mirror of
https://github.com/helm/charts.git
synced 2026-08-23 22:37:45 +00:00
Add distributed tensorflow (#5187)
* Add distributed tensorflow * list indentation * move to stable
This commit is contained in:
cheyang
k8s-ci-robot
@@ -0,0 +1,23 @@
|
||||
# Patterns to ignore when building packages.
|
||||
# This supports shell glob matching, relative path matching, and
|
||||
# negation (prefixed with !). Only one pattern per line.
|
||||
.DS_Store
|
||||
docker
|
||||
*.jpg
|
||||
# Common VCS dirs
|
||||
.git/
|
||||
.gitignore
|
||||
.bzr/
|
||||
.bzrignore
|
||||
.hg/
|
||||
.hgignore
|
||||
.svn/
|
||||
# Common backup files
|
||||
*.swp
|
||||
*.bak
|
||||
*.tmp
|
||||
*~
|
||||
# Various IDEs
|
||||
.project
|
||||
.idea/
|
||||
*.tmproj
|
||||
@@ -0,0 +1,12 @@
|
||||
apiVersion: v1
|
||||
description: A Helm chart for running distributed TensorFlow on Kubernetes
|
||||
name: distributed-tensorflow
|
||||
version: 0.1.0
|
||||
appVersion: 1.6.0
|
||||
sources:
|
||||
- https://github.com/tensorflow/tensorflow
|
||||
- https://www.tensorflow.org/deploy/distributed
|
||||
home: https://www.tensorflow.org
|
||||
maintainers:
|
||||
- name: cheyang
|
||||
email: cheyang@163.com
|
||||
@@ -0,0 +1,85 @@
|
||||
# Distributed TensorFlow
|
||||
|
||||
TensorFlow is an open source software library for numerical computation using data flow graphs, and it supports distributed computing, allowing 'data parallel' or 'model parallel' on different servers. This means data scientists can now scale out distributed training to 100s of GPUs using TensorFlow.
|
||||
|
||||
- https://www.tensorflow.org
|
||||
- https://www.tensorflow.org/deploy/distributed
|
||||
|
||||
|
||||
|
||||
## Prerequisites
|
||||
|
||||
- Kubernetes cluster v1.8+
|
||||
|
||||
## Chart Details
|
||||
|
||||
This chart will create a TensorFlow cluster, and distribute a computation graph across that cluster.
|
||||
|
||||
## Installing the Chart
|
||||
|
||||
* To install the chart with the release name `mnist`:
|
||||
|
||||
```bash
|
||||
$ helm install --name mnist incubator/distributed-tensorflow
|
||||
```
|
||||
|
||||
* To install with custom values via file:
|
||||
|
||||
```
|
||||
$ helm install --values values.yaml --name mnist incubator/distributed-tensorflow
|
||||
```
|
||||
|
||||
Below is an example of the custom value file values.yaml with GPU support.
|
||||
|
||||
```
|
||||
worker:
|
||||
number: 2
|
||||
podManagementPolicy: Parallel
|
||||
image:
|
||||
repository: cheyang/distributed-tf
|
||||
tag: 1.6.0-gpu
|
||||
port: 9090
|
||||
gpuCount: 1
|
||||
|
||||
ps:
|
||||
number: 2
|
||||
podManagementPolicy: Parallel
|
||||
image:
|
||||
repository: cheyang/distributed-tf
|
||||
tag: 1.6.0
|
||||
pullPolicy: IfNotPresent
|
||||
port: 8080
|
||||
|
||||
# optimize for training
|
||||
hyperparams:
|
||||
batchsize: 20
|
||||
learningrate: 0.001
|
||||
```
|
||||
|
||||
> Notice: you can check the details of docker image from [Docker hub](https://hub.docker.com/r/cheyang/distributed-tf/)
|
||||
|
||||
## Uninstalling the Chart
|
||||
|
||||
* To uninstall/delete the `mnist` deployment:
|
||||
|
||||
```bash
|
||||
$ helm delete mnist
|
||||
```
|
||||
|
||||
The command removes all the Kubernetes components associated with the chart and deletes the release.
