summaryrefslogtreecommitdiffstats
path: root/container-search/src/main/java/com/yahoo/search/dispatch
Commit message (Collapse)AuthorAgeFilesLines
* Wait for dispatch nodes config in ReconfigurableDispatcher constructorjonmv2023-08-161-7/+11
|
* Use correct config id for dispatch configjonmv2023-08-161-2/+9
|
* Set up reconfigurable dispatcher with self-subscription insteadjonmv2023-07-141-0/+37
|
* Test delay shutdown of RPC resources in dispatcherjonmv2023-07-133-3/+13
|
* Get info about all nodes before updating search groups referencejonmv2023-07-132-13/+28
|
* Set up constructor wiring for testing rpc connection pool updatejonmv2023-07-134-22/+55
|
* Test nodes are retained when updating search cluster groupsjonmv2023-07-131-3/+4
|
* Set up skeleton for replacing monitored groups and nodesjonmv2023-07-136-50/+77
|
* Add skeleton for delaying search invoker RPC resource shutdownjonmv2023-07-132-12/+74
|
* Simplify a bitjonmv2023-07-134-30/+22
|
* Avoid overflow for 146 years of JVM time for common timersjonmv2023-07-071-10/+4
|
* Another potential overflow (not really, since trunc milis in a long, but ↵jonmv2023-07-071-1/+1
| | | | that could change)
* Avoid overflow for warmup end time computationjonmv2023-07-073-9/+11
|
* Follow API changejonmv2023-07-061-1/+1
|
* move MatchFeatureData class to vespajlibArne Juul2023-04-272-94/+1
|
* Merge pull request #26641 from vespa-engine/balder/control-slime-decodingHenning Baldersheim2023-03-302-3/+21
|\ | | | | Propagate the configured slime decode type into the rpc summary handl…
| * Use the summary decode policy to select Inspector implementations.Henning Baldersheim2023-03-301-1/+6
| |
| * Propagate the configured slime decode type into the rpc summary handling.Henning Baldersheim2023-03-302-2/+15
| |
* | - Use CompoundName.fromHenning Baldersheim2023-03-291-1/+1
| | | | | | | | - Deprecate unused SubProperties
* | Use CompoundName.from in tests and construction of static objects.Henning Baldersheim2023-03-291-1/+1
|/
* Add query trace parameters for profiling backend query evaluation.Geir Storli2023-01-201-1/+22
| | | | | With this change, profiling of matching, first-phase ranking, and second-phase ranking can be tuned separately.
* Don't warn when 1 node is downJon Bratseth2022-12-291-7/+15
|
* Revert "Revert collect(Collectors.toList())"Henning Baldersheim2022-12-042-2/+2
|
* Revert collect(Collectors.toList())Henning Baldersheim2022-12-042-2/+2
|
* collect(Collectors.toList()) -> toList()Henning Baldersheim2022-12-022-2/+2
|
* Need a delayed close for the connections no longer needed.Henning Baldersheim2022-11-261-6/+21
|
* Let RpcResourcePool handle live changes.Henning Baldersheim2022-11-253-16/+37
|
* pathIndex should not be part of equals and hashcode.Henning Baldersheim2022-11-254-16/+9
|
* Using volatiles directly is fine for ensuring thread visibility.Henning Baldersheim2022-11-252-41/+41
| | | | No need to use Atomic when there are no update races.
* Cleanup and throw IllegalArgumentException on invalid pathJon Bratseth2022-11-253-57/+67
|
* Move test utility code to test directory.Henning Baldersheim2022-11-241-24/+0
|
* Prevent division by zeroHenning Baldersheim2022-11-241-1/+2
|
* Prefer Java over com.google.common.collectHenning Baldersheim2022-11-241-5/+5
|
* Put loadbalancer and invokerfactory in a volatile object to ensure atomic ↵Henning Baldersheim2022-11-249-165/+155
| | | | switch when reconfiguring.
* Cleanup the concept of orderedGroups. Just use a single way of accessing the ↵Henning Baldersheim2022-11-247-111/+152
| | | | | | groups. Simplify testing by introducing a GroupList to contain all acces to groups and nodes.
* Ensure you use the same invoker and loadbalancer for the same call.Henning Baldersheim2022-11-232-7/+7
|
* No need to require the full SearchCluster when you only need a list of Groups.Henning Baldersheim2022-11-232-9/+7
|
* Put the RpcClient back in the RpcResourcePool where it belongs.Henning Baldersheim2022-11-232-28/+25
| | | | Remove parts of schema test no longer valid.
* Make a few simpler interfaces instead of carrying one huge implement all ↵Henning Baldersheim2022-11-2313-110/+148
| | | | SearchCluster around.
* Only have the DispatchNodesConfig inject into one component.Henning Baldersheim2022-11-233-30/+53
| | | | | Let RpcResourcePool/RpcClient be owned by the dispatcher. Step 2 in preparing for smooth handling of content cluster changes.
* Use thread local random.Henning Baldersheim2022-11-221-4/+3
|
* GC unused imports.Henning Baldersheim2022-11-221-6/+0
|
* Splitt nodes static and dynamic config in DispatchConfig to prepare for ↵Henning Baldersheim2022-11-213-8/+13
| | | | seamless cluster changes.
* GC unusedHenning Baldersheim2022-11-211-2/+0
| | | | | | | -useMultilevelDispatch -useLocalNode -searchableCopies -mergeGroupingResultInSearchInvokerEnabled
* GC metric that has not had any meaning since we removed fdispatch.Henning Baldersheim2022-11-081-20/+5
|
* Various cleanup and code deduplication.Henning Baldersheim2022-10-132-9/+9
|
* - Use a common scratchpad for serializing the different parts of the query.Henning Baldersheim2022-10-122-29/+33
| | | | | | - Use a threadlocal for the scratchpad. This avoids costly resizing, or initialiing too large buffer for every query. Using a thread local is fine now that we limit the number of search threads to a reasonable number = #cores * 2.
* Allocate the optional buffer lazilyHenning Baldersheim2022-10-121-2/+8
|
* Reuse scratch buffer to avoid allocating and clearing a new buffer for ↵Henning Baldersheim2022-10-121-4/+8
| | | | serializing every tensor.
* - Presize array to its max size.Henning Baldersheim2022-10-121-3/+4
| | | | | - Return the GrowableByteBuffer from TypedBinaryFormat instead of a copy of the buffer. Then buffer is copied only once instead of twice.