X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/88a29cd091468feb98e5cd541c560f4d35bca716..1980d00a9afeaea0bc6f266892e43de14ccd297e:/lib/dispatchcloud/node_size.go

diff --git a/lib/dispatchcloud/node_size.go b/lib/dispatchcloud/node_size.go
index 34f83a6efd..1c36d6cf5b 100644
--- a/lib/dispatchcloud/node_size.go
+++ b/lib/dispatchcloud/node_size.go
@@ -5,10 +5,10 @@
 package dispatchcloud
 
 import (
-	"bytes"
 	"errors"
 	"log"
 	"os/exec"
+	"sort"
 	"strings"
 	"time"
 
@@ -16,11 +16,17 @@ import (
 )
 
 var (
-	ErrConstraintsNotSatisfiable  = errors.New("constraints not satisfiable by any configured instance type")
 	ErrInstanceTypesNotConfigured = errors.New("site configuration does not list any instance types")
 	discountConfiguredRAMPercent  = 5
 )
 
+// ConstraintsNotSatisfiableError includes a list of available instance types
+// to be reported back to the user.
+type ConstraintsNotSatisfiableError struct {
+	error
+	AvailableTypes []arvados.InstanceType
+}
+
 // ChooseInstanceType returns the cheapest available
 // arvados.InstanceType big enough to run ctr.
 func ChooseInstanceType(cc *arvados.Cluster, ctr *arvados.Container) (best arvados.InstanceType, err error) {
@@ -41,20 +47,35 @@ func ChooseInstanceType(cc *arvados.Cluster, ctr *arvados.Container) (best arvad
 	needRAM := ctr.RuntimeConstraints.RAM + ctr.RuntimeConstraints.KeepCacheRAM
 	needRAM = (needRAM * 100) / int64(100-discountConfiguredRAMPercent)
 
-	err = ErrConstraintsNotSatisfiable
+	ok := false
 	for _, it := range cc.InstanceTypes {
 		switch {
-		case err == nil && it.Price > best.Price:
-		case it.Scratch < needScratch:
-		case it.RAM < needRAM:
+		case ok && it.Price > best.Price:
+		case int64(it.Scratch) < needScratch:
+		case int64(it.RAM) < needRAM:
 		case it.VCPUs < needVCPUs:
+		case it.Preemptible != ctr.SchedulingParameters.Preemptible:
 		case it.Price == best.Price && (it.RAM < best.RAM || it.VCPUs < best.VCPUs):
 			// Equal price, but worse specs
 		default:
 			// Lower price || (same price && better specs)
 			best = it
-			err = nil
+			ok = true
+		}
+	}
+	if !ok {
+		availableTypes := make([]arvados.InstanceType, 0, len(cc.InstanceTypes))
+		for _, t := range cc.InstanceTypes {
+			availableTypes = append(availableTypes, t)
 		}
+		sort.Slice(availableTypes, func(a, b int) bool {
+			return availableTypes[a].Price < availableTypes[b].Price
+		})
+		err = ConstraintsNotSatisfiableError{
+			errors.New("constraints not satisfiable by any configured instance type"),
+			availableTypes,
+		}
+		return
 	}
 	return
 }
@@ -69,11 +90,9 @@ func ChooseInstanceType(cc *arvados.Cluster, ctr *arvados.Container) (best arvad
 // it is no longer offered by any node. So, to make a feature name
 // valid, we can add it to a dummy node ("compute0"), then remove it.
 //
-// (2) when srun is given an invalid --gres argument and an invalid
-// --constraint argument, the error message mentions "Invalid feature
-// specification". So, to test whether a feature name is valid without
-// actually submitting a job, we can call srun with the feature name
-// and an invalid --gres argument.
+// (2) To test whether a set of feature names are valid without
+// actually submitting a job, we can call srun --test-only with the
+// desired features.
 //
 // SlurmNodeTypeFeatureKludge does a test-and-fix operation
 // immediately, and then periodically, in case slurm restarts and
@@ -90,39 +109,31 @@ func SlurmNodeTypeFeatureKludge(cc *arvados.Cluster) {
 	}
 	for {
 		slurmKludge(features)
-		time.Sleep(time.Minute)
+		time.Sleep(2 * time.Second)
 	}
 }
 
-var (
-	slurmDummyNode     = "compute0"
-	slurmErrBadFeature = "Invalid feature"
-	slurmErrBadGres    = "Invalid generic resource"
-)
+const slurmDummyNode = "compute0"
 
 func slurmKludge(features []string) {
-	cmd := exec.Command("srun", "--gres=invalid-gres-specification", "--constraint="+strings.Join(features, "&"), "true")
-	out, err := cmd.CombinedOutput()
-	switch {
-	case err == nil:
-		log.Printf("warning: guaranteed-to-fail srun command did not fail: %q %q", cmd.Path, cmd.Args)
-		log.Printf("output was: %q", out)
-
-	case bytes.Contains(out, []byte(slurmErrBadFeature)):
-		log.Printf("temporarily configuring node %q with all node type features", slurmDummyNode)
-		for _, nodeFeatures := range []string{strings.Join(features, ","), ""} {
-			cmd = exec.Command("scontrol", "update", "NodeName="+slurmDummyNode, "Features="+nodeFeatures)
-			log.Printf("running: %q %q", cmd.Path, cmd.Args)
-			out, err := cmd.CombinedOutput()
-			if err != nil {
-				log.Printf("error: scontrol: %s (output was %q)", err, out)
-			}
-		}
+	allFeatures := strings.Join(features, ",")
 
-	case bytes.Contains(out, []byte(slurmErrBadGres)):
-		// Evidently our node-type feature names are all valid.
+	cmd := exec.Command("sinfo", "--nodes="+slurmDummyNode, "--format=%f", "--noheader")
+	out, err := cmd.CombinedOutput()
+	if err != nil {
+		log.Printf("running %q %q: %s (output was %q)", cmd.Path, cmd.Args, err, out)
+		return
+	}
+	if string(out) == allFeatures+"\n" {
+		// Already configured correctly, nothing to do.
+		return
+	}
 
-	default:
-		log.Printf("warning: expected srun error %q or %q, but output was %q", slurmErrBadFeature, slurmErrBadGres, out)
+	log.Printf("configuring node %q with all node type features", slurmDummyNode)
+	cmd = exec.Command("scontrol", "update", "NodeName="+slurmDummyNode, "Features="+allFeatures)
+	log.Printf("running: %q %q", cmd.Path, cmd.Args)
+	out, err = cmd.CombinedOutput()
+	if err != nil {
+		log.Printf("error: scontrol: %s (output was %q)", err, out)
 	}
 }