feat(worker): Add backup remote for exported data

Author: nellh

config.env.example

@@ -39,6 +39,11 @@ AWS_ACCOUNT_ID=
 AWS_BATCH_QUEUE=
 AWS_BATCH_COMPUTE_ENVIRONMENT=
 
+# GCP API keys for S3-compatible access
+GCP_ACCESS_KEY_ID=
+GCP_SECRET_ACCESS_KEY=
+GCP_S3_BACKUP_BUCKET=
+
 # Dataset unpublished/published buckets - used by DataLad service
 AWS_S3_PRIVATE_BUCKET=
 AWS_S3_PUBLIC_BUCKET=

helm/openneuro/templates/secret.yaml

@@ -27,6 +27,9 @@ stringData:
   AWS_REGION: {{ required "AWS_REGION is required" .Values.secrets.aws.AWS_REGION | quote }}
   AWS_ACCOUNT_ID: {{ required "AWS_ACCOUNT_ID is required" .Values.secrets.aws.AWS_ACCOUNT_ID | quote }}
   AWS_ACCESS_KEY_ID: {{ required "AWS_ACCESS_KEY_ID is required" .Values.secrets.aws.AWS_ACCESS_KEY_ID | quote }}
+  GCP_ACCESS_KEY_ID: {{ required "GCP_ACCESS_KEY_ID is required" .Values.secrets.aws.GCP_ACCESS_KEY_ID | quote }}
+  GCP_SECRET_ACCESS_KEY: {{ required "GCP_SECRET_ACCESS_KEY is required" .Values.secrets.aws.GCP_SECRET_ACCESS_KEY | quote }}
+  GCP_S3_BACKUP_BUCKET: {{ required "GCP_S3_BACKUP_BUCKET is required" .Values.secrets.aws.GCP_S3_BACKUP_BUCKET | quote }}
   AWS_SECRET_ACCESS_KEY: {{ required "AWS_SECRET_ACCESS_KEY is required" .Values.secrets.aws.AWS_SECRET_ACCESS_KEY | quote }}
   AWS_S3_PRIVATE_BUCKET: {{ required "AWS_REGION is required" .Values.secrets.aws.AWS_S3_PRIVATE_BUCKET | quote }}
   AWS_S3_PUBLIC_BUCKET: {{ required "AWS_REGION is required" .Values.secrets.aws.AWS_S3_PUBLIC_BUCKET | quote }}

services/datalad/datalad_service/common/s3.py

@@ -12,43 +12,88 @@ def get_s3_remote():
     return 's3-PUBLIC'
 
 
+def get_s3_backup_remote():
+    return 's3-BACKUP'
+
+
 def get_s3_bucket():
     return getattr(datalad_service.config, 'AWS_S3_PUBLIC_BUCKET')
 
 
-def generate_s3_annex_options(dataset_path):
+def get_s3_backup_bucket():
+    return getattr(datalad_service.config, 'GCP_S3_BACKUP_BUCKET')
+
+
+def generate_s3_annex_options(dataset_path, backup=False):
     dataset_id = os.path.basename(dataset_path)
     annex_options = [
         'type=S3',
-        f'bucket={get_s3_bucket()}',
-        'exporttree=yes',
-        'versioning=yes',
         'partsize=1GiB',
         'encryption=none',
         f'fileprefix={dataset_id}/',
-        'autoenable=true',
-        f'publicurl=https://s3.amazonaws.com/{get_s3_bucket()}',
         'public=no',
     ]
+    if backup:
+        annex_options += [
+            f'bucket={get_s3_backup_bucket()}',
+            'cost=400',
+            'host=storage.googleapis.com',
+            'storageclass=ARCHIVE',
+        ]
+    else:
+        annex_options += [
+            'exporttree=yes',
+            'versioning=yes',
+            f'bucket={get_s3_bucket()}',
+            'autoenable=true',
+            f'publicurl=https://s3.amazonaws.com/{get_s3_bucket()}',
+        ]
     return annex_options
 
 
+def backup_remote_env():
+    """Copy and modify the environment for setup/modification of backup remote settings."""
+    backup_remote_env = os.environ.copy()
+    # Overwrite the AWS keys with the GCP key
+    backup_remote_env['AWS_ACCESS_KEY_ID'] = datalad_service.config.GCP_ACCESS_KEY_ID
+    backup_remote_env['AWS_SECRET_ACCESS_KEY'] = (
+        datalad_service.config.GCP_SECRET_ACCESS_KEY
+    )
+    return backup_remote_env
+
+
 def setup_s3_sibling(dataset_path):
     """Add a sibling for an S3 bucket publish."""
-    annex_options = generate_s3_annex_options(dataset_path)
+    # Public remote
+    subprocess.run(
+        ['git-annex', 'initremote', get_s3_remote()]
+        + generate_s3_annex_options(dataset_path),
+        cwd=dataset_path,
+    )
+    # Backup remote
     subprocess.run(
-        ['git-annex', 'initremote', get_s3_remote()] + annex_options, cwd=dataset_path
+        ['git-annex', 'initremote', get_s3_backup_remote()]
+        + generate_s3_annex_options(dataset_path, backup=True),
+        cwd=dataset_path,
+        env=backup_remote_env(),
     )
 
 
 def update_s3_sibling(dataset_path):
     """Update S3 remote with latest config."""
-    annex_options = generate_s3_annex_options(dataset_path)
     # note: enableremote command will only upsert config options, none are deleted
     subprocess.run(
-        ['git-annex', 'enableremote', get_s3_remote()] + annex_options,
+        ['git-annex', 'enableremote', get_s3_remote()]
+        + generate_s3_annex_options(dataset_path),
+        check=True,
+        cwd=dataset_path,
+    )
+    subprocess.run(
+        ['git-annex', 'enableremote', get_s3_backup_remote()]
+        + generate_s3_annex_options(dataset_path, backup=True),
         check=True,
         cwd=dataset_path,
+        env=backup_remote_env(),
     )
 
