1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
15 "git.curoverse.com/arvados.git/sdk/go/arvados"
19 ErrInstanceTypesNotConfigured = errors.New("site configuration does not list any instance types")
20 discountConfiguredRAMPercent = 5
23 // ConstraintsNotSatisfiableError includes a list of available instance types
24 // to be reported back to the user.
25 type ConstraintsNotSatisfiableError struct {
27 AvailableTypes []arvados.InstanceType
30 // ChooseInstanceType returns the cheapest available
31 // arvados.InstanceType big enough to run ctr.
32 func ChooseInstanceType(cc *arvados.Cluster, ctr *arvados.Container) (best arvados.InstanceType, err error) {
33 if len(cc.InstanceTypes) == 0 {
34 err = ErrInstanceTypesNotConfigured
38 needScratch := int64(0)
39 for _, m := range ctr.Mounts {
41 needScratch += m.Capacity
45 needVCPUs := ctr.RuntimeConstraints.VCPUs
47 needRAM := ctr.RuntimeConstraints.RAM + ctr.RuntimeConstraints.KeepCacheRAM
48 needRAM = (needRAM * 100) / int64(100-discountConfiguredRAMPercent)
50 availableTypes := make([]arvados.InstanceType, len(cc.InstanceTypes))
51 copy(availableTypes, cc.InstanceTypes)
52 sort.Slice(availableTypes, func(a, b int) bool {
53 return availableTypes[a].Price < availableTypes[b].Price
55 err = ConstraintsNotSatisfiableError{
56 errors.New("constraints not satisfiable by any configured instance type"),
59 for _, it := range cc.InstanceTypes {
61 case err == nil && it.Price > best.Price:
62 case it.Scratch < needScratch:
63 case it.RAM < needRAM:
64 case it.VCPUs < needVCPUs:
65 case it.Preemptable != ctr.SchedulingParameters.Preemptable:
66 case it.Price == best.Price && (it.RAM < best.RAM || it.VCPUs < best.VCPUs):
67 // Equal price, but worse specs
69 // Lower price || (same price && better specs)
77 // SlurmNodeTypeFeatureKludge ensures SLURM accepts every instance
78 // type name as a valid feature name, even if no instances of that
79 // type have appeared yet.
81 // It takes advantage of some SLURM peculiarities:
83 // (1) A feature is valid after it has been offered by a node, even if
84 // it is no longer offered by any node. So, to make a feature name
85 // valid, we can add it to a dummy node ("compute0"), then remove it.
87 // (2) To test whether a set of feature names are valid without
88 // actually submitting a job, we can call srun --test-only with the
91 // SlurmNodeTypeFeatureKludge does a test-and-fix operation
92 // immediately, and then periodically, in case slurm restarts and
93 // forgets the list of valid features. It never returns (unless there
94 // are no node types configured, in which case it returns
95 // immediately), so it should generally be invoked with "go".
96 func SlurmNodeTypeFeatureKludge(cc *arvados.Cluster) {
97 if len(cc.InstanceTypes) == 0 {
100 var features []string
101 for _, it := range cc.InstanceTypes {
102 features = append(features, "instancetype="+it.Name)
105 slurmKludge(features)
106 time.Sleep(2 * time.Second)
110 const slurmDummyNode = "compute0"
112 func slurmKludge(features []string) {
113 allFeatures := strings.Join(features, ",")
115 cmd := exec.Command("sinfo", "--nodes="+slurmDummyNode, "--format=%f", "--noheader")
116 out, err := cmd.CombinedOutput()
118 log.Printf("running %q %q: %s (output was %q)", cmd.Path, cmd.Args, err, out)
121 if string(out) == allFeatures+"\n" {
122 // Already configured correctly, nothing to do.
126 log.Printf("configuring node %q with all node type features", slurmDummyNode)
127 cmd = exec.Command("scontrol", "update", "NodeName="+slurmDummyNode, "Features="+allFeatures)
128 log.Printf("running: %q %q", cmd.Path, cmd.Args)
129 out, err = cmd.CombinedOutput()
131 log.Printf("error: scontrol: %s (output was %q)", err, out)