This repository has been archived by the owner on Nov 27, 2023. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 252
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Merge pull request #628 from docker/machine
Guess AWS machine type based on service resources reservations
- Loading branch information
Showing
3 changed files
with
329 additions
and
0 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,212 @@ | ||
/* | ||
Copyright 2020 Docker, Inc. | ||
Licensed under the Apache License, Version 2.0 (the "License"); | ||
you may not use this file except in compliance with the License. | ||
You may obtain a copy of the License at | ||
http://www.apache.org/licenses/LICENSE-2.0 | ||
Unless required by applicable law or agreed to in writing, software | ||
distributed under the License is distributed on an "AS IS" BASIS, | ||
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
See the License for the specific language governing permissions and | ||
limitations under the License. | ||
*/ | ||
|
||
package ecs | ||
|
||
import ( | ||
"fmt" | ||
"math" | ||
"strconv" | ||
|
||
"github.com/compose-spec/compose-go/types" | ||
"github.com/docker/go-units" | ||
) | ||
|
||
type machine struct { | ||
id string | ||
cpus float64 | ||
memory types.UnitBytes | ||
gpus int64 | ||
} | ||
|
||
type family []machine | ||
|
||
var p3family = family{ | ||
{ | ||
id: "p3.2xlarge", | ||
cpus: 8, | ||
memory: 64 * units.GiB, | ||
gpus: 2, | ||
}, | ||
{ | ||
id: "p3.8xlarge", | ||
cpus: 32, | ||
memory: 244 * units.GiB, | ||
gpus: 4, | ||
}, | ||
{ | ||
id: "p3.16xlarge", | ||
cpus: 64, | ||
memory: 488 * units.GiB, | ||
gpus: 8, | ||
}, | ||
} | ||
|
||
type filterFn func(machine) bool | ||
|
||
func (f family) filter(fn filterFn) family { | ||
var filtered family | ||
for _, machine := range f { | ||
if fn(machine) { | ||
filtered = append(filtered, machine) | ||
} | ||
} | ||
return filtered | ||
} | ||
|
||
func (f family) firstOrError(msg string, args ...interface{}) (machine, error) { | ||
if len(f) == 0 { | ||
return machine{}, fmt.Errorf(msg, args...) | ||
} | ||
return f[0], nil | ||
} | ||
|
||
func guessMachineType(project *types.Project) (string, error) { | ||
// we select a machine type to match all gpus-bound services requirements | ||
// once https://github.com/aws/containers-roadmap/issues/631 is implemented we can define dedicated CapacityProviders per service. | ||
requirements, err := getResourceRequirements(project) | ||
if err != nil { | ||
return "", err | ||
} | ||
|
||
instanceType, err := p3family. | ||
filter(func(m machine) bool { | ||
return m.memory >= requirements.memory | ||
}). | ||
filter(func(m machine) bool { | ||
return m.cpus >= requirements.cpus | ||
}). | ||
filter(func(m machine) bool { | ||
return m.gpus >= requirements.gpus | ||
}). | ||
firstOrError("none of the Amazon EC2 P3 instance types meet the requirements for memory:%d cpu:%f gpus:%d", requirements.memory, requirements.cpus, requirements.gpus) | ||
if err != nil { | ||
return "", err | ||
} | ||
return instanceType.id, nil | ||
} | ||
|
||
type resourceRequirements struct { | ||
memory types.UnitBytes | ||
cpus float64 | ||
gpus int64 | ||
} | ||
|
||
func getResourceRequirements(project *types.Project) (*resourceRequirements, error) { | ||
return toResourceRequirementsSlice(project). | ||
filter(func(requirements *resourceRequirements) bool { | ||
return requirements.gpus != 0 | ||
}). | ||
max() | ||
} | ||
|
||
type eitherRequirementsOrError struct { | ||
requirements []*resourceRequirements | ||
err error | ||
} | ||
|
||
func toResourceRequirementsSlice(project *types.Project) eitherRequirementsOrError { | ||
var requirements []*resourceRequirements | ||
for _, service := range project.Services { | ||
r, err := toResourceRequirements(service) | ||
if err != nil { | ||
return eitherRequirementsOrError{nil, err} | ||
} | ||
requirements = append(requirements, r) | ||
} | ||
return eitherRequirementsOrError{requirements, nil} | ||
} | ||
|
||
func (r eitherRequirementsOrError) filter(fn func(*resourceRequirements) bool) eitherRequirementsOrError { | ||
if r.err != nil { | ||
return r | ||
} | ||
var requirements []*resourceRequirements | ||
for _, req := range r.requirements { | ||
if fn(req) { | ||
requirements = append(requirements, req) | ||
} | ||