 
@@ -87,3 +132,12 @@ def s3_export(dataset_path, target, treeish):
     subprocess.check_call(
         ['git-annex', 'export', treeish, '--to', target], cwd=dataset_path
     )
+
+
+def s3_backup_push(dataset_path):
+    """Perform an S3 push to the backup remote on a git-annex repo."""
+    subprocess.check_call(
+        ['git-annex', 'push', get_s3_backup_remote()],
+        cwd=dataset_path,
+        env=backup_remote_env(),
+    )

services/datalad/datalad_service/config.py

@@ -19,6 +19,12 @@
 AWS_S3_PRIVATE_BUCKET = os.getenv('AWS_S3_PRIVATE_BUCKET')
 AWS_S3_PUBLIC_BUCKET = os.getenv('AWS_S3_PUBLIC_BUCKET')
 
+# GCP S3 compatible object storage
+GCP_ACCESS_KEY_ID = os.getenv('GCP_ACCESS_KEY_ID')
+GCP_SECRET_ACCESS_KEY = os.getenv('GCP_SECRET_ACCESS_KEY')
+GCP_S3_BACKUP_BUCKET = os.getenv('GCP_S3_BACKUP_BUCKET')
+
+
 # GraphQL URL - override if not docker-compose
 GRAPHQL_ENDPOINT = os.getenv('GRAPHQL_ENDPOINT', 'http://server:8111/crn/graphql')
 

services/datalad/datalad_service/tasks/publish.py

@@ -15,14 +15,18 @@
 from datalad_service.config import DATALAD_GITHUB_EXPORTS_ENABLED
 from datalad_service.config import AWS_ACCESS_KEY_ID
 from datalad_service.config import AWS_SECRET_ACCESS_KEY
+from datalad_service.config import GCP_ACCESS_KEY_ID
+from datalad_service.config import GCP_SECRET_ACCESS_KEY
 from datalad_service.common.annex import get_tag_info, is_git_annex_remote
 from datalad_service.common.openneuro import clear_dataset_cache
 from datalad_service.common.git import git_show, git_tag, git_tag_tree
 from datalad_service.common.github import github_export
 from datalad_service.common.s3 import (
     s3_export,
+    s3_backup_push,
     get_s3_remote,
     get_s3_bucket,
+    get_s3_backup_bucket,
     update_s3_sibling,
 )
 from datalad_service.broker import broker
@@ -90,6 +94,7 @@ def export_dataset(
         # Push the most recent tag
         if tags:
             s3_export(dataset_path, get_s3_remote(), tags[-1].name)
+            s3_backup_push(dataset_path)
             # Once all S3 tags are exported, update GitHub
             if github_enabled:
                 # Perform all GitHub export steps
@@ -129,16 +134,27 @@ def check_remote_has_version(dataset_path, remote, tag):
 def delete_s3_sibling(dataset_id):
     """Run S3 sibling deletion in another process to avoid blocking any callers"""
     delete_executor.submit(delete_s3_sibling_executor, dataset_id)
+    delete_executor.submit(delete_s3_sibling_executor, dataset_id, True)
 
 
-def delete_s3_sibling_executor(dataset_id):
+def delete_s3_sibling_executor(dataset_id, backup=False):
     """Delete all versions of a dataset from S3."""
     try:
-        client = boto3.client(
-            's3',
-            aws_access_key_id=AWS_ACCESS_KEY_ID,
-            aws_secret_access_key=AWS_SECRET_ACCESS_KEY,
-        )
+        if backup:
+            s3_bucket = get_s3_backup_bucket()
+            client = boto3.client(
+                's3',
+                aws_access_key_id=GCP_ACCESS_KEY_ID,
+                aws_secret_access_key=GCP_SECRET_ACCESS_KEY,
+                endpoint_url='https://storage.googleapis.com',
+            )
+        else:
+            s3_bucket = get_s3_bucket()
+            client = boto3.client(
+                's3',
+                aws_access_key_id=AWS_ACCESS_KEY_ID,
+                aws_secret_access_key=AWS_SECRET_ACCESS_KEY,
+            )
         paginator = client.get_paginator('list_object_versions')
         object_delete_list = []
         for response in paginator.paginate(
@@ -154,12 +170,12 @@ def delete_s3_sibling_executor(dataset_id):
             )
         for i in range(0, len(object_delete_list), 1000):
             client.delete_objects(
-                Bucket=get_s3_bucket(),
+                Bucket=s3_bucket,
                 Delete={'Objects': object_delete_list[i : i + 1000], 'Quiet': True},
             )
     except Exception as e:
         raise Exception(
-            f'Attempt to delete dataset {dataset_id} from {get_s3_remote()} has failed. ({e})'
+            f'Attempt to delete dataset {dataset_id} from {s3_bucket} has failed. ({e})'
         )
 
 

services/datalad/tests/test_publish.py

@@ -60,7 +60,7 @@ def test_export_snapshots(no_init_remote, client, new_dataset):
         github_enabled=True,
     )
     # Verify export calls were made
-    assert s3_export_mock.call_count == 1
+    assert s3_export_mock.call_count == 2
     expect_calls = [call(new_dataset.path, 's3-PUBLIC', 'refs/tags/2.0.0')]
     s3_export_mock.assert_has_calls(expect_calls)
     assert github_export_mock.call_count == 1