summaryrefslogtreecommitdiffstats
path: root/searchcore
Commit message (Collapse)AuthorAgeFilesLines
* Modernize code while reading it.Henning Baldersheim2023-05-311-0/+1
|
* Prevent possible deadlock when calling State::cancel by dropping the ↵Henning Baldersheim2023-05-261-5/+9
| | | | taskList lock first.
* Ensure correct destruction order to uphold life time guarantees.Henning Baldersheim2023-05-262-3/+4
|
* The AttributeContext is a short lived cache for attributes guards. Until we ↵Henning Baldersheim2023-05-264-18/+42
| | | | | | | use the thread bundle we do not need to use expensive locking to fill the cache. Most of the attributes are pulled in when building the blueprint tree, and that always happens singlethreaded.
* - Make the MatchContext value object movable.Henning Baldersheim2023-05-2612-62/+78
| | | | - Reduce code visibility.
* - Use a rwlock as reading surpasses writing by a very large factor.Henning Baldersheim2023-05-251-7/+2
| | | | - size() does not need a lock.
* Use vespalib::hash_map over std::unordered_mapHenning Baldersheim2023-05-242-3/+3
|
* Use a smallvector to avoid indirection in most common case, and reserve some ↵Henning Baldersheim2023-05-223-4/+5
| | | | space upfront
* Merge pull request #27144 from vespa-engine/balder/reduce-number-of-allocationsv8.165.119Henning Baldersheim2023-05-211-10/+9
|\ | | | | Balder/reduce number of allocations
| * Make single field createBlueprint accessibleHenning Baldersheim2023-05-161-10/+9
| |
* | GC unused assert includesHenning Baldersheim2023-05-173-2/+3
|/
* Merge pull request #27123 from ↵Henning Baldersheim2023-05-164-53/+76
|\ | | | | | | | | vespa-engine/balder/use-multiple-threads-for-pruning Use multiple threads for pruning the session caches.
| * Use std::make_unique/make_shared instead of explicit new.Henning Baldersheim2023-05-161-24/+20
| |
| * Ensure that we get at least 1 bundle.Henning Baldersheim2023-05-162-6/+5
| |
| * GC unused includeHenning Baldersheim2023-05-151-1/+0
| |
| * Use multiple threads for pruning the session caches.Henning Baldersheim2023-05-153-29/+58
| |
* | To avoid an inherent race in testing flush versus prune instead test for ↵Henning Baldersheim2023-05-151-1/+5
|/ | | | | | what should certainly not be present, and for a smaller subset of what will normally be present in the prune history.
* Use a timeBomb to ensure test terminates on error.Henning Baldersheim2023-05-151-0/+3
|
* Ensure thread safety of handlA.Henning Baldersheim2023-05-151-6/+10
| | | | Use a loop with short wait to speed up test.
* Show log when error happens.Henning Baldersheim2023-05-151-1/+1
|
* Remove unused field/attribute access hinting.Tor Egge2023-05-132-8/+0
|
* Reduce included code.Henning Baldersheim2023-05-1218-89/+91
|
* Merge pull request #27080 from ↵Henning Baldersheim2023-05-123-5/+4
|\ | | | | | | | | vespa-engine/balder/avoid-copying-fieldspecbase-vector Move the fieldspec base vector.
| * Use SmallVector to avoid allocation in the common case.Henning Baldersheim2023-05-121-1/+1
| |
| * As a single field per blueprint is the common use case, make that optimal ↵Henning Baldersheim2023-05-121-1/+1
| | | | | | | | more optimal.
| * Move the fieldspec base vector.Henning Baldersheim2023-05-111-3/+2
| |
* | Only consider the next target in the Q for high priority flush.Henning Baldersheim2023-05-121-4/+2
| |
* | Use c++11 loops.Henning Baldersheim2023-05-121-21/+8
| |
* | Test that high priority tragets do not skip the queue.Henning Baldersheim2023-05-121-28/+63
|/
* Ensure we have enough threads in the flushengine thread pool.Henning Baldersheim2023-05-103-5/+10
|
* Setup search visitor without proton process.Tor Egge2023-05-102-2/+2
|
* Merge pull request #27055 from vespa-engine/balder/propagate-metastorereadguardHenning Baldersheim2023-05-1014-33/+28
|\ | | | | Balder/propagate metastorereadguard
| * Grab the fallback guard in the constructor to ensure it is held fetching the ↵Henning Baldersheim2023-05-104-10/+6
| | | | | | | | postinglists and generating possible bitvector.
| * Wire the metastore read guard all the way and use it, if present.Henning Baldersheim2023-05-094-8/+5
| |
| * Propagate DocumentMetaStore read gaurd to RequestContext.Henning Baldersheim2023-05-098-16/+18
| |
* | Pass transport and file distributor connection spec to SearchEnvironmentTor Egge2023-05-101-4/+20
|/ | | | | in preparation for using RankingAssetsBuilder when handling config in streaming search.
* Merge pull request #27052 from vespa-engine/balder/use-shared-ptr-for-read-guardHenning Baldersheim2023-05-0914-62/+38
|\ | | | | Use shared_ptr for the read guard
| * Simplify to avoid type declarations.Henning Baldersheim2023-05-092-36/+10
| |
| * Use shared_ptr for the read guardHenning Baldersheim2023-05-0913-28/+30
| |
* | Add RankingAssetsBuilder.Tor Egge2023-05-095-104/+52
|/
* Merge pull request #27029 from ↵Henning Baldersheim2023-05-093-21/+91
|\ | | | | | | | | vespa-engine/balder/always-keep-room-for-one-extra-high-priority-flush Leave at least one slot available for high priority flush targets.
| * Test that only 1 high priority target gets a slot when all normal slots are ↵Henning Baldersheim2023-05-091-8/+18
| | | | | | | | taken.
| * Just wait unconditionallyHenning Baldersheim2023-05-081-3/+1
| |
| * Leave at least one slot available for high priority flush targets.Henning Baldersheim2023-05-083-21/+83
| |
* | Export moved config definitions.Tor Egge2023-05-0943-615/+10
| |
* | Move ranking config to configdefinitions module.Tor Egge2023-05-0936-542/+105
| |
* | Revert "Move ranking config to configdefinitions module"Harald Musum2023-05-0836-105/+542
| |
* | Move ranking config to configdefinitions module.Tor Egge2023-05-0836-542/+105
|/
* Remove double bookeeping and simplify.Henning Baldersheim2023-05-082-7/+6
|
* Wait for a pending prune or a high priority slot before continuing.Henning Baldersheim2023-05-082-4/+4
|