aboutsummaryrefslogtreecommitdiffstats
path: root/searchlib
Commit message (Collapse)AuthorAgeFilesLines
* Ensure we retrieve the actual vector (used by nearest neighbor index) for a ↵Geir Storli2020-03-032-3/+3
| | | | | | | | | | | document being added instead of the origo vector. Before the fix, getTensor() would return the default origo tensor as docid would not be less than getCommittedDocIdLimit(). This happened as the docid limit is updated at the end of handling a document being added, which is too late. The fix is to ignore the docid limit in is this particular case. Also update the unit test for tensor attributes to do commit() (which updates the docid limit) later. Before the fix, this triggered the bug in the test "setTensor() updates nearest neighbor index".
* Merge pull request #12381 from vespa-engine/balder/use-single-threaded-executorHenning Baldersheim2020-03-023-4/+22
|\ | | | | Balder/use single threaded executor
| * Wire in control of executor type.Henning Baldersheim2020-02-283-6/+6
| |
| * Add comment about thread safety.Henning Baldersheim2020-02-271-1/+6
| |
| * Add single threaded thoughput optimized executor with high and low watermark ↵Henning Baldersheim2020-02-273-4/+17
| | | | | | | | at 25% / 75%.
* | Merge pull request #12367 from vespa-engine/arnej/actual-random-levelsGeir Storli2020-02-285-7/+82
|\ \ | |/ |/| make actual random levels
| * add class commentArne Juul2020-02-282-0/+26
| |
| * make actual random levelsArne Juul2020-02-275-7/+56
| |
* | Add getStats and setTaskLimit to interface to make it easy to swap ↵Henning Baldersheim2020-02-2714-126/+165
| | | | | | | | | | | | implementation. Also make do with ISequenceHandlerInterface.
* | Add a += operator and test ExecutorStatsHenning Baldersheim2020-02-271-4/+1
|/
* refactor to reuse make_node_for_documentArne Juul2020-02-272-18/+8
|
* remove outdated TODOArne Juul2020-02-271-1/+0
|
* rename config variableArne Juul2020-02-263-7/+7
|
* shrink links if neededArne Juul2020-02-263-17/+159
| | | | | | * select "M" links on all layers * change unit test parameters to avoid triggering shrink early * add symmetry validation method for use from unit test
* Merge pull request #12337 from ↵Arne H Juul2020-02-268-18/+204
|\ | | | | | | | | vespa-engine/geirst/memory-management-in-hnsw-index Add proper memory management to hnsw index and integrate with dense t…
| * Reclaim memory when doing changes to the graph in the hnsw index.Geir Storli2020-02-265-8/+104
| |
| * Add proper memory management to hnsw index and integrate with dense tensor ↵Geir Storli2020-02-257-10/+100
| | | | | | | | attribute.
* | Trunkcate expressions in messagesJon Bratseth2020-02-251-1/+2
|/
* Merge pull request #12328 from vespa-engine/balder/add-executor-benchmarkHenning Baldersheim2020-02-256-5/+33
|\ | | | | Balder/add executor benchmark
| * GC unused include and update copyright.Henning Baldersheim2020-02-254-3/+2
| |
| * Make final and GC unused includeHenning Baldersheim2020-02-242-3/+1
| |
| * Add benchmark for sequenced task executor.Henning Baldersheim2020-02-242-0/+31
| |
* | Merge pull request #12323 from vespa-engine/arnej/add-set-nodeArne H Juul2020-02-253-4/+59
|\ \ | |/ |/| add set_node for unit testing
| * minor fixes after reviewArne Juul2020-02-253-2/+7
| |
| * add set_node for unit testingArne Juul2020-02-243-4/+54
| |
* | Merge pull request #12315 from vespa-engine/balder/follow-intellij-adviseHenning Baldersheim2020-02-2465-387/+345
|\ \ | | | | | | Follow clion advise
| * | Add tests for bool, byte, long, and double attributes.Henning Baldersheim2020-02-241-7/+35
| | |
| * | - Add debug logging.Henning Baldersheim2020-02-2363-296/+247
| | | | | | | | | | | | | | | - std::make_unique - Reduce code visibility.
| * | Follow clion adviseHenning Baldersheim2020-02-232-84/+63
| |/
* | more accurate estimateArne Juul2020-02-221-1/+5
| |
* | keep "Neighbor" as name of structArne Juul2020-02-221-4/+4
| |
* | drop unneccessary passing of unused strict flagArne Juul2020-02-224-5/+2
|/
* Merge pull request #12289 from ↵Håvard Pettersen2020-02-212-0/+33
|\ | | | | | | | | vespa-engine/toregge/fixup-feature-map-on-blueprint-setup-failures Fixup feature map after blueprint setup failure.
| * Call fixup_feature_map() from failed() method.Tor Egge2020-02-211-3/+1
| |
| * Fixup feature map after blueprint setup failure.Tor Egge2020-02-202-0/+35
| |
* | add commentArne Juul2020-02-202-0/+6
| |
* | use NearestNeighborIndex when availableArne Juul2020-02-207-14/+45
| | | | | | | | | | | | | | | | * update find_top_k API in NearestNeighborIndex to return vector with both docids and distances. * constify find_top_k * when possible, call find_top_k from NearestNeighborBlueprint, and use result from that call in iterator.
* | Merge pull request #12281 from vespa-engine/arnej/add-nns-iteratorArne H Juul2020-02-2010-9/+187
|\ \ | | | | | | Arnej/add nns iterator
| * | unit test vector iteratorArne Juul2020-02-201-0/+67
| | |
| * | add search iterator using result from find_top_kArne Juul2020-02-209-9/+120
| |/
* | More use of TensorSpec instead of creating Tensor instance explicit.Geir Storli2020-02-201-60/+51
| |
* | Update nearest neighbor index when tensor values change.Geir Storli2020-02-203-11/+187
| |
* | Instantiate nearest neighbor (hnsw) index in dense tensor attribute when ↵Geir Storli2020-02-2013-49/+212
| | | | | | | | specified in config.
* | Add hnsw index params to config for attribute vector.Geir Storli2020-02-202-0/+20
|/
* explore at least k alwaysArne Juul2020-02-191-1/+1
|
* expose explore_k in top level APIArne Juul2020-02-194-19/+17
|
* test top-level API alsoArne Juul2020-02-171-1/+9
|
* add search API in NearestNeighborIndexArne Juul2020-02-174-3/+22
|
* test find_top_k alsoArne Juul2020-02-171-1/+21
|
* add simple find_top_k methodArne Juul2020-02-172-0/+23
|