|
||||
|
||||
## Configuration
|
||||
|
||||
The following tables lists the configurable parameters of the Service Distributed Tensorflow
|
||||
chart and their default values.
|
||||
|
||||
| Parameter | Description | Default |
|
||||
|-----------|-------------|---------|
|
||||
| `worker.image.repository` | TensorFlow Worker Server's image repository | `cheyang/distributed-tf` |
|
||||
| `worker.image.tag` | TensorFlow Worker Server's image tag | `gpu` |
|
||||
| `worker.image.pullPolicy` | image pullPolicy for the worker | `IfNotPresent` |
|
||||
| `worker.gpuCount` | Set the gpu to be allocated and allowed for the Pods | `0` |
|
||||
| `ps.image.repository` | TensorFlow Parameter Server's image repository | `cheyang/distributed-tf` |
|
||||
| `ps.image.tag` | TensorFlow Parameter Server's image tag | `1.6.0-gpu` |
|
||||
| `ps.image.pullPolicy` | image pullPolicy for the ps | `IfNotPresent` |
|
||||
|
||||
Binary file not shown.
|
After Width: | Height: | Size: 255 KiB |
@@ -0,0 +1,4 @@
|
||||
1. Get the application Status by running these commands:
|
||||
|
||||
echo "Get the statefulset of distributed tensorflow"
|
||||
kubectl get sts --namespace {{ .Release.Namespace }} -l "app={{ template "distributed-tensorflow.name" . }},release={{ .Release.Name }}"
|
||||
@@ -0,0 +1,32 @@
|
||||
{{/* vim: set filetype=mustache: */}}
|
||||
{{/*
|
||||
Expand the name of the chart.
|
||||
*/}}
|
||||
{{- define "distributed-tensorflow.name" -}}
|
||||
{{- default .Chart.Name .Values.nameOverride | trunc 63 | trimSuffix "-" -}}
|
||||
{{- end -}}
|
||||
|
||||
{{/*
|
||||
Create a default fully qualified app name.
|
||||
We truncate at 63 chars because some Kubernetes name fields are limited to this (by the DNS naming spec).
|
||||
If release name contains chart name it will be used as a full name.
|
||||
*/}}
|
||||
{{- define "distributed-tensorflow.fullname" -}}
|
||||
{{- if .Values.fullnameOverride -}}
|
||||
{{- .Values.fullnameOverride | trunc 63 | trimSuffix "-" -}}
|
||||
{{- else -}}
|
||||
{{- $name := default .Chart.Name .Values.nameOverride -}}
|
||||
{{- if contains $name .Release.Name -}}
|
||||
{{- .Release.Name | trunc 63 | trimSuffix "-" -}}
|
||||
{{- else -}}
|
||||
{{- printf "%s-%s" .Release.Name $name | trunc 63 | trimSuffix "-" -}}
|
||||
{{- end -}}
|
||||
{{- end -}}
|
||||
{{- end -}}
|
||||
|
||||
{{/*
|
||||
Create chart name and version as used by the chart label.
