Skip to content

igormusinov/competitions-v1-compute-worker

 
 

Repository files navigation

original code and README here (https://github.com/codalab/competitions-v1-compute-worker)

Why do you need ACI-workers ?

Original codalab's workers use local dockers on provisioned machines. These machines sometimes have to use GPU like Tesla k80 or others. In the clouds VMs with such cool parameters are expensive. If you want to have a scalable system you have to manage these machines (for instance, with Kubernetes), but it could be pretty difficult. That's why we are using ACI - Azure Container Instances (https://azure.microsoft.com/en-us/services/container-instances/) This is Azure service that give us a possibility to launch our serverless docker in the Cloud. You don't have to worry about a difficult infrastructure or scalability.

Deploy

Configs

 ───deploy
│   ├── configs
│   │   ├── afs.cred
│   │   ├── afs_creds.yml
│   │   ├── azure_creds
│   │   ├── env_config

  1. At first, you need to create Azure Resource Group and get azure_creds for this group. With this file your aci-codalab-worker will be able to launch ACI-containers (we use https://github.com/igormusinov/cloud-hunky library inside the aci-codalab-worker's container ).

azure_creds

{
  "clientId": "XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXX",
  "clientSecret": "XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX",
  "subscriptionId": "XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX",
  "tenantId": "XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX",
  "activeDirectoryEndpointUrl": "https://login.microsoftonline.com",
  "resourceManagerEndpointUrl": "https://management.azure.com/",
  "activeDirectoryGraphResourceId": "https://graph.windows.net/",
  "sqlManagementEndpointUrl": "https://management.core.windows.net:8443/",
  "galleryEndpointUrl": "https://gallery.azure.com/",
  "managementEndpointUrl": "https://management.core.windows.net/"
}
  1. You need the Azure File Storage (https://azure.microsoft.com/en-us/services/storage/files/) for mounting your data into the containers.

afs.cred

username=<name of your File Storage account>
password=<key>

afs_creds.yml

AFS_NAME: "name of your File Storage account"
AFS_KEY: "key"
AFS_SHARE: "your share"
AZURE_SECRET: "(optional)"
  1. Create env_config for your aci-codalab-worker.

env_config

# For ACI-container management (cloud-hunky)
AZURE_AUTH_LOCATION=/root/.azure/azure_creds

# For celery.
WORKER=celerymain
WORKER_CONCURRENCY=1

# For local/ACI launch
ACI_PRED=True
ACI_SCORE=True

# Logging config
LOG_CFG=/log/logging.yaml

# This is the URL you get from the Worker Queues Management page on Codalab
MAIN_BROKER_URL=<main virtual host>
BROKER_URL=<additional virtual host> (when codalab creates a new queue - it creates a new virtual host)

# Specify when use alternative queues
MAIN_BROKER_LOGIN=<your / login>
MAIN_BROKER_PASS=<your / pass>

# The temporary directory for submissions to do their work in
# Don't forget specify this path in docker mount
SUBMISSION_TEMP_DIR=/tmp/codalab

# Don't create this env
#BROKER_USE_SSL=True

# Docker Registry creds (Optional). For ACI's start time reduction
IMAGE_REGISTRY_SERVER=''
IMAGE_REGISTRY_USERNAME=''
IMAGE_REGISTRY_PWD=''
  1. Create directory deploy/configs and put these configs there.

Ansible deploy

  1. Set your params in the deploy/ansible.cfg and deploy/hosts.yaml (username and host).

  2. If your rabbitmq broker isn't available from outside the VM, you have to make ssh tunnel to you host. Don't forget to set right URLs in the configs/env_file

  3. Run ansible playbooks.

cd deploy
make mount
make deploy

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Languages

  • Python 95.5%
  • Makefile 2.8%
  • Dockerfile 1.4%
  • Shell 0.3%