Merge branch 'master' into scheduler_config_updae

client/python/pyproject.toml

@@ -9,7 +9,7 @@ license = { text = "Apache Software License" }
 authors = [{ name = "G-Research Open Source Software", email = "armada@armadaproject.io" }]
 
 [project.optional-dependencies]
-format = ["black==23.3.0", "flake8==6.0.0", "pylint==2.17.3"]
+format = ["black==23.3.0", "flake8==6.0.0", "pylint==2.17.4"]
 docs = ["sphinx", "sphinx-jekyll-builder", "sphinx-toolbox==3.2.0b1"]
 test = ["pytest==7.3.1", "coverage>=6.5.0", "pytest-asyncio==0.21.0"]
 

cmd/armada-load-tester/cmd/root.go

@@ -21,15 +21,7 @@ var rootCmd = &cobra.Command{
 Command line utility to submit many jobs to armada
 
 Persistent config can be saved in a config file so it doesn't have to be specified every command.
-
-Example structure:
-armadaUrl: localhost:50051
-basicAuth:
-  username: user1
-  password: password123
-
-The location of this file can be passed in using --config argument or picked from $HOME/.armadactl.yaml.
-`,
+The location of this file can be passed in using --config argument or picked from $HOME/.armadactl.yaml.`,
 }
 
 // Execute adds all child commands to the root command and sets flags appropriately.

cmd/testsuite/cmd/root.go

@@ -18,13 +18,6 @@ func RootCmd() *cobra.Command {
 		Long: `testsuite is a suite of automated tests for Armada deployments.
 
 Persistent config can be saved in a config file so it doesn't have to be specified every command.
-
-Example structure:
-armadaUrl: localhost:50051
-basicAuth:
-username: user1
-password: password123
-
 The location of this file can be passed in using the --config argument.
 If not provided, $HOME/.armadactl.yaml is used.`,
 	}

internal/armada/configuration/types.go

@@ -240,6 +240,9 @@ type PriorityClass struct {
 	// jobs of this priority class are not scheduled if doing so would cause the total resources assigned
 	// to jobs of priority 10 or lower from the same queue to exceed 30% of the total.
 	MaximumResourceFractionPerQueue map[string]float64
+	// Per-pool override of MaximumResourceFractionPerQueue.
+	// If missing for a particular pool, MaximumResourceFractionPerQueue is used instead for that pool.
+	MaximumResourceFractionPerQueueByPool map[string]map[string]float64
 }
 
 func (p PreemptionConfig) PriorityByPriorityClassName() map[string]int32 {

internal/armada/server/lease.go

@@ -267,7 +267,9 @@ func (q *AggregatedQueueServer) getJobs(ctx context.Context, req *api.StreamingL
 	activeClusterReports := scheduling.FilterActiveClusters(usageReports)
 	totalCapacity := make(armadaresource.ComputeResources)
 	for _, clusterReport := range activeClusterReports {
-		totalCapacity.Add(util.GetClusterAvailableCapacity(clusterReport))
+		if clusterReport.Pool == req.Pool {
+			totalCapacity.Add(util.GetClusterAvailableCapacity(clusterReport))
+		}
 	}
 
 	// Collect all allowed priorities.

internal/armada/server/submit_test.go

@@ -1637,7 +1637,7 @@ func withSubmitServerAndRepos(action func(s *SubmitServer, jobRepo repository.Jo
 		MaxPodSpecSizeBytes: 65535,
 		Preemption: configuration.PreemptionConfig{
 			DefaultPriorityClass: "high",
-			PriorityClasses:      map[string]configuration.PriorityClass{"high": {0, false, nil}},
+			PriorityClasses:      map[string]configuration.PriorityClass{"high": {0, false, nil, nil}},
 		},
 		MinTerminationGracePeriod: time.Duration(30 * time.Second),
 		MaxTerminationGracePeriod: time.Duration(300 * time.Second),

internal/scheduler/adapters/adapters_test.go

@@ -18,10 +18,10 @@ import (
 
 var (
 	priorityByPriorityClassName = map[string]configuration.PriorityClass{
-		"priority-0": {0, true, nil},
-		"priority-1": {1, true, nil},
-		"priority-2": {2, true, nil},
-		"priority-3": {3, false, nil},
+		"priority-0": {0, true, nil, nil},
+		"priority-1": {1, true, nil, nil},
+		"priority-2": {2, true, nil, nil},
+		"priority-3": {3, false, nil, nil},
 	}
 
 	priority int32 = 1

internal/scheduler/constraints/constraints.go

@@ -68,15 +68,21 @@ func SchedulingConstraintsFromSchedulingConfig(
 ) SchedulingConstraints {
 	priorityClassSchedulingConstraintsByPriorityClassName := make(map[string]PriorityClassSchedulingConstraints, len(config.Preemption.PriorityClasses))
 	for name, priorityClass := range config.Preemption.PriorityClasses {
+		maximumResourceFractionPerQueue := priorityClass.MaximumResourceFractionPerQueue
+		if m, ok := priorityClass.MaximumResourceFractionPerQueueByPool[pool]; ok {
+			// Use pool-specific config is available.
+			maximumResourceFractionPerQueue = m
+		}
 		priorityClassSchedulingConstraintsByPriorityClassName[name] = PriorityClassSchedulingConstraints{
 			PriorityClassName:                  name,
 			PriorityClassPriority:              priorityClass.Priority,
-			MaximumCumulativeResourcesPerQueue: absoluteFromRelativeLimits(totalResources, priorityClass.MaximumResourceFractionPerQueue),
+			MaximumCumulativeResourcesPerQueue: absoluteFromRelativeLimits(totalResources, maximumResourceFractionPerQueue),
 		}
 	}
 	maximumResourceFractionToSchedule := config.MaximumResourceFractionToSchedule
-	if limit, ok := config.MaximumResourceFractionToScheduleByPool[pool]; ok {
-		maximumResourceFractionToSchedule = limit
+	if m, ok := config.MaximumResourceFractionToScheduleByPool[pool]; ok {
+		// Use pool-specific config is available.
+		maximumResourceFractionToSchedule = m
 	}
 	return SchedulingConstraints{
 		MaximumJobsToSchedule:      config.MaximumJobsToSchedule,