summaryrefslogtreecommitdiffstats
path: root/container-search/src/main/java/com/yahoo/search/dispatch/InterleavedSearchInvoker.java
Commit message (Collapse)AuthorAgeFilesLines
* - Do not use Searcher interface below the prelude cluster searcher.Henning Baldersheim2024-03-081-3/+3
| | | | | | - It is not necessary and provides no value. - It also hides some important aspects, like the schema. - And it simplifies testing.
* Update copyrightJon Bratseth2023-10-091-1/+1
|
* Make a few simpler interfaces instead of carrying one huge implement all ↵Henning Baldersheim2022-11-231-9/+18
| | | | SearchCluster around.
* Various cleanup and code deduplication.Henning Baldersheim2022-10-131-8/+8
|
* Remove all traces from computeCoverageFromTargetActiveDocsHenning Baldersheim2022-09-291-6/+2
|
* Chose correct "redundancy" for coverage estimation. Current default is to ↵Henning Baldersheim2022-09-201-1/+5
| | | | | | use searchable-copies. But when we flip to use target-active directly from the backend we need to use redundancy.
* Control whether to report coverage based on active or target active documents.Henning Baldersheim2022-09-191-1/+1
|
* Clarify semantics by better naming and correct code visibilityHenning Baldersheim2022-09-141-6/+6
|
* Factor out timeout and coverage handling to make the ↵Henning Baldersheim2022-09-141-121/+23
| | | | InterleavedSearchInvoker easier to understand and modify.
* soonActive => targetActiveHenning Baldersheim2022-09-091-5/+5
| | | | Wire in targetActive in MonitorReply/Pong.
* Propagate target-active-docs in monitor reply.Henning Baldersheim2022-09-091-2/+2
|
* Revert "Do not claim that you have asked nodes that were in a known failed ↵Henning Baldersheim2022-07-081-0/+3
| | | | st…"
* Do not claim that you have asked nodes that were in a known failed state.Henning Baldersheim2022-07-081-3/+0
| | | | This makes coverage computation incorrect.
* Remove feature flag 'merge-grouping-result-in-search-invoker'Bjørn Christian Seime2022-03-301-5/+2
|
* Add FlatteningSearcherJon Bratseth2022-03-071-31/+0
|
* Merge grouping results incrementally in search invokerBjørn Christian Seime2022-03-011-3/+14
|
* Log number of working/nodes and nodes.Henning Baldersheim2021-11-051-1/+0
|
* Update 2018 copyright notices.gjoranv2021-10-071-1/+1
|
* Enable top-k optimization only if balanced and non-sparseJon Bratseth2021-07-021-4/+8
|
* Non-functional changes onlyJon Bratseth2021-04-121-11/+1
|
* only do debug logging in very specific casesArne Juul2021-03-261-16/+34
|
* add more debug logging about hit mergingArne Juul2021-03-251-5/+33
|
* Revert "Revert "Disable topk optimisation on dispatch when content ↵Henning Baldersheim2021-01-081-5/+10
| | | | distribution is se…""
* Revert "Disable topk optimisation on dispatch when content distribution is ↵Henning Baldersheim2021-01-081-10/+5
| | | | se…"
* Disable topk optimisation on dispatch when content distribution is severly ↵Henning Baldersheim2021-01-071-5/+10
| | | | | | | | skewed. When the skew is too large the assumption that docs are evenly and randomly distributed hold. The impact and is larger on smaller systems. In large systems the where this optimisation is more important, the probabilitity of large skew will be less.
* Add support for serializing and compressing once, instead of once per ↵Henning Baldersheim2020-06-301-2/+4
| | | | backend node.
* Add query control of top-k-probability.Henning Baldersheim2020-04-151-1/+9
|
* Introduce top-k-probability and use it to fetch correct proper amount of ↵Henning Baldersheim2020-04-151-1/+1
| | | | hits from each partition
* Support max-hits-per-node in DispatcherJon Bratseth2020-01-081-0/+1
|
* Revert "Revert "Revert "Revert "Use a LeanHit until merging is done." MERGEOK"""Henning Baldersheim2019-09-161-49/+25
|
* Revert "Revert "Revert "Use a LeanHit until merging is done." MERGEOK""Henning Baldersheim2019-09-161-25/+49
|
* Revert "Revert "Use a LeanHit until merging is done." MERGEOK"Henning Baldersheim2019-09-161-49/+25
|
* Revert "Use a LeanHit until merging is done."Henning Baldersheim2019-09-161-25/+49
|
* Use a LeanHit until merging is done.Henning Baldersheim2019-09-161-49/+25
|
* Since we are trimming we need to reflect that in the offsets too.Henning Baldersheim2019-09-111-0/+1
|
* Check for auxiliary hits in first chunk explicitHenning Baldersheim2019-09-111-1/+10
|
* Add minmal testing of merge and offset and auxiliary hits.Henning Baldersheim2019-09-111-4/+14
|
* Also consider auxiliary hits.Henning Baldersheim2019-09-101-6/+24
|
* HitGroup.sort and hit.compareTo must ensure same ordering...Henning Baldersheim2019-09-101-21/+19
| | | | Do not use the first result as starting point. Start off with a fresh one.
* Revert "Revert "Append all, the sort all, then trim tail is too expensive.""Henning Baldersheim2019-09-101-14/+33
|
* Revert "Append all, the sort all, then trim tail is too expensive."Henning Baldersheim2019-09-101-33/+14
|
* Append all, the sort all, then trim tail is too expensive.Henning Baldersheim2019-09-081-14/+33
| | | | Continous merge with trim is far more efficient as input is already sorted.
* Trim hits more eagerlyOlli Virtanen2019-06-111-1/+1
|
* Use sort data with java dispatcherOlli Virtanen2019-04-021-12/+4
|
* Change trace level of network errorsOlli Virtanen2019-03-221-2/+2
|
* Report partial group connection failures through trace, not errorOlli Virtanen2019-03-221-4/+27
|
* Remove the packet cache. It will very soon loose its value as we are ↵Henning Baldersheim2019-03-071-3/+2
| | | | | | | changing the protocol. And also as it does not work for grouping I think it is better to just let it go. We might want a cache sometime later, but then we should redesign it.
* Fix search invoker resource leakOlli Virtanen2019-02-121-1/+6
|
* Minor typo fixOlli Virtanen2019-01-161-1/+1
|
* Use the correct error type for timeouts, and only create one errorOlli Virtanen2019-01-161-9/+6
|