|
||||
*/}}
|
||||
{{- define "distributed-tensorflow.chart" -}}
|
||||
{{- printf "%s-%s" .Chart.Name .Chart.Version | replace "+" "_" | trunc 63 | trimSuffix "-" -}}
|
||||
{{- end -}}
|
||||
@@ -0,0 +1,24 @@
|
||||
{{- $workerNum := .Values.worker.number -}}
|
||||
{{- $workerPort := .Values.worker.port -}}
|
||||
{{- $psNum := .Values.ps.number -}}
|
||||
{{- $psPort := .Values.ps.port -}}
|
||||
{{- $tfService := include "distributed-tensorflow.fullname" . }}
|
||||
{{- $releaseName := .Release.Name -}}
|
||||
apiVersion: v1
|
||||
kind: ConfigMap
|
||||
metadata:
|
||||
name: {{ template "distributed-tensorflow.fullname" . }}
|
||||
labels:
|
||||
heritage: {{ .Release.Service | quote }}
|
||||
release: {{ .Release.Name | quote }}
|
||||
chart: {{ template "distributed-tensorflow.chart" . }}
|
||||
app: {{ template "distributed-tensorflow.name" . }}
|
||||
data:
|
||||
ps.hostList: "
|
||||
{{- range $i, $none := until (int $psNum) }}
|
||||
{{- if gt $i 0}},{{- end }}{{ $releaseName }}-ps-{{ $i }}.{{ $tfService }}-ps:{{ $psPort }}
|
||||
{{- end }}"
|
||||
worker.hostList: "
|
||||
{{- range $i, $none := until (int $workerNum) }}
|
||||
{{- if gt $i 0}},{{- end }}{{ $releaseName }}-worker-{{ $i }}.{{ $tfService }}-worker:{{ $workerPort }}
|
||||
{{- end }}"
|
||||
@@ -0,0 +1,20 @@
|
||||
apiVersion: v1
|
||||
kind: Service
|
||||
metadata:
|
||||
name: {{ template "distributed-tensorflow.fullname" . }}-ps
|
||||
labels:
|
||||
app: {{ template "distributed-tensorflow.name" . }}
|
||||
chart: {{ template "distributed-tensorflow.chart" . }}
|
||||
release: {{ .Release.Name }}
|
||||
heritage: {{ .Release.Service }}
|
||||
spec:
|
||||
clusterIP: None
|
||||
ports:
|
||||
- port: {{ .Values.ps.port }}
|
||||
targetPort: {{ .Values.ps.port }}
|
||||
protocol: TCP
|
||||
name: ps
|
||||
selector:
|
||||
app: {{ template "distributed-tensorflow.name" . }}
|
||||
release: {{ .Release.Name }}
|
||||
role: ps
|
||||
@@ -0,0 +1,20 @@
|
||||
apiVersion: v1
|
||||
kind: Service
|
||||
metadata:
|
||||
name: {{ template "distributed-tensorflow.fullname" . }}-worker
|
||||
labels:
|
||||
app: {{ template "distributed-tensorflow.name" . }}
|
||||
chart: {{ template "distributed-tensorflow.chart" . }}
|
||||
release: {{ .Release.Name }}
|
||||
heritage: {{ .Release.Service }}
|
||||
spec:
|
||||
clusterIP: None
|
||||
ports:
|
||||
- port: {{ .Values.worker.port }}
|
||||
targetPort: {{ .Values.worker.port }}
|
||||
protocol: TCP
|
||||
name: worker
|
||||
selector:
|
||||
app: {{ template "distributed-tensorflow.name" . }}
|
||||
release: {{ .Release.Name }}
|
||||
role: worker
|
||||
@@ -0,0 +1,83 @@
|
||||
{{- $lr := .Values.hyperparams.learningrate -}}
|
||||
{{- $batchsize := .Values.hyperparams.batchsize -}}
|
||||
|
||||
apiVersion: apps/v1beta2
|
||||
kind: StatefulSet
|
||||
metadata:
|
||||
name: {{ .Release.Name }}-ps
|
||||
labels:
|
||||
