[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17234755#comment-17234755 ] Adrien Grand commented on LUCENE-9378: -- [~gworah] I'm pretty sure that the drop is due to this change as we had seen a change in the other direction when introducing compression for binary fields. The format with compression looks generally faster for linear scans (like our faceting tasks in nightlies) and slower for selective queries (like some use-cases mentioned above). > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Fix For: 8.8 > > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 5h 40m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17233885#comment-17233885 ] Michael McCandless commented on LUCENE-9378: {quote}Meanwhile, I'll open an issue to disable compression for BinaryDocValues in faceting operations on Lucene's nightly benchmarks! {quote} +1, thanks [~gworah] and [~jpountz]! > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Fix For: 8.8 > > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 5h 40m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17233826#comment-17233826 ] Gautam Worah commented on LUCENE-9378: -- Thanks for catching that regression so quickly Adrien! I had recently made a change in [LUCENE-9450|https://github.com/apache/lucene-solr/pull/1733/] that had switched taxonomy facets to use BinaryDocValues from StoredFields and that could have caused the drop in performance. [Benchmark runs|https://issues.apache.org/jira/browse/LUCENE-9450?focusedCommentId=17180860=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#comment-17180860] for my change had shown a 4.7% improvement in {{BrowseDayOfYearTaxoFacets}} but I had also made some changes after the benchmarks were run. Meanwhile, I'll open an issue to disable compression for BinaryDocValues in faceting operations on Lucene's nightly benchmarks! > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Fix For: 8.8 > > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 5h 40m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17233445#comment-17233445 ] ASF subversion and git services commented on LUCENE-9378: - Commit 4735b4c12fef1723b41ed022ac30340aaea1d372 in lucene-solr's branch refs/heads/master from Adrien Grand [ https://gitbox.apache.org/repos/asf?p=lucene-solr.git;h=4735b4c ] LUCENE-9378: Fix test failure. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Fix For: 8.8 > > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 5h 40m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17232812#comment-17232812 ] Adrien Grand commented on LUCENE-9378: -- This introduced some slowdowns on the nightly benchmarks, e.g. http://people.apache.org/~mikemccand/lucenebench/BrowseMonthTaxoFacets.html. It would be nice if the strategy for BEST_SPEED performed better on linear scans. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Fix For: 8.8 > > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 5h 40m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17232045#comment-17232045 ] ASF subversion and git services commented on LUCENE-9378: - Commit 9b4f8235aac5021aaed348055eb7ca0f593804fd in lucene-solr's branch refs/heads/branch_8x from Adrien Grand [ https://gitbox.apache.org/repos/asf?p=lucene-solr.git;h=9b4f823 ] LUCENE-9378: Fix test failure. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Fix For: 8.8 > > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 5h 40m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17232044#comment-17232044 ] ASF subversion and git services commented on LUCENE-9378: - Commit 9b4f8235aac5021aaed348055eb7ca0f593804fd in lucene-solr's branch refs/heads/branch_8x from Adrien Grand [ https://gitbox.apache.org/repos/asf?p=lucene-solr.git;h=9b4f823 ] LUCENE-9378: Fix test failure. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Fix For: 8.8 > > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 5h 40m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17232043#comment-17232043 ] ASF subversion and git services commented on LUCENE-9378: - Commit 9b4f8235aac5021aaed348055eb7ca0f593804fd in lucene-solr's branch refs/heads/branch_8x from Adrien Grand [ https://gitbox.apache.org/repos/asf?p=lucene-solr.git;h=9b4f823 ] LUCENE-9378: Fix test failure. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Fix For: 8.8 > > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 5h 40m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17232042#comment-17232042 ] ASF subversion and git services commented on LUCENE-9378: - Commit 9b4f8235aac5021aaed348055eb7ca0f593804fd in lucene-solr's branch refs/heads/branch_8x from Adrien Grand [ https://gitbox.apache.org/repos/asf?p=lucene-solr.git;h=9b4f823 ] LUCENE-9378: Fix test failure. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Fix For: 8.8 > > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 5h 40m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17232040#comment-17232040 ] ASF subversion and git services commented on LUCENE-9378: - Commit 9b4f8235aac5021aaed348055eb7ca0f593804fd in lucene-solr's branch refs/heads/branch_8x from Adrien Grand [ https://gitbox.apache.org/repos/asf?p=lucene-solr.git;h=9b4f823 ] LUCENE-9378: Fix test failure. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Fix For: 8.8 > > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 5h 40m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17231472#comment-17231472 ] Michael McCandless commented on LUCENE-9378: Thanks [~jpountz]! > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Fix For: 8.8 > > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 5h 40m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17230911#comment-17230911 ] ASF subversion and git services commented on LUCENE-9378: - Commit a48dc123b3e65928d5e59a76735cf6c88099915a in lucene-solr's branch refs/heads/branch_8x from Adrien Grand [ https://gitbox.apache.org/repos/asf?p=lucene-solr.git;h=a48dc12 ] LUCENE-9378: Make it possible to configure how to trade speed for compression on doc values. (#2069) This adds a switch to `Lucene80DocValuesFormat` which allows to configure whether to prioritize retrieval speed over compression ratio or the other way around. When prioritizing retrieval speed, binary doc values are written using the exact same format as before more aggressive compression got introduced. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 5h 40m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17230912#comment-17230912 ] ASF subversion and git services commented on LUCENE-9378: - Commit a48dc123b3e65928d5e59a76735cf6c88099915a in lucene-solr's branch refs/heads/branch_8x from Adrien Grand [ https://gitbox.apache.org/repos/asf?p=lucene-solr.git;h=a48dc12 ] LUCENE-9378: Make it possible to configure how to trade speed for compression on doc values. (#2069) This adds a switch to `Lucene80DocValuesFormat` which allows to configure whether to prioritize retrieval speed over compression ratio or the other way around. When prioritizing retrieval speed, binary doc values are written using the exact same format as before more aggressive compression got introduced. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 5h 40m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17230801#comment-17230801 ] Nico Tonozzi commented on LUCENE-9378: -- Thank you for the help here folks, and especially [~jpountz]! > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 5h 40m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17230701#comment-17230701 ] ASF subversion and git services commented on LUCENE-9378: - Commit 06877b2c6e47bc481a79d7bedd8ea4fb099f1b4c in lucene-solr's branch refs/heads/master from Adrien Grand [ https://gitbox.apache.org/repos/asf?p=lucene-solr.git;h=06877b2 ] LUCENE-9378: Make it possible to configure how to trade speed for compression on doc values. (#2069) This adds a switch to `Lucene80DocValuesFormat` which allows to configure whether to prioritize retrieval speed over compression ratio or the other way around. When prioritizing retrieval speed, binary doc values are written using the exact same format as before more aggressive compression got introduced. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 5h 40m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17230700#comment-17230700 ] ASF subversion and git services commented on LUCENE-9378: - Commit 06877b2c6e47bc481a79d7bedd8ea4fb099f1b4c in lucene-solr's branch refs/heads/master from Adrien Grand [ https://gitbox.apache.org/repos/asf?p=lucene-solr.git;h=06877b2 ] LUCENE-9378: Make it possible to configure how to trade speed for compression on doc values. (#2069) This adds a switch to `Lucene80DocValuesFormat` which allows to configure whether to prioritize retrieval speed over compression ratio or the other way around. When prioritizing retrieval speed, binary doc values are written using the exact same format as before more aggressive compression got introduced. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 5h 40m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17228681#comment-17228681 ] Michael McCandless commented on LUCENE-9378: {quote}I'll be looking into it for 8.8. {quote} +1, thank you [~jpountz]! > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 4h 20m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17228665#comment-17228665 ] Adrien Grand commented on LUCENE-9378: -- I'll be looking into it for 8.8. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 4h 20m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17228642#comment-17228642 ] Michael McCandless commented on LUCENE-9378: {quote}IMO this is so definitely not a "Minor" matter so I changed it to our default of Major. {quote} +1, thanks [~dsmiley]. This is a major issue for us at Amazon – we are now running the catalog search using a custom Codec that forces (reverts) the whole {{BinaryDocValues}} writing/reading back to before LUCENE-9211, which is not really a comfortable long-term solution. I am hoping that the [ideas being discussed in the PR|https://github.com/apache/lucene-solr/pull/1543] lead to an acceptable solution. I think whether or not {{BinaryDocValues}} fields should be compressed will be very application dependent. Some applications care greatly about the size of the index, and can accept a small hit to search time performance, but for others (like Amazon's!) it is the opposite. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Major > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 4h 10m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17227469#comment-17227469 ] Michael McCandless commented on LUCENE-9378: {quote}I'm wondering if there is anything I should know about if/how it would be a bad idea to continue using the `Lucene70DocValuesFormat` until this issue is resolved? {quote} Just be aware that you don't have the backwards compatibility guarantee you would normally have. E.g. in the future when you upgrade to Lucene 9.x release, it will not be able to read these indices you are creating with {{Lucene70DocValuesFormat}}. Also, few people are fixing bugs in that format, though it is/was widely used so is likely nearly bug free ;) > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 4h 10m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17223292#comment-17223292 ] Alex Klibisz commented on LUCENE-9378: -- Using the `Lucene70DocValuesFormat` gets me back to the pre 8.5.x speed. (Actually a little faster!) I'm wondering if there is anything I should know about if/how it would be a bad idea to continue using the `Lucene70DocValuesFormat` until this issue is resolved? > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 4h 10m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17223132#comment-17223132 ] Alex Klibisz commented on LUCENE-9378: -- > The workaround we were forced to set our codec to a filter codec and use the > {{Lucene70DocValuesFormat}}. Thanks. That gives me an idea.. I think there might be a way to specify an older codec for a specific field in my Elasticsearch plugin. I'll look into this. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 4h 10m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17223016#comment-17223016 ] Nico Tonozzi commented on LUCENE-9378: -- [~alexklibisz] The workaround we were forced to set our codec to a filter codec and use the {{Lucene70DocValuesFormat}}. I'd add another +1 for the option to disable compression. This is going to prevent us from upgrading to a Lucene version that drops support for {{Lucene70DocValuesFormat}}. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 4h 10m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17222962#comment-17222962 ] Michael Gibney commented on LUCENE-9378: {quote}I prefer that BinaryDocValues be simple/fast as it was by default with a configurable option to add some compression. {quote} +1 for an option to bypass compression, however that's accomplished. To note what might be obvious: this change hasn't even fully hit yet in user-land, having been introduced in 8.5. Also maybe worth emphasizing that for bulk-export cases that access values for all docs in arbitrary order (e.g., Solr export handler), iiuc this would result in each block being independently decompressed _for each doc in that block_. Not sure whether there are analogous tools in elasticsearch (or other) ... > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 4h 10m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17222935#comment-17222935 ] Alex Klibisz commented on LUCENE-9378: -- Are there any suggested workarounds for this issue at the moment? The performance hit has basically made it impossible for me to upgrade past 8.4. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 4h 10m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17198646#comment-17198646 ] David Smiley commented on LUCENE-9378: -- Given that docValues (compared to "stored" values) is supposed to optimize for fast field lookups above all else, I prefer that BinaryDocValues be simple/fast as it was by default with a configurable option to add some compression. CompressingStoredFieldsFormat works similarly albeit doesn't have an option to turn compression off altogether. This is easier to support that an entirely separate DocValuesFormat. It's settings could be randomized in the test framework. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 4h 10m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17198570#comment-17198570 ] Michael McCandless commented on LUCENE-9378: {quote}I'm not sure what the path forward is. {color:#172b4d}I'm a bit unhappy of abandoning completely the idea of leveraging redundancy we might see across values.{color} {quote} {color:#172b4d}Can't we leave the (great!) compression on by default, maybe with shared dictionaries across blocks (later), but then allow a backwards compatible option for expert users passing arguments to the default doc values format ctor at segment write time? Can we find a way to do that, so that we [do not really have to support two different doc values formats|https://github.com/apache/lucene-solr/pull/1543#issuecomment-669927391]?{color} > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 4h 10m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17198432#comment-17198432 ] Adrien Grand commented on LUCENE-9378: -- I'm surprised it's slower since other cases that were about scanning the entire data on non-compressible data reported speedups with this new format. I wonder why this one is different. Indeed by asking for use-cases, I was trying to understand whether the solution to this would be to introduce a new type of doc values rather than disabling compression on binary doc values, or even a new format like we're considering for feature vectors. I'm not sure what the path forward is. This trade-off we're making with binary doc values is not unique, it's the same one we're making with postings where you can't decode a single doc ID, you have to decode whole blocks of 128 values. While reverting to the previous format would address the performance regressions that have been reported here, I'm a bit unhappy of abandoning completely the idea of leveraging redundancy we might see across values. Maybe we should look into using dictionaries like we do for stored fields in order to be able to achieve good compression ratios with smaller block sizes. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 4h 10m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17196567#comment-17196567 ] Julie Tibshirani commented on LUCENE-9378: -- Adding another datapoint in case it’s helpful: Elasticsearch uses binary doc values to store high-dimensional float vectors (the `dense_vector` type). The vectors are typically used to score documents. This is a very similar use case to the one [~alexklibisz] describes. In a best case scenario where we index random 100-dimensional float vectors (so they don’t compress well), and scan over all documents (so we don’t decompress data that’s not needed), we see a QPS drop of ~10%: {{Algorithm QPS}} {{Old version 66.989}} {{New version 60.906}} I would guess [~nicot]'s example also falls under the general category of "storing/ loading numeric feature vectors"? For this type of data, it’s common to take a specialized approach instead of applying a general compression algorithm (like using a sparse vector representation, using fewer bytes for each vector element, etc.) > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 4h 10m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17185324#comment-17185324 ] Nico Tonozzi commented on LUCENE-9378: -- Oh, and to clarify, we are not sorting by this field, but we retrieve the field for every hit. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 4h 10m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17183431#comment-17183431 ] Nico Tonozzi commented on LUCENE-9378: -- The most expensive data to load is a serialized [Long2DoubleOpenHashMap|http://fastutil.di.unimi.it/docs/it/unimi/dsi/fastutil/longs/Long2DoubleOpenHashMap.html]. The maps have up to 150 elements in them, and the keys/values are all over the place. The serialized data is up to 2,400 bytes. We are seeing this in a few different cases though, so it's not just one particular type of stored binary data that we are having issues with. We have seen regressions in a few services, but this one with large binary doc values is the only one that has seen drastically worse latency. I'm not sure what proportion of documents are hits, but I suspect that most of the time there is only one hit in the block of 32 compressed doc values, so we are doing much more work to load doc values than we did in the past. Thanks for the help! > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 4h 10m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17183291#comment-17183291 ] Adrien Grand commented on LUCENE-9378: -- [~nicot] Can you give more details about what you're storing in binary doc values. Is it numeric vectors like [~alexklibisz]? > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 4h 10m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17181436#comment-17181436 ] Nico Tonozzi commented on LUCENE-9378: -- We have also noticed this problem at Twitter. The search latency in one of our services went from 2 milliseconds to 15 milliseconds with this change. It would be great to be able to configure the compression or disable it. I can provide profiling data or try to reproduce with minimal code if it would be helpful. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 4h 10m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17140163#comment-17140163 ] Alex Klibisz commented on LUCENE-9378: -- [~jpountz] It's about 2x slower. I re-ran a benchmark to be sure. Here is the setup: * Storing a corpus of 18K binary vectors in a single shard. * Each vector contains ~500 ints denoting the positive indices. So each one is storing a bytearray of 500 * 4 = 2000 bytes in the binary doc values. * Running 2000 serial searches against these vectors. Each search reads, deserializes, and computes the Jaccard similarity against every vector in the corpus. So a total of 18K * 2K reads from the shard. * The read order is defined by Elasticsearch. Internally I'm using a FunctionScoreQuery, code here: [https://github.com/alexklibisz/elastiknn/blob/5246a26f76791362482a98066e31071cb03e0a74/plugin/src/main/scala/com/klibisz/elastiknn/query/ExactQuery.scala#L22-L29] * Ubuntu 20 on an Intel i7-8750H 2.20GHz x 12cores * Running Oracle Jdk 14 : ``` $ java -version java version "14" 2020-03-17 Java(TM) SE Runtime Environment (build 14+36-1461) Java HotSpot(TM) 64-Bit Server VM (build 14+36-1461, mixed mode, sharing) ``` * Running all 2000 searches once, then again, and reporting the time from second run (JVM warmup, etc.). Results: * Using Elasticsearch 7.6.2 w/ Lucene 8.4.0: ** 212 seconds for 2000 searches ** Search threads spend 95.5% of time computing similarities, 0.2% in the LZ4.decompress() method. * Using Elasticsearch 7.7.1 w/ Lucene 8.5.1: ** 445 seconds for 2000 searches ** Search threads spend 56% of total time computing similarities, 42% in the decompress method. VisualVM screenshot for 7.6.x: !hotspots-v76x.png! VisualVM screenshot for 7.7.x: !hotspots-v77x.png! Attaching snapshots from VisualVM: [^snapshots-v76x.nps] [^snapshot-v77x.nps] Thank you all for your help! :) > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: hotspots-v76x.png, hotspots-v76x.png, hotspots-v76x.png, > hotspots-v76x.png, hotspots-v76x.png, hotspots-v77x.png, hotspots-v77x.png, > hotspots-v77x.png, hotspots-v77x.png, image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png, snapshot-v77x.nps, snapshot-v77x.nps, > snapshot-v77x.nps, snapshots-v76x.nps, snapshots-v76x.nps, snapshots-v76x.nps > > Time Spent: 3h 40m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17139977#comment-17139977 ] Adrien Grand commented on LUCENE-9378: -- [~alexklibisz] Thanks for the details, what is the order of magnitude of the slowdown that you are observing? > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png > > Time Spent: 3h 40m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17137925#comment-17137925 ] Alex Klibisz commented on LUCENE-9378: -- [~mgibney] > quick clarification regarding "for every doc in the lucene shard": do your >benchmarks illustrating the regression evaluate the vector query over the full >domain (i.e., literally every (live) doc in the index, without any >pre-filtering of the search domain)? It's reading every live doc in the elasticsearch index, which consists of multiple Lucene shards. I'm not controlling the order. From the perspective of my plugin, I'm just getting a docId from Elasticsearch and using `advanceNext(docId)` to lookup the binary value. Also, there are no deleted docs in this particular case, though there could be in practice. Here's the exact snippet: [https://github.com/alexklibisz/elastiknn/blob/benchmarks-4/plugin/src/main/scala/com/klibisz/elastiknn/query/ExactQuery.scala#L25-L29] > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png > > Time Spent: 3h 20m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17137820#comment-17137820 ] Michael Gibney commented on LUCENE-9378: [~alexklibisz], quick clarification regarding "for every doc in the lucene shard": do your benchmarks illustrating the regression evaluate the vector query over the full domain (i.e., literally every (live) doc in the index, without any pre-filtering of the search domain)? This question is related to [~jpountz]'s comment above: "decompress all values when we need a single one in a block". It would make sense that docId-order access to docValues over the full domain could be faster (e.g., full-domain facets?); selective docId-order access (i.e. over a filtered domain) could be slower; arbitrary (non-docId-order) access over the full domain would likely be a worst-case scenario (wrt the impact of block size), all else being equal. The last of these would affect bulk-export-type use cases, accessing docValues for each doc in arbitrary order. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png > > Time Spent: 3h 20m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17136668#comment-17136668 ] Alex Klibisz commented on LUCENE-9378: -- [~jpountz] Sure, I'll explain below: My plugin is doing nearest neighbors search on sparse and dense vectors. Neighbors are "near" based on a similarity score like L1, L2, Angular, Hamming, or Jaccard similarity. Also, when I say "Vector" I mean it in the math/physics sense, not the "Vector" style of data structure. The only relevant data structure for storing a vector is a simple arrays of floats or ints. I'm storing the contents of each vector in binary doc values. So for dense floating point vectors, I store the literal numbers (e.g. 0.9,0.22,1.234,...) as a `float[]`. And for sparse boolean vectors, I store indices where the vector is "true" (1,22,99,101,...) as a `int[]`. In both cases the ints and floats are serialized as a byte array using the sun.misc.Unsafe module and passed to Lucene as a `new BinaryDocValuesField()`. From the perspective of Lucene, the serialization protocol shouldn't matter. I could just as well be using an ObjectOutputStream, DataOutputStream, etc. Typical vector length is ~1000, so 1000 4-byte ints/floats produces `byte[]` with length 4000. I also experimented with variable-length encoding schemes, but determined it wasn't saving much space at all since Lucene was already compressing the byte array. My benchmark just repeatedly runs queries against a corpus of these stored vectors. So it's loop like: * get a query vector * for every doc in the lucene shard ** read the vector corresponding to the doc from binary doc values (this is LZ4.decompress() part that got much slower with the upgrade to 8.5.0). ** convert the bytearray to an array of floats or ints ** compute the similarity score of the array against the query vector (this is the `sortedIntersectionCount` in the screenshots I posted). ** return the score The code is in a very experimental state, but if it helps I can try to clean it up and make it reproducible for others. It seems like a nice solution would be the ability to configure or disable the level of compression when I store a BinaryDocValuesField. Or maybe there is another way to store these vectors that avoid the compression overhead? I'm open to other options. I'm much more familiar with Elasticsearch internals than I am with Lucene internals. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png > > Time Spent: 3h 20m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17136454#comment-17136454 ] Adrien Grand commented on LUCENE-9378: -- [~alexklibisz] Can you give more details about what you are storing in binary doc values and what your benchmark does? > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png > > Time Spent: 3h 20m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17134629#comment-17134629 ] Alex Klibisz commented on LUCENE-9378: -- Hi, just as another datapoint: I'm using BinaryDocValues to store vectors for this elasticsearch plugin: [https://github.com/alexklibisz/elastiknn, |https://github.com/alexklibisz/elastiknn] The usecase is actually very similar to what [~sokolov] described. I saw a large regression after switching from elasticsearch 7.6.x to 7.7.x, which introduces Lucene 8.5.0. For instance, here are two screenshots from visualvm running the same benchmark on 7.6.x and then 7.7.x. 7.7.x spends a lot more time in the `decompress` method, and actually overtakes the `sortedIntersectionCount` method that was previously most expensive. !image-2020-06-12-22-18-48-919.png|width=732,height=50! !image-2020-06-12-22-18-24-527.png! Note that this is also comparing Oracle JDK 13 (7.6.x) to Oracle JDK 14 (7.7.x). As a sanity check, I benchmarked the sortedIntersectionCount independently and it did get faster after the JDK switch. I can provide more detailed info if necessary. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > Attachments: image-2020-06-12-22-17-30-339.png, > image-2020-06-12-22-17-53-961.png, image-2020-06-12-22-18-24-527.png, > image-2020-06-12-22-18-48-919.png > > Time Spent: 3h 20m > Remaining Estimate: 0h > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17116719#comment-17116719 ] Michael Sokolov commented on LUCENE-9378: - Another use case I'd like to plan for is using {{BinaryDocValues}} to store numeric vectors and retrieve them for time-sensitive computations (think 64-256 dimension floating point numbers, or maybe bytes, so something in the range of 64-1024 bytes per document). Probably these would be somewhat compressible if a number of the dimensions take on zero values? We can revisit later if/when this use case becomes problematic, but I think it could require some manual control to preserve best query performance. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17116713#comment-17116713 ] Michael Sokolov commented on LUCENE-9378: - {quote}... disabled compression for short binary values ... {quote} That seems like a good compromise. It should preserve the main benefit of compression. Our main use case is 10-character values, although we have a related field that could be larger - a threshold somewhere in 32-64 bytes makes sense to me. That would capture eg UUIDs even with their hyphens. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17116518#comment-17116518 ] Adrien Grand commented on LUCENE-9378: -- I profiled some of these sorting tasks to understand where time is spent, and while there is non-negligible time spent reading lengths, the bulk of the CPU time is spent decompressing bytes given how highly compressible titles are in wikimedium with lots of exact duplicates. Furthermore, in your case, decoding the lengths is likely even cheaper given that all documents have the same length. We have discussed building dictionaries in the past, in order to not have to decompress all values when we need a single one in a block. This was initially for stored fields, but I believe that this could help in this case here. It's not really a low hanging fruit though. Would it be an ok workaround for you if, in the meantime, we just disabled compression for short binary values? E.g. assuming that the average length of values in a block is small enough, then we'd write values without compression. This would help avoid introducing a flag. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17116167#comment-17116167 ] Michael Sokolov commented on LUCENE-9378: - {quote}... binary doc values at Amazon, ie. are adjacent doc IDs likely to get the exact same value? {quote} Not identical values, but the field we use as a sorting tiebreaker, that is binary docvalues, will be in sorted order in the index, so adjacent values may be very similar, all the same length, drawn from a limited set of characters, and maybe sequential ones differ only in the last character or few. Highly compressible I would think? And we probably usually only need a relatively small number from each block. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17115885#comment-17115885 ] Adrien Grand commented on LUCENE-9378: -- bq. Another factor that probably plays a role here is how compressible the data is I looked a bit more into the data we use for benchmarking. wikibigall already makes titles quite compressible given how linefile docs are sorted by title in the input file. wikimedium makes it way more compressible given how it splits articles into 1kB chunks that all share the same title, creating many duplicate values for adjacent doc IDs. This makes wikimedium titles a best-case scenario for compression (and thus a worst-case scenario for search speed) and I'd expect performance numbers to be significantly different between wikimediumall and wikibigall, and again between wikibigall and a shuffled copy of wikibigall that would no longer sort by title. Are wikimedium titles representative of the data that you are indexing into binary doc values at Amazon, ie. are adjacent doc IDs likely to get the exact same value? If that's the case, then we could probably add ad-hoc compression for this case which would have a better runtime than LZ4, and we could automatically make the decision at index time instead of requiring users to configure a flag. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17115553#comment-17115553 ] Michael McCandless commented on LUCENE-9378: {quote}As an alternative to API changes and on/off stuff (which is costly, it means supporting multiple formats), maybe the compression should be less aggressive, similar to what was implemented for the terms dictionary? {quote} +1, if we can find that compromise. Stored fields offers {{BEST}} and {{FAST}}; maybe we can find the same balanced options here? > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17114727#comment-17114727 ] Robert Muir commented on LUCENE-9378: - As an alternative to API changes and on/off stuff (which is costly, it means supporting multiple formats), maybe the compression should be less aggressive, similar to what was implemented for the terms dictionary? > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17114706#comment-17114706 ] Michael Sokolov commented on LUCENE-9378: - This is the output of the "normal" wikimediumall tests plus one additional task I'm proposing to add for the BDV sorting case: || Task|| QPS before|| StdDev|| QPS after|| StdDev|| Pct diff|| | HighTermTitleBDVSort| 5.65| (2.9%)| 1.03| (0.7%)| -81.7% ( -82% - -80%)| | Respell| 95.20| (1.9%)| 75.76| (1.2%)| -20.4% ( -23% - -17%)| | Fuzzy2| 83.74| (6.1%)| 79.80| (3.8%)| -4.7% ( -13% - 5%)| | HighIntervalsOrdered| 3.45| (1.8%)| 3.31| (2.5%)| -4.0% ( -8% - 0%)| | Wildcard| 95.69| (11.9%)| 91.86| (11.6%)| -4.0% ( -24% - 22%)| | OrHighNotMed| 735.62| (3.3%)| 707.58| (3.3%)| -3.8% ( -10% - 2%)| | LowTerm| 1409.80| (5.4%)| 1356.41| (3.7%)| -3.8% ( -12% - 5%)| | OrHighNotLow| 602.60| (3.8%)| 582.17| (3.6%)| -3.4% ( -10% - 4%)| | MedTerm| 1067.74| (3.4%)| 1034.85| (3.0%)| -3.1% ( -9% - 3%)| |BrowseDayOfYearSSDVFacets| 3.23| (5.0%)| 3.14| (10.0%)| -2.6% ( -16% - 13%)| |BrowseDayOfYearTaxoFacets| 5096.02| (2.9%)| 4962.76| (3.0%)| -2.6% ( -8% - 3%)| | BrowseMonthTaxoFacets| 5131.16| (2.9%)| 5007.29| (3.2%)| -2.4% ( -8% - 3%)| | MedPhrase| 453.30| (3.1%)| 446.26| (2.3%)| -1.6% ( -6% - 3%)| | BrowseMonthSSDVFacets| 3.91| (11.4%)| 3.86| (13.0%)| -1.0% ( -22% - 26%)| | HighPhrase| 37.99| (3.1%)| 37.59| (3.4%)| -1.0% ( -7% - 5%)| | OrNotHighLow| 552.20| (3.7%)| 546.81| (3.1%)| -1.0% ( -7% - 6%)| | OrHighNotHigh| 679.14| (2.8%)| 673.74| (3.2%)| -0.8% ( -6% - 5%)| | MedSpanNear| 34.13| (2.2%)| 33.91| (3.4%)| -0.6% ( -6% - 5%)| | OrHighLow| 219.19| (2.9%)| 217.83| (2.9%)| -0.6% ( -6% - 5%)| | OrHighMed| 25.65| (2.9%)| 25.51| (2.4%)| -0.6% ( -5% - 4%)| | LowSloppyPhrase| 62.29| (5.0%)| 61.96| (5.9%)| -0.5% ( -10% - 10%)| | AndHighLow| 387.57| (2.6%)| 385.73| (3.6%)| -0.5% ( -6% - 5%)| | Prefix3| 60.91| (4.1%)| 60.68| (3.4%)| -0.4% ( -7% - 7%)| | HighTerm| 806.80| (5.8%)| 804.20| (4.8%)| -0.3% ( -10% - 10%)| | LowPhrase| 59.60| (3.4%)| 59.46| (3.2%)| -0.2% ( -6% - 6%)| | LowSpanNear| 36.39| (2.0%)| 36.38| (2.6%)| -0.0% ( -4% - 4%)| | HighSloppyPhrase| 4.79| (8.8%)| 4.79| (9.2%)| 0.0% ( -16% - 19%)| | OrHighHigh| 8.88| (3.2%)| 8.89| (2.5%)| 0.1% ( -5% - 5%)| | MedSloppyPhrase| 19.58| (4.8%)| 19.61| (4.2%)| 0.1% ( -8% - 9%)| | Fuzzy1| 110.96| (6.5%)| 111.14| (4.2%)| 0.2% ( -9% - 11%)| | AndHighMed| 20.09| (4.5%)| 20.15| (4.8%)| 0.3% ( -8% - 10%)| | AndHighHigh| 13.78| (4.5%)| 13.82| (6.2%)| 0.3% ( -9% - 11%)| | HighSpanNear| 1.53| (2.6%)| 1.53| (3.5%)| 0.3% ( -5% - 6%)| | HighTermDayOfYearSort| 16.38| (7.4%)| 16.47| (6.2%)| 0.5% ( -12% - 15%)| | HighTermMonthSort| 14.09| (5.5%)| 14.19| (5.1%)| 0.8% ( -9% - 12%)| | IntNRQ| 37.12| (1.9%)| 37.47| (0.8%)| 0.9% ( -1% - 3%)| | OrNotHighMed| 615.21| (2.6%)| 624.34| (2.5%)| 1.5% ( -3% - 6%)| | OrNotHighHigh| 475.15| (3.3%)| 486.16| (2.7%)| 2.3% ( -3% - 8%)| | BrowseDateTaxoFacets| 0.61| (5.2%)| 0.96| (4.5%)| 57.7% ( 45% - 71%)| > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene80DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17114010#comment-17114010 ] Adrien Grand commented on LUCENE-9378: -- The code suggests that block decompression happens only once per block indeed. I'm not very familiar with the facets tasks, do they consume all docs by any chance? A side-effect of bulk-decoding multiple values at once is that selective queries get slower because they likely decompress values that they don't need, but queries that match most documents like MatchAllDocsQuery might get faster. Another factor that probably plays a role here is how compressible the data is. The compression logic we're using is fast when data is barely compressible and gets slower if the data is highly compressible. So depending on how compressible the data is, performance results could be extremely different. Maybe we should update the Disk usage tool (https://github.com/mikemccand/luceneutil/blob/master/src/main/perf/DiskUsage.70.java) to work with the Lucene84 and Lucene86 codecs to get a clearer picture about the storage savings on a per-field basis. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene84DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17113960#comment-17113960 ] Michael McCandless commented on LUCENE-9378: Thanks for running the luceneutil benchmarks Michael Sokolov! {quote}Interestingly, BrowseDateTaxoFacets shows a big improvement! But otherwise we see a pretty significant degradation in performance. {quote} That is fascinating, because faceting uses BINARY DV to hold all ordinals. I wonder whether the BINARY DV compression somehow makes faceting faster!? Could you try running the tasks w/ normal relevance sort to see impact on {{BrowseDateTaxoFacets}}? (So we can separate "sorting by BINARY compressed" from "faceting on BINARY compressed"). Robert Muir also suggested this idea: have we verified that the block decompression only happens once per block, when we {{.advance}} to multiple (increasing) docids in the block? The sizable performance hits are so big in the results above that it makes me wonder if we are accidentally decompressing on every {{.advance}} rather than once per block. Also, I wonder why the original benchmarks on the issue didn't uncover similar performance changes. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene84DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17113535#comment-17113535 ] Michael Sokolov commented on LUCENE-9378: - Here are the index file sizes (after merging to a single segment). In total there was a ~6.5% reduction in index size, although the doc values (dvd) file reduced quite a bit more, ~28% h3. Before |4|_h4.dii| |276168|_h4.dim| |892220|_h4.fdt| |840|_h4.fdx| |4|_h4.fnm| |1981564|_h4_Lucene80_0.dvd| |24|_h4_Lucene80_0.dvm| |5111752|_h4_Lucene84_0.doc| |4108112|_h4_Lucene84_0.pos| |1145544|_h4_Lucene84_0.tim| |23268|_h4_Lucene84_0.tip| |65104|_h4.nvd| |4|_h4.nvm| |4|_h4.si| |4|segments_3| |0|write.lock| |13604636|TOTAL| h3. After |4|_h5.dii| |276480|_h5.dim| |12|_h5.fdm| |889700|_h5.fdt| |820|_h5.fdx| |4|_h5.fnm| |1421700|_h5_Lucene80_0.dvd| |4|_h5_Lucene80_0.dvm| |5111616|_h5_Lucene84_0.doc| |4108024|_h5_Lucene84_0.pos| |848876|_h5_Lucene84_0.tim| |23244|_h5_Lucene84_0.tip| |65104|_h5.nvd| |4|_h5.nvm| |4|_h5.si| |4|segments_3| |12745620|TOTAL| > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene84DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17113506#comment-17113506 ] Adrien Grand commented on LUCENE-9378: -- If you still have the indices, I'd be curious to know what the difference in disk usage is too. > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene84DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[jira] [Commented] (LUCENE-9378) Configurable compression for BinaryDocValues
[ https://issues.apache.org/jira/browse/LUCENE-9378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17113489#comment-17113489 ] Michael Sokolov commented on LUCENE-9378: - I updated luceneutil to enable sorting by a BinaryDocValues field over title and ran a test t across a wide range of tasks, comparing branch_8_4 (before) and branch_8_5 (after). In this test, all tasks have a title sort criterion applied. Interestingly, BrowseDateTaxoFacets shows a big improvement! But otherwise we see a pretty significant degradation in performance. ||Task||QPS before||StdDev||QPS after||StdDev||Pct diff|| |MedTerm|9.40|(3.1%)|1.68|(0.4%)|-82.1% ( -83% - -81%)| |LowTerm|20.17|(1.8%)|3.74|(0.4%)|-81.5% ( -82% - -80%)| |Wildcard|5.25|(3.3%)|1.02|(0.4%)|-80.6% ( -81% - -79%)| |Prefix3|12.83|(2.3%)|2.52|(0.4%)|-80.4% ( -81% - -79%)| |OrHighLow|3.07|(4.1%)|0.71|(0.6%)|-76.9% ( -78% - -75%)| |HighTerm|2.79|(4.6%)|0.72|(0.5%)|-74.1% ( -75% - -72%)| |Fuzzy2|19.88|(2.7%)|5.16|(0.5%)|-74.0% ( -75% - -72%)| |IntNRQ|329.04|(1.4%)|85.42|(0.4%)|-74.0% ( -74% - -73%)| |AndHighHigh|5.44|(3.1%)|1.52|(0.6%)|-72.1% ( -73% - -70%)| |AndHighMed|7.85|(2.4%)|2.55|(0.6%)|-67.4% ( -68% - -65%)| |LowSloppyPhrase|5.11|(2.4%)|1.90|(0.6%)|-62.9% ( -64% - -61%)| |OrHighHigh|1.47|(4.2%)|0.56|(1.0%)|-61.7% ( -64% - -58%)| |LowPhrase|8.21|(1.9%)|3.23|(0.6%)|-60.6% ( -61% - -59%)| |HighSloppyPhrase|1.48|(3.2%)|0.61|(0.9%)|-58.9% ( -61% - -56%)| |Fuzzy1|112.25|(5.7%)|46.46|(1.1%)|-58.6% ( -61% - -54%)| |MedSloppyPhrase|2.16|(3.0%)|0.94|(0.7%)|-56.5% ( -58% - -54%)| |OrHighMed|1.23|(4.4%)|0.54|(1.2%)|-55.9% ( -58% - -52%)| |MedPhrase|2.87|(2.6%)|1.77|(1.0%)|-38.5% ( -40% - -35%)| |HighPhrase|0.28|(3.3%)|0.21|(1.9%)|-24.1% ( -28% - -19%)| |HighIntervalsOrdered|0.48|(4.7%)|0.41|(2.9%)|-16.2% ( -22% - -9%)| |Respell|99.24|(1.7%)|86.51|(0.8%)|-12.8% ( -15% - -10%)| |AndHighLow|302.35|(2.5%)|276.95|(2.6%)|-8.4% ( -13% - -3%)| |BrowseDayOfYearTaxoFacets|4202.04|(3.0%)|4057.48|(2.6%)|-3.4% ( -8% - 2%)| |BrowseMonthTaxoFacets|4160.07|(2.8%)|4080.02|(2.2%)|-1.9% ( -6% - 3%)| |BrowseDayOfYearSSDVFacets|3.29|(4.9%)|3.29|(7.1%)|0.0% ( -11% - 12%)| |BrowseMonthSSDVFacets|3.68|(15.7%)|3.69|(16.9%)|0.3% ( -27% - 39%)| |BrowseDateTaxoFacets|0.54|(6.3%)|0.96|(5.8%)|77.3% ( 61% - 95%)| > Configurable compression for BinaryDocValues > > > Key: LUCENE-9378 > URL: https://issues.apache.org/jira/browse/LUCENE-9378 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Viral Gandhi >Priority: Minor > > Lucene 8.5.1 includes a change to always [compress > BinaryDocValues|https://issues.apache.org/jira/browse/LUCENE-9211]. This > caused (~30%) reduction in our red-line QPS (throughput). > We think users should be given some way to opt-in for this compression > feature instead of always being enabled which can have a substantial query > time cost as we saw during our upgrade. [~mikemccand] suggested one possible > approach by introducing a *mode* in Lucene84DocValuesFormat (COMPRESSED and > UNCOMPRESSED) and allowing users to create a custom Codec subclassing the > default Codec and pick the format they want. > Idea is similar to Lucene50StoredFieldsFormat which has two modes, > Mode.BEST_SPEED and Mode.BEST_COMPRESSION. > Here's related issues for adding benchmark covering BINARY doc values > query-time performance - [https://github.com/mikemccand/luceneutil/issues/61] -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org