where all the merge-participating segments are already sorted... I understand that I need to define a DocMap with old-new doc-id mappings. Is it possible to optimize the eager loading of DocMap and make it kind of lazy load on-demand? Ex: Pass ListAtomicReader to the caller and ask for next new-old doc mapping.. Since my segments are already sorted, I could save on memory a little-bit this way, instead of loading the full DocMap upfront -- Ravi
Re: SortingMergePolicy for already sorted segments
Therefore the DocMap is initialized only when the merge actually executes ... what is there more to postpone? Agreed. However, what I am asking is, if there is an alternative to DocMap, will that be better? Plz read-on And besides, if the segments are already sorted, you should return a null DocMap, like Lucene code does ... What I am trying to say is, my individual segments are sorted. However, when a merge combines N individual sorted-segments, there needs to be a global sort-order for writing the new segment. Passing null DocMap won't work here, no? DocMap is one-way of bringing the global order during a merge. Another way is to use something like a MergedIteratorSegmentReader instead of DocMap, which doesn't need any memory I was trying to get a heads-up on these 2 approaches. Please do let me know if I have understood correctly -- Ravi On Tue, Jun 17, 2014 at 5:42 PM, Shai Erera ser...@gmail.com wrote: I am afraid the DocMap still maintains doc-id mappings till merge and I am trying to avoid it... What do you mean 'till merge'? The method OneMerge.getMergeReaders() is called only when the merge is executed, not when the MergePolicy decided to merge those segments. Therefore the DocMap is initialized only when the merge actually executes ... what is there more to postpone? And besides, if the segments are already sorted, you should return a null DocMap, like Lucene code does ... If I miss your point, I'd appreciate if you can point me to a code example, preferably in Lucene source, which demonstrates the problem. Shai On Tue, Jun 17, 2014 at 3:03 PM, Ravikumar Govindarajan ravikumar.govindara...@gmail.com wrote: I am afraid the DocMap still maintains doc-id mappings till merge and I am trying to avoid it... I think lucene itself has a MergeIterator in o.a.l.util package. A MergePolicy can wrap a simple MergeIterator for iterating docs across different AtomicReaders in correct sort-order for a given field/term That should be fine right? -- Ravi -- Ravi On Tue, Jun 17, 2014 at 1:24 PM, Shai Erera ser...@gmail.com wrote: loadSortTerm is your method right? In the current Sorter.sort implementation, I see this code: boolean sorted = true; for (int i = 1; i maxDoc; ++i) { if (comparator.compare(i-1, i) 0) { sorted = false; break; } } if (sorted) { return null; } Perhaps you can write similar code? Also note that the sorting interface has changed, I think in 4.8, and now you don't really need to implement a Sorter, but rather pass a SortField, if that works for you. Shai On Tue, Jun 17, 2014 at 9:41 AM, Ravikumar Govindarajan ravikumar.govindara...@gmail.com wrote: Shai, This is the code snippet I use inside my class... public class MySorter extends Sorter { @Override public DocMap sort(AtomicReader reader) throws IOException { final MapInteger, BytesRef docVsId = loadSortTerm(reader); final Sorter.DocComparator comparator = new Sorter.DocComparator() { @Override public int compare(int docID1, int docID2) { BytesRef v1 = docVsId.get(docID1); BytesRef v2 = docVsId.get(docID2); return v1.compareTo(v2); } }; return sort(reader.maxDoc(), comparator); } } My Problem is, the AtomicReader passed to Sorter.sort method is actually a SlowCompositeReader, composed of a list of AtomicReaders each of which is already sorted. I find this loadSortTerm(compositeReader) to be a bit heavy where it tries to all load the doc-to-term mappings eagerly... Are there some alternatives for this? -- Ravi On Tue, Jun 17, 2014 at 10:58 AM, Shai Erera ser...@gmail.com wrote: I'm not sure that I follow ... where do you see DocMap being loaded up front? Specifically, Sorter.sort may return null of the readers are already sorted ... I think we already optimized for the case where the readers are sorted. Shai On Tue, Jun 17, 2014 at 4:04 AM, Ravikumar Govindarajan ravikumar.govindara...@gmail.com wrote: I am planning to use SortingMergePolicy where all the merge-participating segments are already sorted... I understand that I need to define a DocMap with old-new doc-id mappings. Is it possible to optimize the eager loading of DocMap and make it kind of lazy load on-demand? Ex: Pass ListAtomicReader to the caller and ask for next new-old doc mapping.. Since my segments are already sorted, I could save on memory a little-bit this way, instead of loading the full DocMap upfront -- Ravi
Re: Facets in Lucene 4.7.2
Hi, Thanks again! This time, I have indexed data with the following specs. I run into 40 seconds for the FastTaxonomyFacetCounts to create all the facets. Is this as per your measurements? Subsequent runs fare much better probably because of the Windows file system cache. How can I speed this up? I believe there was a CategoryListCache earlier. Is there any cache or other implementation that I can use? Secondly, I had a general question. If I extrapolate these numbers for a billion documents, my search and facet number may probably be unusable in a real time scenario. What are the strategies employed when you deal with such large scale? I am new to Lucene so please also direct me to the relevant info sources. Thanks! Corpus: Count: 20M, Size: 51GB Index: Size (w/o Facets): 19GB, Size (w/Facets): 20.12GB Creation Time (w/o Facets): 3.46hrs, Creation Time (w/Facets): 3.49hrs Search Performance: With 29055 hits (5 terms in query): Query Execution: 8 seconds Facet counts execution: 40-45 seconds With 4.22M hits (2 terms in query): Query Execution: 3 seconds Facet counts execution: 42-46 seconds With 15.1M hits (1 term in query): Query Execution: 2 seconds Facet counts execution: 45-53 seconds With 6183 hits (5 different values for the same 5 terms): (Without Flushing Windows File Cache on Next run) Query Execution: 11 seconds Facet counts execution: 1 second With 4.9M hits (1 different value for the 1 term): (Without Flushing Windows File Cache on Next run) Query Execution: 2 seconds Facet counts execution: 3 seconds --- Thanks n Regards, Sandeep Ramesh Khanzode On Monday, June 16, 2014 8:11 PM, Shai Erera ser...