summaryrefslogtreecommitdiffstats
path: root/container-search
Commit message (Collapse)AuthorAgeFilesLines
* Support cloning of primitive arraysJon Bratseth2017-04-271-2/+2
|
* Override cloneJon Bratseth2017-04-271-2/+11
|
* Don't fall throughJon Bratseth2017-04-261-5/+3
|
* Correct heuristic for urlsJon Bratseth2017-04-261-13/+9
|
* explicit lint optionsArne H Juul2017-04-261-0/+12
| | | | | * this module can't really benefit from linting currently, but make it explicit for clarity.
* remove unneeded castsArne H Juul2017-04-262-6/+6
|
* suppress fallthrough warningsArne H Juul2017-04-262-1/+7
| | | | * add comments where they occurred, somebody should look at that
* Merge pull request #2210 from yahoo/arnej/silence-warnings-2Arne H Juul2017-04-214-0/+11
|\ | | | | Arnej/silence warnings 2
| * add @OverrideArne H Juul2017-04-211-0/+1
| |
| * use Objects.hash() to implement hashCode()Arne H Juul2017-04-211-8/+4
| |
| * fix fallthrough warningArne H Juul2017-04-201-0/+1
| |
| * implement hashCode to fix warningArne H Juul2017-04-201-0/+11
| |
| * fix warningArne H Juul2017-04-201-0/+1
| |
| * fix fallthrough warningArne H Juul2017-04-201-0/+1
| |
* | Don't depend on wallclock timeJon Bratseth2017-04-201-13/+7
|/
* Sample max QPS every secondBjorn Meland2017-03-311-2/+2
|
* Increase timeout because factory is so slowJon Bratseth2017-03-271-11/+11
|
* Don't call Query.setTimeout() with negative valueBjørn Christian Seime2017-03-161-14/+30
|
* Merge pull request #2012 from ↵Bjørn Christian Seime2017-03-153-94/+68
|\ | | | | | | | | yahoo/bjorncs/fix-trace-level-bug-streaming-searcher Bjorncs/fix trace level bug streaming searcher
| * Modify test to explicitly test behaviour for tracelevel=0Bjørn Christian Seime2017-03-151-33/+4
| |
| * Only render trace when tracelevel is higher than 0Bjørn Christian Seime2017-03-152-73/+78
| |
| * Revert "Don't add statistics property as trace level 0"Bjørn Christian Seime2017-03-031-11/+9
| | | | | | | | This reverts commit 640880aad9d5585c9d079717a1ae5ce03993fcce.
* | raw user input will be an forwarded as exact match.Henning Baldersheim2017-03-067-16/+48
| |
* | Shutdown renderer threads on deconstructJon Bratseth2017-03-062-6/+24
|/
* Revert "Store streaming stats in query instead of query context"Bjørn Christian Seime2017-03-023-11/+17
|
* Store streaming stats in query instead of query contextBjørn Christian Seime2017-03-023-17/+11
| | | | | | All properties in the query context are modelled as trace nodes and will be part of the query trace regardless of level. The streaming statistics should only be visible in trace for levels >=2.
* Revert "Store streaming stats in query instead of query context"Arnstein Ressem2017-03-013-10/+14
|
* Store streaming stats in query instead of query contextBjørn Christian Seime2017-03-013-14/+10
| | | | | | All properties in the query context are modelled as trace nodes and will be part of the query trace regardless of level. The streaming statistics should only be visible in trace for levels >=2.
* revert qrserver part of docsum protocol changesArne H Juul2017-02-284-14/+16
| | | | | | | * when the qrserver was upgraded first, it would expect the backend to send schemaless docsums without asking for it, leading to lots of warnings and performance problems. We'll need to do this protocol change in two steps.
* small fixes after reviewArne H Juul2017-02-231-1/+3
|
* remove setting of "allow slime" flagArne H Juul2017-02-234-16/+12
|
* ask for and expect slime docsumsArne H Juul2017-02-212-0/+2
|
* Increase timeout in case of slow executionJon Bratseth2017-02-201-2/+2
|
* Don't add statistics property as trace level 0Bjørn Christian Seime2017-02-151-9/+11
| | | | | Fix bug where statistics are always visible in trace output, independent of trace level. Also remove duplicate visitor statistics in trace.
* Add some more debug information to figure out what we actually send down in ↵Henning Baldersheim2017-02-152-4/+3
| | | | fill.
* Use direct dispatch for backend too.Henning Baldersheim2017-02-141-1/+1
|
* Remove hits we could not fill in timeJon Bratseth2017-02-144-15/+65
|
* Update with class comments.Henning Baldersheim2017-02-131-0/+3
|
* Add back some public static methods to avoid api breakage, however unlikely.Henning Baldersheim2017-02-131-2/+8
|
* Fix java doc warnings.Henning Baldersheim2017-02-131-2/+2
|
* When code is duplicated is will only be fixed in one of the places.Henning Baldersheim2017-02-132-57/+20
| | | | | At least now the cloning magic is only present in one place. Shall we deprecate/deduplicate any of this processing/search code ?
* Revert "ask for and expect slime docsums"Arne H Juul2017-02-132-2/+0
|
* ask for and expect slime docsumsArne H Juul2017-02-132-0/+2
|
* Merge pull request #1708 from yahoo/arnej/followup-reviewJon Bratseth2017-02-131-3/+2
|\ | | | | use documentation comment for expected invariant
| * use documentation comment for expected invariantArne H Juul2017-02-071-3/+2
| | | | | | | | * use "assert" only for internally-guaranteed logic
* | Merge branch 'master' into bratseth/add-filter-testJon Bratseth2017-02-0812-53/+320
|\|
| * handle duplicate addArne H Juul2017-02-071-22/+27
| | | | | | | | | | | | | | * handle adding the same node to Hasher multiple times * handle removing the same node from Hasher multiple times * simplify select() logic to be more readable * sprinkle some asserts to ensure consistency
| * changing initial state did not workArne H Juul2017-02-072-4/+5
| | | | | | | | | | | | | | | | * turns out ClusterSearcher depends on initial state "working", so it got duplicates in the Hasher when transitioning from initial to working, then got null pointers when transitioning to down. Go back to initial isWorking=true. * When a backend fails, also clear its searchNodesOnline flag.
| * Merge pull request #1686 from yahoo/arnej/assume-node-not-workingJon Bratseth2017-02-062-16/+12
| |\ | | | | | | assume node down until proven otherwise
| | * assume node down until proven otherwiseArne H Juul2017-02-062-16/+12
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | * change initializer values in NodeMonitor to assume the node is down and has no search nodes online until we see some data from it. Also, update vip status as soon as we have some nodes attached. This is to avoid the container believing it should be in service at startup and then changing its mind when it discovers that underlying services are down.