app: {{ template "distributed-tensorflow.name" . }}
|
||||
chart: {{ template "distributed-tensorflow.chart" . }}
|
||||
release: {{ .Release.Name }}
|
||||
heritage: {{ .Release.Service }}
|
||||
role: ps
|
||||
spec:
|
||||
selector:
|
||||
matchLabels:
|
||||
app: {{ template "distributed-tensorflow.name" . }}
|
||||
release: {{ .Release.Name }}
|
||||
role: ps
|
||||
serviceName: {{ template "distributed-tensorflow.fullname" . }}-ps
|
||||
podManagementPolicy: {{ .Values.ps.podManagementPolicy }}
|
||||
replicas: {{.Values.ps.number}}
|
||||
template:
|
||||
metadata:
|
||||
labels:
|
||||
app: {{ template "distributed-tensorflow.name" . }}
|
||||
chart: {{ template "distributed-tensorflow.chart" . }}
|
||||
release: {{ .Release.Name }}
|
||||
heritage: {{ .Release.Service }}
|
||||
role: ps
|
||||
spec:
|
||||
{{- if .Values.volumes }}
|
||||
volumes:
|
||||
{{ toYaml .Values.volumes | indent 6 }}
|
||||
{{- end }}
|
||||
containers:
|
||||
- name: ps
|
||||
image: "{{ .Values.ps.image.repository }}:{{ .Values.ps.image.tag }}"
|
||||
imagePullPolicy: {{ .Values.ps.image.pullPolicy }}
|
||||
command:
|
||||
- "python"
|
||||
- "train_distributed.py"
|
||||
- --learning_rate
|
||||
- {{ $lr | quote }}
|
||||
- --batch_size
|
||||
- {{ $batchsize | quote }}
|
||||
env:
|
||||
- name: WORKER_HOSTS
|
||||
valueFrom:
|
||||
configMapKeyRef:
|
||||
name: {{ template "distributed-tensorflow.fullname" . }}
|
||||
key: worker.hostList
|
||||
- name: PS_HOSTS
|
||||
valueFrom:
|
||||
configMapKeyRef:
|
||||
name: {{ template "distributed-tensorflow.fullname" . }}
|
||||
key: ps.hostList
|
||||
- name: POD_NAME
|
||||
valueFrom:
|
||||
fieldRef:
|
||||
fieldPath: metadata.name
|
||||
- name: JOB_NAME
|
||||
value: ps
|
||||
{{- if .Values.ps.env }}
|
||||
{{- range $key, $value := .Values.ps.env }}
|
||||
- name: "{{ $key }}"
|
||||
value: "{{ $value }}"
|
||||
{{- end }}
|
||||
{{- end }}
|
||||
{{- if .Values.ps.privileged }}
|
||||
securityContext:
|
||||
privileged: true
|
||||
{{- end }}
|
||||
ports:
|
||||
- containerPort: {{ .Values.ps.port }}
|
||||
{{- if .Values.volumeMounts }}
|
||||
volumeMounts:
|
||||
{{ toYaml .Values.volumeMounts | indent 8 }}
|
||||
{{- end }}
|
||||
{{- if .Values.ps.resources }}
|
||||
resources:
|
||||
{{ toYaml .Values.ps.resources | indent 10 }}
|
||||
{{- end }}
|
||||
@@ -0,0 +1,86 @@
|
||||
{{- $lr := .Values.hyperparams.learningrate -}}
|
||||
{{- $batchsize := .Values.hyperparams.batchsize -}}
|
||||
|
||||
apiVersion: apps/v1beta2
|
||||
kind: StatefulSet
|
||||
metadata:
|
||||
name: {{ .Release.Name }}-worker
|
||||
labels:
|
||||
app: {{ template "distributed-tensorflow.name" . }}
|
||||
chart: {{ template "distributed-tensorflow.chart" . }}
|
||||
release: {{ .Release.Name }}
|
||||
heritage: {{ .Release.Service }}
|
||||
role: worker
|
||||
spec:
|
||||
selector:
|
||||
matchLabels:
|
||||
app: {{ template "distributed-tensorflow.name" . }}
|