@gmail.com wrote: Hi 1.] Is there any API that gives me the count of a specific dimension from FacetCollector in response to a search query. Currently, I use the getTopChildren() with some value and then check the FacetResult object for the actual number of dimensions hit along with their occurrences. Also, the getSpecificValue() does not work without a path attribute to the API. To get the value of the dimension itself, you should call getTopChildren(1, dim). Note that getSpecificValue does not allow to pass only the dimension, and getTopChildren requires topN to be 0. Passing 1 is a hack, but I'm not sure we should specifically support getting the aggregated value of just the dimension ... once you get that, the FacetResult.value tells you the aggregated count. 2.] Can I find the MAX or MIN value of a Numeric type field written to the index? Depends how you index them. If you index the field as a numeric field (e.g. LongField), I believe you can use NumericUtils.getMaxLong. If it's a DocValues field, I don't know of a built-in function that does it, but this thread has a demo code: http://www.gossamer-threads.com/lists/lucene/java-user/195594. 3.] I am trying to compare and contrast Lucene Facets with Elastic Search. I could determine that ES does search time faceting and dynamically returns the response without any prior faceting during indexing time. Is index time lag is not my concern, can I assume that, in general, performance-wise Lucene facets would be faster? I will start by saying that I don't know much about how ES facets work. We have some committers who know both how Lucene and ES facets work, so they can comment on that. But I personally don't think there's no index-time decision when it comes to faceting. Well .. not unless you're faceting on arbitrary terms. Otherwise, you already make decision such as indexing the field as not tokenized/analyzed/lowercased/doc-values etc. Note that Lucene facets also support non-taxonomy based faceting option, using the DocValues fields. Look at SortedSetDocValuesFacetField. This too can be perceived as an index-time decision though... And there are some built-in dynamic faceting capabilities too, like range facets (LongRangeFacetCounts), which can work on any NumericDocValuesField, as well as any ValueSource (such as Expressions). I cannot compare ES facets to Lucene's in terms of performance, as I haven't benchmarked them yet. 4.] I index a semi-large-ish corpus of 20M files across 50GB. If I do not use IndexWriter.commit(), I get standard files like cfe/cfs/si in the index directory. However, if I do use the commit(), then as I understand it, the state is persisted to the disk. But this time, there are additional file extensions like doc/pos/tim/tip/dvd/dvm, etc. I am not sure about this difference and its cause. The information of the doc/tim/tip etc. is buffered in memory (controlled by ramBufferSizeMB) and when they are flushed (on commit or when the RAM buffer fills up), those files materialize on disk. When you call
Facet migration 4.6.1 to 4.7.0
Hi, I'm migrating from lucene 4.6.1 to 4.8.1 and I noticed some Facet API changes happened on 4.7.0 probably mostly related to this ticket: http://issues.apache.org/jira/browse/LUCENE-5339 Here are few question about some customization/extension we did and seem not having a direct counterpart/extension point in the new API; can someone help with these questions? - we are extending FacetResultsHandler to change the order of the facet results (i.e. date facets ordered by date instead of count). How can I achieve this now? - we have usual IndexReaders opened in groups with MultiReader, than we're merging in RAM the TaxonomyReaders to obtain a correspondence of the MultiReader for the taxonomies. Do you think I can still do this? - at some point you removed the residue information from facets and we calculated it differently; am I right I can now calculate it as FacetResult.childCount - FacetResult.labelValues.length? - we are extending TaxonomyFacetsAccumulator to provide: - specific FacetResultsHandler(s) depeding on the facet - add facet other than the topk if the user selected some facet values from the residue. where does the API permit to extends the behavior to achieve this? Any help will be really apreciated, Nicola. -- Nicola Buso Software Engineer - Web Production Team European Bioinformatics Institute (EMBL-EBI) European Molecular Biology Laboratory Wellcome Trust Genome Campus Hinxton Cambridge CB10 1SD United Kingdom URL: http://www.ebi.ac.uk - To unsubscribe, e-mail: java-user-unsubscr...@lucene.apache.org For additional commands, e-mail: java-user-h...@lucene.apache.org
Indexing size increase 20% after switching from lucene 4.4 to 4.5 or 4.8 with BinaryDocValuesField
I used lucene 4.4 to create index for some documents. One of the indexing fields is BinaryDocValuesField. After I change the dependency to lucene 4.5. The index size for 1 million documents increases from 293MB to 357MB. If I did not use BinaryDocValuesField, the index size increases only about 2%. I also tried lucene 4.8. The index size is similar to index size with lucene 4.5. I am wondering what the change for handling BinaryDocValuesField from 4.4 to 4.5 or 4.8 is. Gang Zhao Software Engineer - EA Digital Platform 207 Redwood Shores Parkway Redwood City, CA 94065 Direct Line: 650-628-3719 [cid:image001.png@01CD68F0.6239B040]
