7478: Unify arvados node size access across different drivers

[arvados.git] / services / nodemanager / arvnodeman / computenode / driver / gce.py

diff --git a/services/nodemanager/arvnodeman/computenode/driver/gce.py b/services/nodemanager/arvnodeman/computenode/driver/gce.py
index bbabdd4c761b5a0e3809449878227adea7db0a5a..f8d81cab041d974ac253e3d11f68e95dcfd1abd3 100644 (file)
--- a/services/nodemanager/arvnodeman/computenode/driver/gce.py
+++ b/services/nodemanager/arvnodeman/computenode/driver/gce.py
@@ -1,4 +1,7 @@
 #!/usr/bin/env python
+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: AGPL-3.0
 
 from __future__ import absolute_import, print_function
 
@@ -31,10 +34,11 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
         create_kwargs = create_kwargs.copy()
         create_kwargs.setdefault('external_ip', None)
         create_kwargs.setdefault('ex_metadata', {})
+        self._project = auth_kwargs.get("project")
         super(ComputeNodeDriver, self).__init__(
             auth_kwargs, list_kwargs, create_kwargs,
             driver_class)
-        self._sizes_by_name = {sz.name: sz for sz in self.sizes.itervalues()}
+        self._sizes_by_id = {sz.id: sz for sz in self.sizes.itervalues()}
         self._disktype_links = {dt.name: self._object_link(dt)
                                 for dt in self.real.ex_list_disktypes()}
 
@@ -44,7 +48,7 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
 
     def _init_image(self, image_name):
         return 'image', self.search_for(
-            image_name, 'list_images', self._name_key)
+            image_name, 'list_images', self._name_key, ex_project=self._project)
 
     def _init_network(self, network_name):
         return 'ex_network', self.search_for(
@@ -60,9 +64,16 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
             self.create_kwargs['ex_metadata']['sshKeys'] = (
                 'root:' + ssh_file.read().strip())
 
+    def create_cloud_name(self, arvados_node):
+        uuid_parts = arvados_node['uuid'].split('-', 2)
+        return 'compute-{parts[2]}-{parts[0]}'.format(parts=uuid_parts)
+
     def arvados_create_kwargs(self, size, arvados_node):
-        cluster_id, _, node_id = arvados_node['uuid'].split('-')
-        name = 'compute-{}-{}'.format(node_id, cluster_id)
+        name = self.create_cloud_name(arvados_node)
+
+        if size.scratch > 375000:
+            self._logger.warning("Requested %d MB scratch space, but GCE driver currently only supports attaching a single 375 GB disk.", size.scratch)
+
         disks = [
             {'autoDelete': True,
              'boot': True,
@@ -98,6 +109,11 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
         return result
 
 
+    def create_node(self, size, arvados_node):
+        # Set up tag indicating the Arvados assigned Cloud Size id.
+        self.create_kwargs['ex_metadata'].update({'arvados_node_size': size.id})
+        return super(ComputeNodeDriver, self).create_node(size, arvados_node)
+
     def list_nodes(self):
         # The GCE libcloud driver only supports filtering node lists by zone.
         # Do our own filtering based on tag list.
@@ -109,7 +125,8 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
         # and monkeypatch the results when that's the case.
         if nodelist and not hasattr(nodelist[0].size, 'id'):
             for node in nodelist:
-                node.size = self._sizes_by_name[node.size]
+                node.size = self._sizes_by_id[node.size]
+                node.extra['arvados_node_size'] = node.extra.get('metadata', {}).get('arvados_node_size')
         return nodelist
 
     @classmethod
@@ -132,6 +149,10 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
             raise
 
     def sync_node(self, cloud_node, arvados_node):
+        # Update the cloud node record to ensure we have the correct metadata
+        # fingerprint.
+        cloud_node = self.real.ex_get_node(cloud_node.name, cloud_node.extra['zone'])
+
         # We can't store the FQDN on the name attribute or anything like it,
         # because (a) names are static throughout the node's life (so FQDN
         # isn't available because we don't know it at node creation time) and
@@ -143,12 +164,8 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
             self._find_metadata(metadata_items, 'hostname')['value'] = hostname
         except KeyError:
             metadata_items.append({'key': 'hostname', 'value': hostname})
-        response = self.real.connection.async_request(
-            '/zones/{}/instances/{}/setMetadata'.format(
-                cloud_node.extra['zone'].name, cloud_node.name),
-            method='POST', data=metadata_req)
-        if not response.success():
-            raise Exception("setMetadata error: {}".format(response.error))
+
+        self.real.ex_set_node_metadata(cloud_node, metadata_items)
 
     @classmethod
     def node_fqdn(cls, node):
@@ -163,3 +180,7 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
                     node.extra['metadata']['items'], 'booted_at'))
         except KeyError:
             return 0
+
+    @classmethod
+    def node_id(cls, node):
+        return node.id