||||
release: {{ .Release.Name }}
|
||||
role: worker
|
||||
serviceName: {{ template "distributed-tensorflow.fullname" . }}-worker
|
||||
podManagementPolicy: {{ .Values.worker.podManagementPolicy }}
|
||||
replicas: {{.Values.worker.number}}
|
||||
template:
|
||||
metadata:
|
||||
labels:
|
||||
app: {{ template "distributed-tensorflow.name" . }}
|
||||
chart: {{ template "distributed-tensorflow.chart" . }}
|
||||
release: {{ .Release.Name }}
|
||||
heritage: {{ .Release.Service }}
|
||||
role: worker
|
||||
spec:
|
||||
{{- if .Values.volumes }}
|
||||
volumes:
|
||||
{{ toYaml .Values.volumes | indent 6 }}
|
||||
{{- end }}
|
||||
containers:
|
||||
- name: worker
|
||||
image: "{{ .Values.worker.image.repository }}:{{ .Values.worker.image.tag }}"
|
||||
imagePullPolicy: {{ .Values.worker.image.pullPolicy }}
|
||||
command:
|
||||
- "python"
|
||||
- "train_distributed.py"
|
||||
{{- if gt (int .Values.worker.gpuCount) 0 }}
|
||||
- --num_gpus
|
||||
- "{{ .Values.worker.gpuCount }}"
|
||||
{{- end }}
|
||||
- --learning_rate
|
||||
- {{ $lr | quote }}
|
||||
- --batch_size
|
||||
- {{ $batchsize | quote }}
|
||||
env:
|
||||
- name: WORKER_HOSTS
|
||||
valueFrom:
|
||||
configMapKeyRef:
|
||||
name: {{ template "distributed-tensorflow.fullname" . }}
|
||||
key: worker.hostList
|
||||
- name: PS_HOSTS
|
||||
valueFrom:
|
||||
configMapKeyRef:
|
||||
name: {{ template "distributed-tensorflow.fullname" . }}
|
||||
key: ps.hostList
|
||||
- name: POD_NAME
|
||||
valueFrom:
|
||||
fieldRef:
|
||||
fieldPath: metadata.name
|
||||
- name: JOB_NAME
|
||||
value: worker
|
||||
{{- if .Values.worker.env }}
|
||||
{{- range $key, $value := .Values.worker.env }}
|
||||
- name: "{{ $key }}"
|
||||
value: "{{ $value }}"
|
||||
{{- end }}
|
||||
{{- end }}
|
||||
ports:
|
||||
- containerPort: {{ .Values.worker.port }}
|
||||
{{- if .Values.volumeMounts }}
|
||||
volumeMounts:
|
||||
{{ toYaml .Values.volumeMounts | indent 8 }}
|
||||
{{- end }}
|
||||
{{- if gt (int .Values.worker.gpuCount) 0 }}
|
||||
resources:
|
||||
limits:
|
||||
nvidia.com/gpu: {{- .Values.worker.gpuCount }}
|
||||
request:
|
||||
nvidia.com/gpu: {{- .Values.worker.gpuCount }}
|
||||
{{- end }}
|
||||
@@ -0,0 +1,24 @@
|
||||
# Default values for distributed-tensorflow.
|
||||
# This is a YAML-formatted file.
|
||||
# Declare variables to be passed into your templates.
|
||||
worker:
|
||||
number: 2
|
||||
podManagementPolicy: Parallel
|
||||
image:
|
||||
repository: cheyang/distributed-tf
|
||||
tag: 1.6.0
|
||||
pullPolicy: IfNotPresent
|
||||
port: 9000
|
||||
# gpuCount: 2
|
||||
ps:
|
||||
number: 2
|
||||
podManagementPolicy: Parallel
|
||||
image:
|
||||
repository: cheyang/distributed-tf
|
||||
tag: 1.6.0
|
||||
pullPolicy: IfNotPresent
|
||||
port: 8000
|
||||
# optimize for training
|
||||
hyperparams:
|
||||
batchsize: 20
|
||||
learningrate: 0.001
|
||||
Reference in New Issue
Block a user