Re: Indexing size increase 20% after switching from lucene 4.4 to 4.5 or 4.8 with BinaryDocValuesField
Again, because merging is based on byte size, you have to be careful how you measure (hint: use LogDocMergePolicy). Otherwise you are comparing apples and oranges. Separately, your configuration is using experimental codecs like disk/memory which arent as heavily benchmarked etc as the default index format. On Fri, Jun 13, 2014 at 8:09 PM, Zhao, Gang gz...@ea.com wrote: I used lucene 4.4 to create index for some documents. One of the indexing fields is BinaryDocValuesField. After I change the dependency to lucene 4.5. The index size for 1 million documents increases from 293MB to 357MB. If I did not use BinaryDocValuesField, the index size increases only about 2%. I also tried lucene 4.8. The index size is similar to index size with lucene 4.5. I am wondering what the change for handling BinaryDocValuesField from 4.4 to 4.5 or 4.8 is. Gang Zhao Software Engineer - EA Digital Platform 207 Redwood Shores Parkway Redwood City, CA 94065 Direct Line: 650-628-3719 [image: cid:image001.png@01CD68F0.6239B040]
Re: Facets in Lucene 4.7.2
Hi 40 seconds for faceted search is ... crazy. Also, note how the times don't differ much even though the number of hits is much higher (29K vs 15.1M) ... That, w/ that you say that subsequent queries are much faster (few seconds) suggests that something is seriously messed up w/ your environment. Maybe it's a faulty disk? E.g. after the file system cache is warm, you no longer hit the disk? In general, the more hits you have, the more expensive is faceted search. It's also true for scoring as well (i.e. even without facets). There's just more work to determine the top results (docs, facets...). With facets, you can use sampling (see RandomSamplingFacetsCollector), but I would do that only after you verify that collecting 15M docs is very expensive for you, even when the file system cache is hot. I've never seen those numbers before, therefore it's difficult for me to relate to them. There's a caching mechanism for facets, through CachedOrdinalsReader. But I wouldn't go there until you verify that your IO system is good (try another machine, OS, disk ...)., and that the 40s times are truly from the faceting code. Shai On Tue, Jun 17, 2014 at 4:21 PM, Sandeep Khanzode sandeep_khanz...@yahoo.com.invalid wrote: Hi, Thanks again! This time, I have indexed data with the following specs. I run into 40 seconds for the FastTaxonomyFacetCounts to create all the facets. Is this as per your measurements? Subsequent runs fare much better probably because of the Windows file system cache. How can I speed this up? I believe there was a CategoryListCache earlier. Is there any cache or other implementation that I can use? Secondly, I had a general question. If I extrapolate these numbers for a billion documents, my search and facet number may probably be unusable in a real time scenario. What are the strategies employed when you deal with such large scale? I am new to Lucene so please also direct me to the relevant info sources. Thanks! Corpus: Count: 20M, Size: 51GB Index: Size (w/o Facets): 19GB, Size (w/Facets): 20.12GB Creation Time (w/o Facets): 3.46hrs, Creation Time (w/Facets): 3.49hrs Search Performance: With 29055 hits (5 terms in query): Query Execution: 8 seconds Facet counts execution: 40-45 seconds With 4.22M hits (2 terms in query): Query Execution: 3 seconds Facet counts execution: 42-46 seconds With 15.1M hits (1 term in query): Query Execution: 2 seconds Facet counts execution: 45-53 seconds With 6183 hits (5 different values for the same 5 terms): (Without Flushing Windows File Cache on Next run) Query Execution: 11 seconds Facet counts execution: 1 second With 4.9M hits (1 different value for the 1 term): (Without Flushing Windows File Cache on Next run) Query Execution: 2 seconds Facet counts execution: 3 seconds --- Thanks n Regards, Sandeep Ramesh Khanzode On Monday, June 16, 2014 8:11 PM, Shai Erera ser...@gmail.com wrote: Hi 1.] Is there any API that gives me the count of a specific dimension from FacetCollector in response to a search query. Currently, I use the getTopChildren() with some value and then check the FacetResult object for the actual number of dimensions hit along with their occurrences. Also, the getSpecificValue() does not work without a path attribute to the API. To get the value of the dimension itself, you should call getTopChildren(1, dim). Note that getSpecificValue does not allow to pass only the dimension, and getTopChildren requires topN to be 0. Passing 1 is a hack, but I'm not sure we should specifically support getting the aggregated value of just the dimension ... once you get that, the FacetResult.value tells you the aggregated count. 2.] Can I find the MAX or MIN value of a Numeric type field written to the index? Depends how you index them. If you index the field as a numeric field (e.g. LongField), I believe you can use NumericUtils.getMaxLong. If it's a DocValues field, I don't know of a built-in function that does it, but this thread has a demo code: http://www.gossamer-threads.com/lists/lucene/java-user/195594. 3.] I am trying to compare and contrast Lucene Facets with Elastic Search. I could determine that ES does search time faceting and dynamically returns the response without any prior faceting during indexing time. Is index time lag is not my concern, can I assume that, in general, performance-wise Lucene facets would be faster? I will start by saying that I don't know much about how ES facets work. We have some committers who know both how Lucene and ES facets work, so they can comment on that. But I personally don't think there's no index-time decision when it comes to faceting. Well ..
