X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/e8f99cfef7cfbfcf1a1485d69250f24ced3fd609..94942f7b2f35a775aea5b22d2be637022e6b4fb7:/lib/dispatchcloud/node_size.go

diff --git a/lib/dispatchcloud/node_size.go b/lib/dispatchcloud/node_size.go
index 2fac799f40..7c7643bfc7 100644
--- a/lib/dispatchcloud/node_size.go
+++ b/lib/dispatchcloud/node_size.go
@@ -6,12 +6,11 @@ package dispatchcloud
 
 import (
 	"errors"
-	"log"
 	"regexp"
 	"sort"
 	"strconv"
 
-	"git.curoverse.com/arvados.git/sdk/go/arvados"
+	"git.arvados.org/arvados.git/sdk/go/arvados"
 )
 
 var ErrInstanceTypesNotConfigured = errors.New("site configuration does not list any instance types")
@@ -34,12 +33,10 @@ var pdhRegexp = regexp.MustCompile(`^[0-9a-f]{32}\+(\d+)$`)
 func estimateDockerImageSize(collectionPDH string) int64 {
 	m := pdhRegexp.FindStringSubmatch(collectionPDH)
 	if m == nil {
-		log.Printf("estimateDockerImageSize: '%v' did not match pdhRegexp, returning 0", collectionPDH)
 		return 0
 	}
 	n, err := strconv.ParseInt(m[1], 10, 64)
 	if err != nil || n < 122 {
-		log.Printf("estimateDockerImageSize: short manifest %v or error (%v), returning 0", n, err)
 		return 0
 	}
 	// To avoid having to fetch the collection, take advantage of
@@ -49,7 +46,7 @@ func estimateDockerImageSize(collectionPDH string) int64 {
 	// the size of the manifest.
 	//
 	// Use the following heuristic:
-	// - Start with the length of the mainfest (n)
+	// - Start with the length of the manifest (n)
 	// - Subtract 80 characters for the filename and file segment
 	// - Divide by 42 to get the number of block identifiers ('hash\+size\ ' is 32+1+8+1)
 	// - Assume each block is full, multiply by 64 MiB
@@ -86,6 +83,19 @@ func EstimateScratchSpace(ctr *arvados.Container) (needScratch int64) {
 	return
 }
 
+// compareVersion returns true if vs1 < vs2, otherwise false
+func versionLess(vs1 string, vs2 string) (bool, error) {
+	v1, err := strconv.ParseFloat(vs1, 64)
+	if err != nil {
+		return false, err
+	}
+	v2, err := strconv.ParseFloat(vs2, 64)
+	if err != nil {
+		return false, err
+	}
+	return v1 < v2, nil
+}
+
 // 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) {
@@ -99,19 +109,29 @@ func ChooseInstanceType(cc *arvados.Cluster, ctr *arvados.Container) (best arvad
 	needVCPUs := ctr.RuntimeConstraints.VCPUs
 
 	needRAM := ctr.RuntimeConstraints.RAM + ctr.RuntimeConstraints.KeepCacheRAM
+	needRAM += int64(cc.Containers.ReserveExtraRAM)
+	needRAM += int64(cc.Containers.LocalKeepBlobBuffersPerVCPU * needVCPUs * (1 << 26))
 	needRAM = (needRAM * 100) / int64(100-discountConfiguredRAMPercent)
 
 	ok := false
 	for _, it := range cc.InstanceTypes {
+		driverInsuff, driverErr := versionLess(it.CUDA.DriverVersion, ctr.RuntimeConstraints.CUDA.DriverVersion)
+		capabilityInsuff, capabilityErr := versionLess(it.CUDA.HardwareCapability, ctr.RuntimeConstraints.CUDA.HardwareCapability)
+
 		switch {
-		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
+		// reasons to reject a node
+		case ok && it.Price > best.Price: // already selected a node, and this one is more expensive
+		case int64(it.Scratch) < needScratch: // insufficient scratch
+		case int64(it.RAM) < needRAM: // insufficient RAM
+		case it.VCPUs < needVCPUs: // insufficient VCPUs
+		case it.Preemptible != ctr.SchedulingParameters.Preemptible: // wrong preemptable setting
+		case it.Price == best.Price && (it.RAM < best.RAM || it.VCPUs < best.VCPUs): // same price, worse specs
+		case it.CUDA.DeviceCount < ctr.RuntimeConstraints.CUDA.DeviceCount: // insufficient CUDA devices
+		case ctr.RuntimeConstraints.CUDA.DeviceCount > 0 && (driverInsuff || driverErr != nil): // insufficient driver version
+		case ctr.RuntimeConstraints.CUDA.DeviceCount > 0 && (capabilityInsuff || capabilityErr != nil): // insufficient hardware capability
+			// Don't select this node
 		default:
+			// Didn't reject the node, so select it
 			// Lower price || (same price && better specs)
 			best = it
 			ok = true