aboutsummaryrefslogtreecommitdiffstats
path: root/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/autoscale
Commit message (Collapse)AuthorAgeFilesLines
* memoryGb => memoryGiBHenning Baldersheim2024-06-113-12/+11
|
* Merge pull request #31446 from vespa-engine/bratseth/container-ideal-memoryBjørn Meland2024-06-051-1/+1
|\ | | | | Increase container ideal memory load to 0.9
| * Increase container ideal memory load to 0.9Jon Bratseth2024-06-051-1/+1
| | | | | | | | | | | | | | Vespa will now attempt to allocate memory to this level. If we don't allow it, all allocations will be unfulfiled, since adding more resources won't decrease used memory per container.
* | Remove printlnsJon Bratseth2024-06-051-2/+0
| |
* | Test that we don't rescale when container memory is unfulfiledJon Bratseth2024-06-052-0/+19
|/
* Avoid injecting ExclusivityHåkon Hallingstad2024-05-191-3/+6
|
* Remove flags dependency from config-modelHåkon Hallingstad2024-05-182-14/+9
|
* Move CapacityPolicies to config-privisioningJon Bratseth2024-05-121-1/+1
|
* Make CapacityPolicies independent of NodeRepositoryJon Bratseth2024-05-121-1/+4
|
* Lower ideal query cpuJon Bratseth2024-04-243-59/+59
| | | | 80% has observed to cause problems, so lowering to 75%.
* Add host-memory-ratio flag and add flavor dimension to host-memoryHåkon Hallingstad2024-04-221-1/+1
|
* Avoid Optional with CloudAccountHåkon Hallingstad2024-04-183-8/+6
|
* Propagate cloud account to ResourcesCalculatorHåkon Hallingstad2024-04-173-5/+9
|
* Store multiple resource suggestionsOla Aunronning2024-01-122-4/+4
|
* Store GPU metrics and load separatelyMartin Polden2023-12-205-31/+35
|
* Revert "Expose GPU load in nodes API"Harald Musum2023-12-195-35/+31
|
* Store GPU metrics and load separatelyMartin Polden2023-12-195-31/+35
|
* Remove unused parametersMartin Polden2023-12-081-1/+0
|
* Let distributor metric decide cluster stabilityMartin Polden2023-12-061-1/+20
|
* Revert "Evaluate flags referenced multiple times at the start of prepare"Håkon Hallingstad2023-11-073-10/+6
|
* Include NodeRepository, ApplcationId, and ClusterSpec as parametersHåkon Hallingstad2023-11-053-10/+7
|
* Rename to AllocationParamsHåkon Hallingstad2023-11-052-5/+5
|
* Rename to ClusterAllocationParamsHåkon Hallingstad2023-11-043-9/+9
|
* Evaluate flags referenced multiple times at the start of prepareHåkon Hallingstad2023-11-043-4/+11
|
* Update ↵Bjørn Meland2023-11-031-1/+1
| | | | | node-repository/src/test/java/com/yahoo/vespa/hosted/provision/autoscale/AutoscalingTest.java Co-authored-by: Jon Bratseth <bratseth@gmail.com>
* Clarification of messagebjormel2023-11-021-1/+1
|
* Don't autoscale single node clustersbjormel2023-11-021-3/+1
|
* Update copyrightJon Bratseth2023-10-0915-13/+15
|
* Propagate GPU resources when determining current resources for autoscalingValerij Fredriksen2023-09-281-0/+19
|
* Update AwsNodeTypes with GPU instancesValerij Fredriksen2023-09-281-11/+62
|
* Compare advertised resources in flavor satisified check in testsValerij Fredriksen2023-09-283-37/+51
|
* Use at least 8GB memory nodes in unit testsValerij Fredriksen2023-09-081-3/+1
|
* Adjust by idealJon Bratseth2023-08-161-19/+19
|
* Don't use BCPInfo to scale downJon Bratseth2023-08-161-1/+16
|
* More accurate accounting of redistribution costJon Bratseth2023-08-161-30/+1
|
* Test account for change costJon Bratseth2023-08-152-31/+97
|
* AllocatableClusterresources -> AllocatableResourcesJon Bratseth2023-08-152-9/+5
|
* No functional changes only: Wrap current resources in modelJon Bratseth2023-08-152-13/+16
|
* Consider switching cost when choosing resourcesJon Bratseth2023-08-151-1/+1
|
* Leave 15% headroom when scaling downJon Bratseth2023-07-312-6/+6
|
* Set aside enough headroom rather than not scaling down at allJon Bratseth2023-07-252-6/+6
|
* Set aside space for headroom in one passJon Bratseth2023-07-251-3/+3
|
* Don't scale down if we are likely to scale back upJon Bratseth2023-07-203-10/+41
| | | | | | | | | When we decide to scale number of nodes down we'll see a lower ideal load because we need to be able to handle one node going down. This may lead us to be closer to ideal (at current peak load) than otherwise anticipated, such that we are quite likely to soon scale back up. This checks for that and avoids scaling down dimensions where this is the case.
* Merge pull request #27787 from vespa-engine/bratseth/no-scalingDuration-minJon Bratseth2023-07-142-29/+49
|\ | | | | Remove scaling duration min for content clusters
| * Remove scaling duration min for content clustersJon Bratseth2023-07-142-29/+49
| | | | | | | | This has been a safety measure, which doesn't seem necessary any more.
* | Deduce tight group limits when using group-sizeJon Bratseth2023-07-131-1/+1
|/
* Don't pick resources which are against recommendationsJon Bratseth2023-07-071-22/+66
|
* Test warningJon Bratseth2023-07-051-1/+12
|
* Reduce minimum content scaling duration from 8 to 6hJon Bratseth2023-06-132-7/+6
|
* Reduce min content scaling duration frpm 12 to 8Jon Bratseth2023-05-242-10/+10
|