Re: SortingMergePolicy for already sorted segments
OK I think I now understand what you're asking :). It's unrelated though to SortingMergePolicy. You propose to do the merge part of a merge-sort, since we know the indexes are already sorted, right? This is something we've considered in the past, but it is very tricky (see below) and we went with the SortingAR for simplicity and speed of coding. If however you have an idea how we can easily implement that, that would be awesome. So let's consider merging the posting lists of f:val from the N readers. Say that each returns docs 0-3, and the merged posting will have 4*N entries (say we don't have deletes). To properly merge them, you need to lookup the sort-value of each document from each reader, and compare according to it. Now you move on to f:val2 (another posting) and it wants to merge 100 other docs. So you need to lookup the value of each document, compare by it, and merge them. And the process continues ... These lookups are expensive and will be done millions of times (each term, each DV field, each .. everything). More than that, there's a serious issue of correctness, because you never make a global sorting decision. So if f:val sees only a single document - 0, in all segments, you want to map them to 4 GLOBALLY SORTED documents. If you make a local decision based on these 4 documents, you will end up w/ a completely messed up segment. I think the global DocMap is really required. Forget about that that other code, e.g. IndexWriter relies on this in order to properly apply incoming document deletions and field updates while the segments were merging. It's just a matter of correctness - we need to know the global sorted segment map. Shai On Tue, Jun 17, 2014 at 3:41 PM, Ravikumar Govindarajan ravikumar.govindara...@gmail.com wrote: Therefore the DocMap is initialized only when the merge actually executes ... what is there more to postpone? Agreed. However, what I am asking is, if there is an alternative to DocMap, will that be better? Plz read-on And besides, if the segments are already sorted, you should return a null DocMap, like Lucene code does ... What I am trying to say is, my individual segments are sorted. However, when a merge combines N individual sorted-segments, there needs to be a global sort-order for writing the new segment. Passing null DocMap won't work here, no? DocMap is one-way of bringing the global order during a merge. Another way is to use something like a MergedIteratorSegmentReader instead of DocMap, which doesn't need any memory I was trying to get a heads-up on these 2 approaches. Please do let me know if I have understood correctly -- Ravi On Tue, Jun 17, 2014 at 5:42 PM, Shai Erera ser...@gmail.com wrote: I am afraid the DocMap still maintains doc-id mappings till merge and I am trying to avoid it... What do you mean 'till merge'? The method OneMerge.getMergeReaders() is called only when the merge is executed, not when the MergePolicy decided to merge those segments. Therefore the DocMap is initialized only when the merge actually executes ... what is there more to postpone? And besides, if the segments are already sorted, you should return a null DocMap, like Lucene code does ... If I miss your point, I'd appreciate if you can point me to a code example, preferably in Lucene source, which demonstrates the problem. Shai On Tue, Jun 17, 2014 at 3:03 PM, Ravikumar Govindarajan ravikumar.govindara...@gmail.com wrote: I am afraid the DocMap still maintains doc-id mappings till merge and I am trying to avoid it... I think lucene itself has a MergeIterator in o.a.l.util package. A MergePolicy can wrap a simple MergeIterator for iterating docs across different AtomicReaders in correct sort-order for a given field/term That should be fine right? -- Ravi -- Ravi On Tue, Jun 17, 2014 at 1:24 PM, Shai Erera ser...@gmail.com wrote: loadSortTerm is your method right? In the current Sorter.sort implementation, I see this code: boolean sorted = true; for (int i = 1; i maxDoc; ++i) { if (comparator.compare(i-1, i) 0) { sorted = false; break; } } if (sorted) { return null; } Perhaps you can write similar code? Also note that the sorting interface has changed, I think in 4.8, and now you don't really need to implement a Sorter, but rather pass a SortField, if that works for you. Shai On Tue, Jun 17, 2014 at 9:41 AM, Ravikumar Govindarajan ravikumar.govindara...@gmail.com wrote: Shai, This is the code snippet I use inside my class... public class MySorter extends Sorter { @Override public DocMap sort(AtomicReader reader) throws IOException { final MapInteger, BytesRef docVsId = loadSortTerm(reader);
Re: SortingMergePolicy for already sorted segments
That said... if we generate the global DocMap up front, there's no reason to not execute the merge of the segments more efficiently, i.e. without wrapping them in a SlowCompositeReaderWrapper. But that's not work for SortingMergePolicy, it's either a special SortingAtomicReader which wraps a group of readers + a global DocMap, and then merge-sorts them more efficiently than how it's done now. Or we tap into SegmentMerger .. which is way more complicated. Perhaps it would be worth to explore a SortingMultiSortedAtomicReader which merge-sorts the postings and other data that way ... I look at e.g how doc-values are merged .. not sure it will improve performance. But if you want to cons up a patch, that'd be awesome! Shai On Tue, Jun 17, 2014 at 8:01 PM, Shai Erera ser...@gmail.com wrote: OK I think I now understand what you're asking :). It's unrelated though to SortingMergePolicy. You propose to do the merge part of a merge-sort, since we know the indexes are already sorted, right? This is something we've considered in the past, but it is very tricky (see below) and we went with the SortingAR for simplicity and speed of coding. If however you have an idea how we can easily implement that, that would be awesome. So let's consider merging the posting lists of f:val from the N readers. Say that each returns docs 0-3, and the merged posting will have 4*N entries (say we don't have deletes). To properly merge them, you need to lookup the sort-value of each document from each reader, and compare according to it. Now you move on to f:val2 (another posting) and it wants to merge 100 other docs. So you need to lookup the value of each document, compare by it, and merge them. And the process continues ... These lookups are expensive and will be done millions of times (each term, each DV field, each .. everything). More than that, there's a serious issue of correctness, because you never make a global sorting decision. So if f:val sees only a single document - 0, in all segments, you want to map them to 4 GLOBALLY SORTED documents. If you make a local decision based on these 4 documents, you will end up w/ a completely messed up segment. I think the global DocMap is really required. Forget about that that other code, e.g. IndexWriter relies on this in order to properly apply incoming document deletions and field updates while the segments were merging. It's just a matter of correctness - we need to know the global sorted segment map. Shai On Tue, Jun 17, 2014 at 3:41 PM, Ravikumar Govindarajan ravikumar.govindara...