Google Cloud Platform¶
||Cluster running on GCP VM Instances.|
In order to create clusters on GCP you need to set your authentication credentials.
You can do this via the
gcloud command line tool.
$ gcloud auth login
Alternatively you can use a service account which provides credentials in a JSON file.
You must set the
GOOGLE_APPLICATION_CREDENTIALS environment variable to the path to the JSON file.
$ export GOOGLE_APPLICATION_CREDENTIALS=/path/to/credentials.json
To use Dask Cloudprovider with GCP you must also configure your Project ID. Generally when creating a GCP account you will create a default project. This can be found at the top of the GCP dashboard.
Your Project ID must be added to your Dask config file.
# ~/.config/dask/cloudprovider.yaml cloudprovider: gcp: projectid: "YOUR PROJECT ID"
Or via an environment variable.
$ export DASK_CLOUDPROVIDER__GCP__PROJECTID="YOUR PROJECT ID"
Google Cloud VMs¶
GCPCluster(projectid=None, zone=None, machine_type=None, source_image=None, docker_image=None, ngpus=None, gpu_type=None, filesystem_size=None, auto_shutdown=None, bootstrap=True, **kwargs)¶
Cluster running on GCP VM Instances.
This cluster manager constructs a Dask cluster running on Google Cloud Platform 67VMs.
When configuring your cluster you may find it useful to install the
gcloudtool for querying the GCP API for available options.
- projectid: str
Your GCP project ID. This must be set either here or in your Dask config.
See the GCP docs page for more info.
- zone: str
The GCP zone to launch you cluster in. A full list can be obtained with
gcloud compute zones list.
- machine_type: str
The VM machine_type. You can get a full list with
gcloud compute machine-types list. The default is
n1-standard-1which is 3.75GB RAM and 1 vCPU
- source_image: str
The OS image to use for the VM. Dask Cloudprovider will boostrap Ubuntu based images automatically. Other images require Docker and for GPUs the NVIDIA Drivers and NVIDIA Docker.
A list of available images can be found with
gcloud compute images list
- Valid values are:
- The short image name provided it is in
- The full image name
- The full image URI such as those listed in
gcloud compute images list --uri.
- The short image name provided it is in
The default is
- docker_image: string (optional)
The Docker image to run on all instances.
This image must have a valid Python environment and have
daskinstalled in order for the
dask-workercommands to be available. It is recommended the Python environment matches your local environment where
EC2Clusteris being created from.
For GPU instance types the Docker image much have NVIDIA drivers and
By default the
daskdev/dask:latestimage will be used.
- ngpus: int (optional)
The number of GPUs to atatch to the instance. Default is
- gpu_type: str (optional)
The name of the GPU to use. This must be set if
ngpus>0. You can see a list of GPUs available in each zone with
gcloud compute accelerator-types list.
- filesystem_size: int (optional)
The VM filesystem size in GB. Defaults to
- n_workers: int (optional)
Number of workers to initialise the cluster with. Defaults to
- bootstrap: bool (optional)
Install Docker and NVIDIA drivers if
ngpus>0. Set to
Falseif you are using a custom
source_imagewhich already has these requirements. Defaults to
- worker_class: str
The Python class to run for the worker. Defaults to
- worker_options: dict (optional)
Params to be passed to the worker class. See
distributed.worker.Workerfor default worker class. If you set
worker_classthen refer to the docstring for the custom worker class.
- env_vars: dict (optional)
Environment variables to be passed to the worker.
- scheduler_options: dict (optional)
Params to be passed to the scheduler class. See
- silence_logs: bool (optional)
Whether or not we should silence logging when setting up the cluster.
- asynchronous: bool (optional)
If this is intended to be used directly within an event loop with async/await
- security : Security or bool (optional)
Configures communication security in this cluster. Can be a security object, or True. If True, temporary self-signed credentials will be created automatically.
Create the cluster.
>>> from dask_cloudprovider.gcp import GCPCluster >>> cluster = GCPCluster(n_workers=1) Launching cluster with the following configuration: Source Image: projects/ubuntu-os-cloud/global/images/ubuntu-minimal-1804-bionic-v20201014 Docker Image: daskdev/dask:latest Machine Type: n1-standard-1 Filesytsem Size: 50 N-GPU Type: Zone: us-east1-c Creating scheduler instance dask-acc897b9-scheduler Internal IP: 10.142.0.37 External IP: 22.214.171.124 Waiting for scheduler to run Scheduler is running Creating worker instance dask-acc897b9-worker-bfbc94bc Internal IP: 10.142.0.39 External IP: 126.96.36.199
Connect a client.
>>> from dask.distributed import Client >>> client = Client(cluster)
Do some work.
>>> import dask.array as da >>> arr = da.random.random((1000, 1000), chunks=(100, 100)) >>> arr.mean().compute() 0.5001550986751964
Close the cluster
>>> cluster.close() Closing Instance: dask-acc897b9-worker-bfbc94bc Closing Instance: dask-acc897b9-scheduler
You can also do this all in one go with context managers to ensure the cluster is created and cleaned up.
>>> with GCPCluster(n_workers=1) as cluster: ... with Client(cluster) as client: ... print(da.random.random((1000, 1000), chunks=(100, 100)).mean().compute()) Launching cluster with the following configuration: Source Image: projects/ubuntu-os-cloud/global/images/ubuntu-minimal-1804-bionic-v20201014 Docker Image: daskdev/dask:latest Machine Type: n1-standard-1 Filesytsem Size: 50 N-GPU Type: Zone: us-east1-c Creating scheduler instance dask-19352f29-scheduler Internal IP: 10.142.0.41 External IP: 188.8.131.52 Waiting for scheduler to run Scheduler is running Creating worker instance dask-19352f29-worker-91a6bfe0 Internal IP: 10.142.0.48 External IP: 184.108.40.206 0.5000812282861661 Closing Instance: dask-19352f29-worker-91a6bfe0 Closing Instance: dask-19352f29-scheduler
adapt(*args[, minimum, maximum])
Turn on adaptivity
get_logs([cluster, scheduler, workers])
Return logs for the cluster, scheduler and workers
Generate tags to be applied to all resources.
Return name and spec for the next worker
scale([n, memory, cores])
Scale cluster to n workers
scale_up([n, memory, cores])
Scale cluster to n workers close get_cloud_init logs render_cloud_init scale_down sync