} | ||
return eitherRequirementsOrError{requirements, nil} | ||
} | ||
|
||
func toResourceRequirements(service types.ServiceConfig) (*resourceRequirements, error) { | ||
if service.Deploy == nil { | ||
return nil, nil | ||
} | ||
reservations := service.Deploy.Resources.Reservations | ||
if reservations == nil { | ||
return nil, nil | ||
} | ||
|
||
var requiredGPUs int64 | ||
for _, r := range reservations.GenericResources { | ||
if r.DiscreteResourceSpec.Kind == "gpus" { | ||
requiredGPUs = r.DiscreteResourceSpec.Value | ||
break | ||
} | ||
} | ||
|
||
var nanocpu float64 | ||
if reservations.NanoCPUs != "" { | ||
v, err := strconv.ParseFloat(reservations.NanoCPUs, 64) | ||
if err != nil { | ||
return nil, err | ||
} | ||
nanocpu = v | ||
} | ||
return &resourceRequirements{ | ||
memory: reservations.MemoryBytes, | ||
cpus: nanocpu, | ||
gpus: requiredGPUs, | ||
}, nil | ||
} | ||
|
||
func (r resourceRequirements) combine(o *resourceRequirements) resourceRequirements { | ||
if o == nil { | ||
return r | ||
} | ||
return resourceRequirements{ | ||
memory: maxUnitBytes(r.memory, o.memory), | ||
cpus: math.Max(r.cpus, o.cpus), | ||
gpus: maxInt64(r.gpus, o.gpus), | ||
} | ||
} | ||
|
||
func (r eitherRequirementsOrError) max() (*resourceRequirements, error) { | ||
if r.err != nil { | ||
return nil, r.err | ||
} | ||
min := resourceRequirements{} | ||
for _, req := range r.requirements { | ||
min = min.combine(req) | ||
} | ||
return &min, nil | ||
} | ||
|
||
func maxInt64(a, b int64) int64 { | ||
if a > b { | ||
return a | ||
} | ||
return b | ||
} | ||
|
||
func maxUnitBytes(a, b types.UnitBytes) types.UnitBytes { | ||
if a > b { | ||
return a | ||
} | ||
return b | ||
} |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,115 @@ | ||
/* | ||
Copyright 2020 Docker, Inc. | ||
Licensed under the Apache License, Version 2.0 (the "License"); | ||
you may not use this file except in compliance with the License. | ||
You may obtain a copy of the License at | ||
http://www.apache.org/licenses/LICENSE-2.0 | ||
Unless required by applicable law or agreed to in writing, software | ||
distributed under the License is distributed on an "AS IS" BASIS, | ||
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
See the License for the specific language governing permissions and | ||
limitations under the License. | ||
*/ | ||
|
||
package ecs | ||
|
||
import ( | ||
"testing" | ||
) | ||
|
||
func TestGuessMachineType(t *testing.T) { | ||
tests := []struct { | ||
name string | ||
yaml string | ||
want string | ||
wantErr bool | ||
}{ | ||
{ | ||
name: "1-gpus", | ||
yaml: ` | ||
services: | ||
learning: | ||
image: tensorflow/tensorflow:latest-gpus | ||
deploy: | ||
resources: | ||
reservations: | ||
generic_resources: | ||
- discrete_resource_spec: | ||
kind: gpus | ||
value: 1 | ||
`, | ||
want: "p3.2xlarge", | ||
wantErr: false, | ||
}, | ||
{ | ||
name: "4-gpus", | ||
yaml: ` | ||
services: | ||
learning: | ||
image: tensorflow/tensorflow:latest-gpus | ||
deploy: | ||
resources: | ||
reservations: | ||
generic_resources: | ||
- discrete_resource_spec: | ||
kind: gpus | ||
value: 4 | ||
`, | ||
want: "p3.8xlarge", | ||
wantErr: false, | ||
}, | ||
{ | ||
name: "1-gpus, high-memory", | ||
yaml: ` | ||
services: | ||
learning: | ||
image: tensorflow/tensorflow:latest-gpus | ||
deploy: | ||
resources: | ||
reservations: | ||
memory: 300Gb | ||
generic_resources: | ||
- discrete_resource_spec: | ||
kind: gpus | ||
value: 2 | ||
`, | ||
want: "p3.16xlarge", | ||
wantErr: false, | ||
}, | ||
{ | ||
name: "1-gpus, high-cpu", | ||
yaml: ` | ||
services: | ||
learning: | ||
image: tensorflow/tensorflow:latest-gpus | ||
deploy: | ||
resources: | ||
reservations: | ||
memory: 32Gb | ||
cpus: "32" | ||
generic_resources: | ||
- discrete_resource_spec: | ||
kind: gpus | ||
value: 2 | ||
`, | ||
want: "p3.8xlarge", | ||
wantErr: false, | ||
}, | ||
} | ||
for _, tt := range tests { | ||
t.Run(tt.name, func(t *testing.T) { | ||
project := loadConfig(t, tt.yaml) | ||
got, err := guessMachineType(project) | ||
if (err != nil) != tt.wantErr { | ||
t.Errorf("guessMachineType() error = %v, wantErr %v", err, tt.wantErr) | ||
return | ||
} | ||
if got != tt.want { | ||
t.Errorf("guessMachineType() got = %v, want %v", got, tt.want) | ||
} | ||
}) | ||
} | ||
} |