@gmail.com wrote: Therefore the DocMap is initialized only when the merge actually executes ... what is there more to postpone? Agreed. However, what I am asking is, if there is an alternative to DocMap, will that be better? Plz read-on And besides, if the segments are already sorted, you should return a null DocMap, like Lucene code does ... What I am trying to say is, my individual segments are sorted. However, when a merge combines N individual sorted-segments, there needs to be a global sort-order for writing the new segment. Passing null DocMap won't work here, no? DocMap is one-way of bringing the global order during a merge. Another way is to use something like a MergedIteratorSegmentReader instead of DocMap, which doesn't need any memory I was trying to get a heads-up on these 2 approaches. Please do let me know if I have understood correctly -- Ravi On Tue, Jun 17, 2014 at 5:42 PM, Shai Erera ser...@gmail.com wrote: I am afraid the DocMap still maintains doc-id mappings till merge and I am trying to avoid it... What do you mean 'till merge'? The method OneMerge.getMergeReaders() is called only when the merge is executed, not when the MergePolicy decided to merge those segments. Therefore the DocMap is initialized only when the merge actually executes ... what is there more to postpone? And besides, if the segments are already sorted, you should return a null DocMap, like Lucene code does ... If I miss your point, I'd appreciate if you can point me to a code example, preferably in Lucene source, which demonstrates the problem. Shai On Tue, Jun 17, 2014 at 3:03 PM, Ravikumar Govindarajan ravikumar.govindara...@gmail.com wrote: I am afraid the DocMap still maintains doc-id mappings till merge and I am trying to avoid it... I think lucene itself has a MergeIterator in o.a.l.util package. A MergePolicy can wrap a simple MergeIterator for iterating docs across different AtomicReaders in correct sort-order for a given field/term That should be fine right? -- Ravi -- Ravi On Tue, Jun 17, 2014 at 1:24 PM, Shai Erera ser...@gmail.com wrote: loadSortTerm is your method right? In the current Sorter.sort implementation, I see this code: boolean sorted = true; for (int i = 1; i
Re: Facets in Lucene 4.7.2
Hi, Thanks for your response. It does sound pretty bad which is why I am not sure whether there is an issue with the code, the index, the searcher, or just the machine, as you say. I will try with another machine just to make sure and post the results. Meanwhile, can you tell me if there is anything wrong in the below measurement? Or is the API usage or the pattern incorrect? I used a tool called RAMMap to clean the Windows cache. If I do not, the results are very fast as I mentioned already. If I do, then the total time is 40s. Can you please provide any pointers on what could be wrong? I will be checking on a Linux box anyway. = System.out.println(1. Start Date: + new Date()); TopDocs topDocs = FacetsCollector.search(searcher, query, 100, fc); System.out.println(1. End Date: + new Date()); // Above part takes approx 2-12 seconds depending on the query System.out.println(2. Start Date: + new Date()); ListFacetResult results = new ArrayListFacetResult(); Facets facets = new FastTaxonomyFacetCounts(taxoReader, config, fc); System.out.println(2. End Date: + new Date()); // Above part takes approx 40-53 seconds depending on the query for the first time on Windows System.out.println(3. Start Date: + new Date()); results.add(facets.getTopChildren(1000, F1)); results.add(facets.getTopChildren(1000, F2)); results.add(facets.getTopChildren(1000, F3)); results.add(facets.getTopChildren(1000, F4)); results.add(facets.getTopChildren(1000, F5)); results.add(facets.getTopChildren(1000, F6)); results.add(facets.getTopChildren(1000, F7)); System.out.println(3. End Date: + new Date()); // Above part takes approx less than 1 second = --- Thanks n Regards, Sandeep Ramesh Khanzode On Tuesday, June 17, 2014 10:15 PM, Shai Erera ser...@gmail.com wrote: Hi 40 seconds for faceted search is ... crazy. Also, note how the times don't differ much even though the number of hits is much higher (29K vs 15.1M) ... That, w/ that you say that subsequent queries are much faster (few seconds) suggests that something is seriously messed up w/ your environment. Maybe it's a faulty disk? E.g. after the file system cache is warm, you no longer hit the disk? In general, the more hits you have, the more expensive is faceted search. It's also true for scoring as well (i.e. even without facets). There's just more work to determine the top results (docs, facets...). With facets, you can use sampling (see RandomSamplingFacetsCollector), but I would do that only after you verify that collecting 15M docs is very expensive for you, even when the file system cache is hot. I've never seen those numbers before, therefore it's difficult for me to relate to them. There's a caching mechanism for facets, through CachedOrdinalsReader. But I wouldn't go there until you verify that your IO system is good (try another machine, OS, disk ...)., and that the 40s times are truly from the faceting code. Shai On Tue, Jun 17, 2014 at 4:21 PM, Sandeep Khanzode sandeep_khanz...@yahoo.com.invalid wrote: Hi, Thanks again! This time, I have indexed data with the following specs. I run into 40 seconds for the FastTaxonomyFacetCounts to create all the facets. Is this as per your measurements? Subsequent runs fare much better probably because of the Windows file system cache. How can I speed this up? I believe there was a CategoryListCache earlier. Is there any cache or other implementation that I can use? Secondly, I had a general question. If I extrapolate these numbers for a billion documents, my search and facet number may probably be unusable in a real time scenario. What are the strategies employed when you deal with such large scale? I am new to Lucene so please also direct me to the relevant info sources. Thanks! Corpus: Count: 20M, Size: 51GB Index: Size (w/o Facets): 19GB, Size (w/Facets): 20.12GB Creation Time (w/o Facets): 3.46hrs, Creation Time (w/Facets): 3.49hrs Search Performance: With 29055 hits (5 terms in query): Query Execution: 8 seconds Facet counts execution: 40-45 seconds With 4.22M hits (2 terms in query): Query Execution: 3 seconds Facet counts execution: 42-46 seconds With 15.1M hits (1 term in query): Query Execution: 2 seconds Facet counts execution: 45-53 seconds With 6183 hits (5 different values for the same 5 terms): (Without Flushing Windows File Cache on Next run) Query Execution: 11 seconds Facet counts execution: 1 second With 4.9M hits (1 different value for the 1 term): (Without Flushing Windows File Cache on Next run) Query Execution: 2 seconds Facet counts execution: 3 seconds
