1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
14 "git.curoverse.com/arvados.git/sdk/go/arvados"
18 ErrConstraintsNotSatisfiable = errors.New("constraints not satisfiable by any configured instance type")
19 ErrInstanceTypesNotConfigured = errors.New("site configuration does not list any instance types")
20 discountConfiguredRAMPercent = 5
23 // ChooseInstanceType returns the cheapest available
24 // arvados.InstanceType big enough to run ctr.
25 func ChooseInstanceType(cc *arvados.Cluster, ctr *arvados.Container) (best arvados.InstanceType, err error) {
26 if len(cc.InstanceTypes) == 0 {
27 err = ErrInstanceTypesNotConfigured
31 needScratch := int64(0)
32 for _, m := range ctr.Mounts {
34 needScratch += m.Capacity
38 needVCPUs := ctr.RuntimeConstraints.VCPUs
40 needRAM := ctr.RuntimeConstraints.RAM + ctr.RuntimeConstraints.KeepCacheRAM
41 needRAM = (needRAM * 100) / int64(100-discountConfiguredRAMPercent)
43 err = ErrConstraintsNotSatisfiable
44 for _, it := range cc.InstanceTypes {
46 case err == nil && it.Price > best.Price:
47 case it.Scratch < needScratch:
48 case it.RAM < needRAM:
49 case it.VCPUs < needVCPUs:
50 case it.Preemptable != ctr.SchedulingParameters.Preemptable:
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) To test whether a set of feature names are valid without
73 // actually submitting a job, we can call srun --test-only with the
76 // SlurmNodeTypeFeatureKludge does a test-and-fix operation
77 // immediately, and then periodically, in case slurm restarts and
78 // forgets the list of valid features. It never returns (unless there
79 // are no node types configured, in which case it returns
80 // immediately), so it should generally be invoked with "go".
81 func SlurmNodeTypeFeatureKludge(cc *arvados.Cluster) {
82 if len(cc.InstanceTypes) == 0 {
86 for _, it := range cc.InstanceTypes {
87 features = append(features, "instancetype="+it.Name)
91 time.Sleep(2 * time.Second)
95 const slurmDummyNode = "compute0"
97 func slurmKludge(features []string) {
98 allFeatures := strings.Join(features, ",")
100 cmd := exec.Command("sinfo", "--nodes="+slurmDummyNode, "--format=%f", "--noheader")
101 out, err := cmd.CombinedOutput()
103 log.Printf("running %q %q: %s (output was %q)", cmd.Path, cmd.Args, err, out)
106 if string(out) == allFeatures+"\n" {
107 // Already configured correctly, nothing to do.
111 log.Printf("configuring node %q with all node type features", slurmDummyNode)
112 cmd = exec.Command("scontrol", "update", "NodeName="+slurmDummyNode, "Features="+allFeatures)
113 log.Printf("running: %q %q", cmd.Path, cmd.Args)
114 out, err = cmd.CombinedOutput()
116 log.Printf("error: scontrol: %s (output was %q)", err, out)