Skip to content

Latest commit

 

History

History
281 lines (203 loc) · 9.45 KB

README.md

File metadata and controls

281 lines (203 loc) · 9.45 KB

This is a work in progress to use Terraform to manage our cloud infrastructure.

Instructions:

  • You will need a GCP project. Configure gcloud to point at your project:

    gcloud config set project <gcp-project-id>
gcloud config set compute/zone <gcp-zone>

  • Enable the GCP services needed by Hail:

    gcloud services enable \
    container.googleapis.com \
    compute.googleapis.com \
    cloudkms.googleapis.com \
    cloudresourcemanager.googleapis.com \
    servicenetworking.googleapis.com \
    sqladmin.googleapis.com \
    serviceusage.googleapis.com \
    dns.googleapis.com \
    logging.googleapis.com \
    cloudprofiler.googleapis.com \
    monitoring.googleapis.com \
    iam.googleapis.com \
    artifactregistry.googleapis.com \
    cloudbilling.googleapis.com

  • Delete the default network if it exists. Enabling the networking API creates it.

  • Go to the Google Cloud console, API & Services, Credentials. Configure the consent screen. Add the scope: https://www.googleapis.com/auth/userinfo.email. Back in Credentials, create an OAuth client ID. Authorize the redirect URIs:

    Download the client secret as /tmp/auth_oauth2_client_secret.json.

  • Create infra/gcp/$GITHUB_ORGANIZATION/global.tfvars based on the template below, where $GITHUB_ORGANIZATION corresponds to the GitHub organization used for your Hail Batch deployment (e.g. hail-is). This avoids collisions between configuration files from different Hail deployments.

    # organization_domain is a string that is the domain of the organization
# E.g. "hail.is"
organization_domain = "<domain>"

# The GitHub organization hosting your Hail Batch repository, e.g. "hail-is".
github_organization = "<github-organization>"

# batch_gcp_regions is a JSON array of string, the names of the gcp
# regions to schedule over in Batch. E.g. "[\"us-central1\"]"
batch_gcp_regions = "<batch-gcp-regions>"

gcp_project = "<gcp-project-id>"

# This is the bucket location that spans the regions you're going to
# schedule across in Batch.  If you are running on one region, it can
# just be that region. E.g. "US"
batch_logs_bucket_location = "<bucket-location>"

# The storage class for the batch logs bucket.  It should span the
# batch regions and be compatible with the bucket location.
batch_logs_bucket_storage_class = "MULTI_REGIONAL"

# Similarly, bucket locations and storage classes are specified
# for other services:
hail_query_bucket_location = "<bucket-location>"
hail_query_bucket_storage_class = "MULTI_REGIONAL"
hail_test_gcs_bucket_location = "<bucket-location>"
hail_test_gcs_bucket_storage_class = "MULTI_REGIONAL"

gcp_region = "<gcp-region>"

gcp_zone = "<gcp-zone>"

gcp_location = "<gcp-region>"

domain = "<domain>"

# If set to true, pull the base ubuntu image from Artifact Registry.
# Otherwise, assumes GCR.
use_artifact_registry = false

  • You can optionally create a /tmp/ci_config.json file to enable CI triggered by GitHub events. Note that github_oauth_token is not necessarily an OAuth2 access token. In fact, it should be a fine-grained personal access token. The currently public documentation on fine-grained access tokens is not very good. Check this page in github/docs for information on how to create a personal access token that is privileged to access the hail-is organization. Note in particular that personal access tokens have a "resource owner" field which is fixed at creation time. The token can only read or write to repositories owned by the "resource owner".

    {
    "bucket_location": "<gcp-zone>",
    "bucket_storage_class": "STANDARD",
    "deploy_steps": [
        "deploy_batch",
        "test_batch_0",
        "deploy_ci"
    ],
    "github_context": "ci-gcp",
    "github_oauth_token": "<TOKEN>",
    "github_user1_oauth_token": "<TOKEN>",
    "watched_branches": [
        [
            "hail-is/hail:main",
            true,
            false
        ]
    ]
}

  • Install sops.

  • Set up a key for sops to use:

    gcloud auth application-default login

gcloud kms keyrings create sops --location global

gcloud kms keys create sops-key --location global --keyring sops --purpose encryption

gcloud kms keys list --location global --keyring sops

    You should see:

    NAME                                                                         PURPOSE          PRIMARY_STATE
projects/<gcp-project-id>/locations/global/keyRings/sops/cryptoKeys/sops-key ENCRYPT_DECRYPT  ENABLED

    This key can be shared with other developers in your team, controlling access through IAM. It needs to be created outside of Terraform to avoid a cyclic dependency: the Terraform configuration needs to decrypt sops files.