Re: Facets in Lucene 4.7.2
Nothing suspicious ... code looks fine. The call to FastTaxoFacetCounts actually computes the counts ... that's the expensive part of faceted search. How big is your taxonomy (number categories)? Is it hierarchical (i.e. are your dimensions flat, or deep like A/1/2/3/)? What does your FacetsConfig look like? Still, well maybe if your taxonomy is huge (hundreds of millions of categories), I don't think you can intentionally mess up something that much to end up w/ 40-45s response times! Shai On Tue, Jun 17, 2014 at 8:51 PM, Sandeep Khanzode sandeep_khanz...@yahoo.com.invalid wrote: Hi, Thanks for your response. It does sound pretty bad which is why I am not sure whether there is an issue with the code, the index, the searcher, or just the machine, as you say. I will try with another machine just to make sure and post the results. Meanwhile, can you tell me if there is anything wrong in the below measurement? Or is the API usage or the pattern incorrect? I used a tool called RAMMap to clean the Windows cache. If I do not, the results are very fast as I mentioned already. If I do, then the total time is 40s. Can you please provide any pointers on what could be wrong? I will be checking on a Linux box anyway. = System.out.println(1. Start Date: + new Date()); TopDocs topDocs = FacetsCollector.search(searcher, query, 100, fc); System.out.println(1. End Date: + new Date()); // Above part takes approx 2-12 seconds depending on the query System.out.println(2. Start Date: + new Date()); ListFacetResult results = new ArrayListFacetResult(); Facets facets = new FastTaxonomyFacetCounts(taxoReader, config, fc); System.out.println(2. End Date: + new Date()); // Above part takes approx 40-53 seconds depending on the query for the first time on Windows System.out.println(3. Start Date: + new Date()); results.add(facets.getTopChildren(1000, F1)); results.add(facets.getTopChildren(1000, F2)); results.add(facets.getTopChildren(1000, F3)); results.add(facets.getTopChildren(1000, F4)); results.add(facets.getTopChildren(1000, F5)); results.add(facets.getTopChildren(1000, F6)); results.add(facets.getTopChildren(1000, F7)); System.out.println(3. End Date: + new Date()); // Above part takes approx less than 1 second = --- Thanks n Regards, Sandeep Ramesh Khanzode On Tuesday, June 17, 2014 10:15 PM, Shai Erera ser...@gmail.com wrote: Hi 40 seconds for faceted search is ... crazy. Also, note how the times don't differ much even though the number of hits is much higher (29K vs 15.1M) ... That, w/ that you say that subsequent queries are much faster (few seconds) suggests that something is seriously messed up w/ your environment. Maybe it's a faulty disk? E.g. after the file system cache is warm, you no longer hit the disk? In general, the more hits you have, the more expensive is faceted search. It's also true for scoring as well (i.e. even without facets). There's just more work to determine the top results (docs, facets...). With facets, you can use sampling (see RandomSamplingFacetsCollector), but I would do that only after you verify that collecting 15M docs is very expensive for you, even when the file system cache is hot. I've never seen those numbers before, therefore it's difficult for me to relate to them. There's a caching mechanism for facets, through CachedOrdinalsReader. But I wouldn't go there until you verify that your IO system is good (try another machine, OS, disk ...)., and that the 40s times are truly from the faceting code. Shai On Tue, Jun 17, 2014 at 4:21 PM, Sandeep Khanzode sandeep_khanz...@yahoo.com.invalid wrote: Hi, Thanks again! This time, I have indexed data with the following specs. I run into 40 seconds for the FastTaxonomyFacetCounts to create all the facets. Is this as per your measurements? Subsequent runs fare much better probably because of the Windows file system cache. How can I speed this up? I believe there was a CategoryListCache earlier. Is there any cache or other implementation that I can use? Secondly, I had a general question. If I extrapolate these numbers for a billion documents, my search and facet number may probably be unusable in a real time scenario. What are the strategies employed when you deal with such large scale? I am new to Lucene so please also direct me to the relevant info sources. Thanks! Corpus: Count: 20M, Size: 51GB Index: Size (w/o Facets): 19GB, Size (w/Facets): 20.12GB Creation Time (w/o Facets): 3.46hrs, Creation Time (w/Facets): 3.49hrs Search Performance: With 29055 hits (5 terms in query): Query Execution: 8 seconds Facet counts execution: 40-45 seconds With 4.22M hits (2 terms in query):
Lucene QueryParser/Analyzer inconsistency
Hi, I'm experience a puzzling behaviour with the QueryParser and was hoping someone around here can help me. I have a very simple Analyzer that tries to replace forward slashes (/) by spaces. Because QueryParser forces me to escape strings with slashes before parsing, I added a MappingCharFilter to the analyzer that replaces \/ with a single space. The analyzer is defined as follows: @Override protected TokenStreamComponents createComponents(String field, Reader in) { NormalizeCharMap.Builder builder = new NormalizeCharMap.Builder(); builder.add(\\/, ); Reader mappingFilter = new MappingCharFilter(builder.build(), in); Tokenizer tokenizer = new WhitespaceTokenizer(version, mappingFilter); return new TokenStreamComponents(tokenizer); } Then I use this analyzer in the QueryParser to parse a string with dashes: String text = QueryParser.escape(one/two); QueryParser parser = new QueryParser(Version.LUCENE_48, f, new MyAnalyzer(Version.LUCENE_48)); System.err.println(parser.parse(text)); The expected output would be f:one f:two However, I get: f:one/two The puzzling thing is that when I debug the analyzer, it tokenizes the input string correctly, returning two tokens instead of one. What is going on? Many thanks, Luís Pureza P.S.: I was able to fix this issue temporarily by creating my own tokenizer that tokenizes on whitespace and slashes. However, I still don't understand what's going on.
