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 ErrConstraintsNotSatisfiable = errors.New("constraints not satisfiable by any configured instance type")
20 ErrInstanceTypesNotConfigured = errors.New("site configuration does not list any instance types")
21 discountConfiguredRAMPercent = 5
24 // ChooseInstanceType returns the cheapest available
25 // arvados.InstanceType big enough to run ctr.
26 func ChooseInstanceType(cc *arvados.Cluster, ctr *arvados.Container) (best arvados.InstanceType, err error) {
27 if len(cc.InstanceTypes) == 0 {
28 err = ErrInstanceTypesNotConfigured
32 needScratch := int64(0)
33 for _, m := range ctr.Mounts {
35 needScratch += m.Capacity
39 needVCPUs := ctr.RuntimeConstraints.VCPUs
41 needRAM := ctr.RuntimeConstraints.RAM + ctr.RuntimeConstraints.KeepCacheRAM
42 needRAM = (needRAM * 100) / int64(100-discountConfiguredRAMPercent)
44 err = ErrConstraintsNotSatisfiable
45 for _, it := range cc.InstanceTypes {
47 case err == nil && it.Price > best.Price:
48 case it.Scratch < needScratch:
49 case it.RAM < needRAM:
50 case it.VCPUs < needVCPUs:
51 case it.Price == best.Price && (it.RAM < best.RAM || it.VCPUs < best.VCPUs):
52 // Equal price, but worse specs
54 // Lower price || (same price && better specs)
62 // SlurmNodeTypeFeatureKludge ensures SLURM accepts every instance
63 // type name as a valid feature name, even if no instances of that
64 // type have appeared yet.
66 // It takes advantage of some SLURM peculiarities:
68 // (1) A feature is valid after it has been offered by a node, even if
69 // it is no longer offered by any node. So, to make a feature name
70 // valid, we can add it to a dummy node ("compute0"), then remove it.
72 // (2) when srun is given an invalid --gres argument and an invalid
73 // --constraint argument, the error message mentions "Invalid feature
74 // specification". So, to test whether a feature name is valid without
75 // actually submitting a job, we can call srun with the feature name
76 // and an invalid --gres argument.
78 // SlurmNodeTypeFeatureKludge does a test-and-fix operation
79 // immediately, and then periodically, in case slurm restarts and
80 // forgets the list of valid features. It never returns (unless there
81 // are no node types configured, in which case it returns
82 // immediately), so it should generally be invoked with "go".
83 func SlurmNodeTypeFeatureKludge(cc *arvados.Cluster) {
84 if len(cc.InstanceTypes) == 0 {
88 for _, it := range cc.InstanceTypes {
89 features = append(features, "instancetype="+it.Name)
93 time.Sleep(time.Minute)
98 slurmDummyNode = "compute0"
99 slurmErrBadFeature = "Invalid feature"
100 slurmErrBadGres = "Invalid generic resource"
103 func slurmKludge(features []string) {
104 cmd := exec.Command("srun", "--gres=invalid-gres-specification", "--constraint="+strings.Join(features, "&"), "true")
105 out, err := cmd.CombinedOutput()
108 log.Printf("warning: guaranteed-to-fail srun command did not fail: %q %q", cmd.Path, cmd.Args)
109 log.Printf("output was: %q", out)
111 case bytes.Contains(out, []byte(slurmErrBadFeature)):
112 log.Printf("temporarily configuring node %q with all node type features", slurmDummyNode)
113 for _, nodeFeatures := range []string{strings.Join(features, ","), ""} {
114 cmd = exec.Command("scontrol", "update", "NodeName="+slurmDummyNode, "Features="+nodeFeatures)
115 log.Printf("running: %q %q", cmd.Path, cmd.Args)
116 out, err := cmd.CombinedOutput()
118 log.Printf("error: scontrol: %s (output was %q)", err, out)
122 case bytes.Contains(out, []byte(slurmErrBadGres)):
123 // Evidently our node-type feature names are all valid.
126 log.Printf("warning: expected srun error %q or %q, but output was %q", slurmErrBadFeature, slurmErrBadGres, out)