  • Create a service account for Terraform with Owner role. We use service account name terraform. Create a JSON service account key and place it in /tmp/terraform_sa_key.json.

    gcloud iam service-accounts create terraform --display-name="Terraform Account"

gcloud projects add-iam-policy-binding <project-id> --member='serviceAccount:terraform@<project-id>.iam.gserviceaccount.com' --role='roles/owner'

gcloud iam service-accounts keys create /tmp/terraform_sa_key.json  --iam-account=terraform@<project-id>.iam.gserviceaccount.com

  • Encrypt the above files and add them to the repository.

    sops --encrypt --gcp-kms projects/<gcp-project-id>/locations/global/keyRings/sops/cryptoKeys/sops-key /tmp/auth_oauth2_client_secret.json > $HAIL/infra/gcp/$GITHUB_ORGANIZATION/auth_oauth2_client_secret.enc.json

# Optional
sops --encrypt --gcp-kms projects/<gcp-project-id>/locations/global/keyRings/sops/cryptoKeys/sops-key /tmp/ci_config.json > $HAIL/infra/gcp/$GITHUB_ORGANIZATION/ci_config.enc.json

sops --encrypt --gcp-kms projects/<gcp-project-id>/locations/global/keyRings/sops/cryptoKeys/sops-key /tmp/terraform_sa_key.json > $HAIL/infra/gcp/$GITHUB_ORGANIZATION/terraform_sa_key.enc.json

git add $HAIL/infra/gcp/$GITHUB_ORGANIZATION/*

# git commit and push as desired.

  • If you want Zulip integration for alerts from CI and Grafana, create a zuliprc file:

    cat /tmp/zuliprc <<EOF
[api]
key=SECRET_KEY_HERE
email=YOUR_BOT_EMAIL_HERE
site=YOUR_SITE_HERE
EOF

  • Encrypt the zuliprc with SOPS:

    sops --encrypt --gcp-kms projects/<gcp-project-id>/locations/global/keyRings/sops/cryptoKeys/sops-key /tmp/zuliprc \
     >$HAIL/infra/gcp/$GITHUB_ORGANIZATION/zuliprc.enc

  • Install terraform.

  • Run terraform init.

  • Run terraform apply -var-file=$GITHUB_ORGANIZATION/global.tfvars. At the time of writing, this takes ~15m.

  • Terraform created a GKE cluster named vdc. Configure kubectl to point at the vdc cluster:

    gcloud container clusters get-credentials --zone <gcp-zone> vdc

    Register domain with a DNS registry with the ip field in the Kubernetes global-config. This should point to the kubernetes external load balancer.

You can now install Hail:

  • Create a VM on the internal network, standard-8, 100GB PD-SSD, Ubuntu 22.04 TLS, allow full access to all Cloud APIs, use the Terraform service account. 10GB will run out of space. We assume the rest of the commands are run on the VM. You will need to connect to this instance with ssh. You may want to add a suiteable ssh forwarding rule to the default network.

  • Clone the Hail Github repository:

    git clone https://github.com/hail-is/hail.git

  • In the $HAIL/infra directory, run

    ./install_bootstrap_dependencies.sh

    At this point, log out and ssh back in (so that changes to group settings for Docker can be applied). The following steps should be completed from the $HAIL/infra/gcp directory, unless otherwise stated.

  • Run the following to authenticate docker and kubectl with the new artifact registry and kubernetes cluster, respectively. The GKE_ZONE is the zone of the GKE cluster and the GAR_REGION is the region of the artifact registry.

    ./bootstrap.sh configure_gcloud <GKE_ZONE> <GAR_REGION>

  • Edit $HAIL/letsencrypt/subdomains.txt to include just the services you plan to use in this deployment, e.g. auth, batch and batch-driver.

  • Deploy unmanaged resources by running

    ./bootstrap.sh deploy_unmanaged

  • Create the batch worker VM image. Run:

    NAMESPACE=default $HAIL/batch/gcp-create-worker-image.sh

  • Download the global-config to be used by bootstrap.py.

    mkdir /global-config
kubectl -n default get secret global-config -o json | jq -r '.data | map_values(@base64d) | to_entries|map("echo -n \(.value) > /global-config/\(.key)") | .[]' | bash

  • Bootstrap the cluster.

    ./bootstrap.sh bootstrap $GITHUB_ORGANIZATION/hail:<BRANCH> deploy_batch

  • Deploy the gateway: run make -C $HAIL/gateway envoy-xds-config deploy.

  • Create the initial (developer) user.

    ./bootstrap.sh bootstrap $GITHUB_ORGANIZATION/hail:<BRANCH> create_initial_user <USERNAME> <EMAIL>

    Additional users can be added by the initial user by going to auth./users.