Re: Facets in Lucene 4.7.2
If I am counting correctly, the $facets field in the index shows a count of approx. 28k. That does not sound like much, I guess. All my facets are flat and the FacetsConfig only defines a couple of them to be multi-valued. Let me know if I am not counting the taxonomy size correctly. The taxoReader.getSize() also shows this count. I will check on a Linux box to make sure. Thanks, --- Thanks n Regards, Sandeep Ramesh Khanzode On Tuesday, June 17, 2014 11:28 PM, Shai Erera ser...@gmail.com wrote: Nothing suspicious ... code looks fine. The call to FastTaxoFacetCounts actually computes the counts ... that's the expensive part of faceted search. How big is your taxonomy (number categories)? Is it hierarchical (i.e. are your dimensions flat, or deep like A/1/2/3/)? What does your FacetsConfig look like? Still, well maybe if your taxonomy is huge (hundreds of millions of categories), I don't think you can intentionally mess up something that much to end up w/ 40-45s response times! Shai On Tue, Jun 17, 2014 at 8:51 PM, Sandeep Khanzode sandeep_khanz...@yahoo.com.invalid wrote: Hi, Thanks for your response. It does sound pretty bad which is why I am not sure whether there is an issue with the code, the index, the searcher, or just the machine, as you say. I will try with another machine just to make sure and post the results. Meanwhile, can you tell me if there is anything wrong in the below measurement? Or is the API usage or the pattern incorrect? I used a tool called RAMMap to clean the Windows cache. If I do not, the results are very fast as I mentioned already. If I do, then the total time is 40s. Can you please provide any pointers on what could be wrong? I will be checking on a Linux box anyway. = System.out.println(1. Start Date: + new Date()); TopDocs topDocs = FacetsCollector.search(searcher, query, 100, fc); System.out.println(1. End Date: + new Date()); // Above part takes approx 2-12 seconds depending on the query System.out.println(2. Start Date: + new Date()); ListFacetResult results = new ArrayListFacetResult(); Facets facets = new FastTaxonomyFacetCounts(taxoReader, config, fc); System.out.println(2. End Date: + new Date()); // Above part takes approx 40-53 seconds depending on the query for the first time on Windows System.out.println(3. Start Date: + new Date()); results.add(facets.getTopChildren(1000, F1)); results.add(facets.getTopChildren(1000, F2)); results.add(facets.getTopChildren(1000, F3)); results.add(facets.getTopChildren(1000, F4)); results.add(facets.getTopChildren(1000, F5)); results.add(facets.getTopChildren(1000, F6)); results.add(facets.getTopChildren(1000, F7)); System.out.println(3. End Date: + new Date()); // Above part takes approx less than 1 second = --- Thanks n Regards, Sandeep Ramesh Khanzode On Tuesday, June 17, 2014 10:15 PM, Shai Erera ser...@gmail.com wrote: Hi 40 seconds for faceted search is ... crazy. Also, note how the times don't differ much even though the number of hits is much higher (29K vs 15.1M) ... That, w/ that you say that subsequent queries are much faster (few seconds) suggests that something is seriously messed up w/ your environment. Maybe it's a faulty disk? E.g. after the file system cache is warm, you no longer hit the disk? In general, the more hits you have, the more expensive is faceted search. It's also true for scoring as well (i.e. even without facets). There's just more work to determine the top results (docs, facets...). With facets, you can use sampling (see RandomSamplingFacetsCollector), but I would do that only after you verify that collecting 15M docs is very expensive for you, even when the file system cache is hot. I've never seen those numbers before, therefore it's difficult for me to relate to them. There's a caching mechanism for facets, through CachedOrdinalsReader. But I wouldn't go there until you verify that your IO system is good (try another machine, OS, disk ...)., and that the 40s times are truly from the faceting code. Shai On Tue, Jun 17, 2014 at 4:21 PM, Sandeep Khanzode sandeep_khanz...@yahoo.com.invalid wrote: Hi, Thanks again! This time, I have indexed data with the following specs. I run into 40 seconds for the FastTaxonomyFacetCounts to create all the facets. Is this as per your measurements? Subsequent runs fare much better probably because of the Windows file system cache. How can I speed this up? I believe there was a CategoryListCache earlier. Is there any cache or other implementation that I can use? Secondly, I had a general question. If I extrapolate these numbers for a billion documents, my search and facet number may probably be unusable in a real time scenario. What are the strategies employed when you deal
Re: Facets in Lucene 4.7.2
You can get the size of the taxonomy by calling taxoReader.getSize(). What does the 28K of the $facets field denote - the number of terms (drill-down)? If so, that sounds like your taxonomy is of that size. And indeed, this is a tiny taxonomy ... How many facets do you record per document? This also affects the amount of IO that's done during search, as we traverse the BinaryDocValues field, reading the categories of each document. Shai On Tue, Jun 17, 2014 at 9:32 PM, Sandeep Khanzode sandeep_khanz...@yahoo.com.invalid wrote: If I am counting correctly, the $facets field in the index shows a count of approx. 28k. That does not sound like much, I guess. All my facets are flat and the FacetsConfig only defines a couple of them to be multi-valued. Let me know if I am not counting the taxonomy size correctly. The taxoReader.getSize() also shows this count. I will check on a Linux box to make sure. Thanks, --- Thanks n Regards, Sandeep Ramesh Khanzode On Tuesday, June 17, 2014 11:28 PM, Shai Erera ser...@gmail.com wrote: Nothing suspicious ... code looks fine. The call to FastTaxoFacetCounts actually computes the counts ... that's the expensive part of faceted search. How big is your taxonomy (number categories)? Is it hierarchical (i.e. are your dimensions flat, or deep like A/1/2/3/)? What does your FacetsConfig look like? Still, well maybe if your taxonomy is huge (hundreds of millions of categories), I don't think you can intentionally mess up something that much to end up w/ 40-45s response times! Shai On Tue, Jun 17, 2014 at 8:51 PM, Sandeep Khanzode sandeep_khanz...@yahoo.com.invalid wrote: Hi, Thanks for your response. It does sound pretty bad which is why I am not sure whether there is an issue with the code, the index, the searcher, or just the machine, as you say. I will try with another machine just to make sure and post the results. Meanwhile, can you tell me if there is anything wrong in the below measurement? Or is the API usage or the pattern incorrect? I used a tool called RAMMap to clean the Windows cache. If I do not, the results are very fast as I mentioned already. If I do, then the total time is 40s. Can you please provide any pointers on what could be wrong? I will be checking on a Linux box anyway. = System.out.println(1. Start Date: + new Date()); TopDocs topDocs = FacetsCollector.search(searcher, query, 100, fc); System.out.println(1. End Date: + new Date()); // Above part takes approx 2-12 seconds depending on the query System.out.println(2. Start Date: + new Date()); ListFacetResult results = new ArrayListFacetResult(); Facets facets = new FastTaxonomyFacetCounts(taxoReader, config, fc); System.out.println(2. End Date: + new Date()); // Above part takes approx 40-53 seconds depending on the query for the first time on Windows System.out.println(3. Start Date: + new Date()); results.add(facets.getTopChildren(1000, F1)); results.add(facets.getTopChildren(1000, F2)); results.add(facets.getTopChildren(1000, F3)); results.add(facets.getTopChildren(1000, F4)); results.add(facets.getTopChildren(1000, F5)); results.add(facets.getTopChildren(1000, F6)); results.add(facets.getTopChildren(1000, F7)); System.out.println(3. End Date: + new Date()); // Above part takes approx less than 1 second = --- Thanks n Regards, Sandeep Ramesh Khanzode On Tuesday, June 17, 2014 10:15 PM, Shai Erera ser...@gmail.com wrote: Hi 40 seconds for faceted search is ... crazy. Also, note how the times don't differ much even though the number of hits is much higher (29K vs 15.1M) ... That, w/ that you say that subsequent queries are much faster (few seconds) suggests that something is seriously messed up w/ your environment. Maybe it's a faulty disk? E.g. after the file system cache is warm, you no longer hit the disk? In general, the more hits you have, the more expensive is faceted search. It's also true for scoring as well (i.e. even without facets). There's just more work to determine the top results (docs, facets...). With facets, you can use sampling (see RandomSamplingFacetsCollector), but I would do that only after you verify that collecting 15M docs is very expensive for you, even when the file system cache is hot. I've never seen those numbers before, therefore it's difficult for me to relate to them. There's a caching mechanism for facets, through CachedOrdinalsReader. But I wouldn't go there until you verify that your IO system is good (try another machine, OS, disk ...)., and that the 40s times are truly from the faceting code. Shai On Tue, Jun 17, 2014 at 4:21 PM, Sandeep Khanzode sandeep_khanz...@yahoo.com.invalid wrote: Hi,
Re: Lucene QueryParser/Analyzer inconsistency
Yeah, this is kind of tricky and confusing! Here's what happens: 1. The query parser parses the input string into individual source terms, each delimited by white space. The escape is removed in this process, but... no analyzer has been called at this stage. 2. The query parser (generator) calls the analyzer for each source term. Your analyzer is called at this stage, but... the escape is already gone, so... the backslashslash mapping rule is not triggered, leaving the slash recorded in the source term from step 1. You do need the backslash in your original query because a slash introduces a regex query term. It is added by the escape method you call, but the escaping will be gone by the time your analyzer is called. So, just try a simple, unescaped slash in your char mapping table. -- Jack Krupansky -Original Message- From: Luis Pureza Sent: Tuesday, June 17, 2014 1:43 PM To: java-user@lucene.apache.org Subject: Lucene QueryParser/Analyzer inconsistency Hi, I'm experience a puzzling behaviour with the QueryParser and was hoping someone around here can help me. I have a very simple Analyzer that tries to replace forward slashes (/) by spaces. Because QueryParser forces me to escape strings with slashes before parsing, I added a MappingCharFilter to the analyzer that replaces \/ with a single space. The analyzer is defined as follows: @Override protected TokenStreamComponents createComponents(String field, Reader in) { NormalizeCharMap.Builder builder = new NormalizeCharMap.Builder(); builder.add(\\/, ); Reader mappingFilter = new MappingCharFilter(builder.build(), in); Tokenizer tokenizer = new WhitespaceTokenizer(version, mappingFilter); return new TokenStreamComponents(tokenizer); } Then I use this analyzer in the QueryParser to parse a string with dashes: String text = QueryParser.escape(one/two); QueryParser parser = new QueryParser(Version.LUCENE_48, f, new MyAnalyzer(Version.LUCENE_48)); System.err.println(parser.parse(text)); The expected output would be f:one f:two However, I get: f:one/two The puzzling thing is that when I debug the analyzer, it tokenizes the input string correctly, returning two tokens instead of one. What is going on? Many thanks, Luís Pureza P.S.: I was able to fix this issue temporarily by creating my own tokenizer that tokenizes on whitespace and slashes. However, I still don't understand what's going on. - To unsubscribe, e-mail: java-user-unsubscr...@lucene.apache.org For additional commands, e-mail: java-user-h...@lucene.apache.org