Merge branch 'master' into 13822-nm-delayed-daemon

[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 db7b162b324766c224c9770988c206211179d39f..11025f7840bc00fe6c188ad6b0f9e9bea1795cba 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
 
@@ -31,24 +34,20 @@ 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._disktype_links = {dt.name: self._object_link(dt)
                                 for dt in self.real.ex_list_disktypes()}
 
-    @staticmethod
-    def _name_key(cloud_object):
-        return cloud_object.name
-
     @staticmethod
     def _object_link(cloud_object):
         return cloud_object.extra.get('selfLink')
 
     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(
@@ -64,9 +63,16 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
             self.create_kwargs['ex_metadata']['sshKeys'] = (
                 'root:' + ssh_file.read().strip())
 
-    def arvados_create_kwargs(self, arvados_node, size):
-        cluster_id, _, node_id = arvados_node['uuid'].split('-')
-        name = 'compute-{}-{}'.format(node_id, cluster_id)
+    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):
+        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,
@@ -95,10 +101,11 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
                   'ex_disks_gce_struct': disks,
                   }
         result['ex_metadata'].update({
-                'arv-ping-url': self._make_ping_url(arvados_node),
-                'booted_at': time.strftime(ARVADOS_TIMEFMT, time.gmtime()),
-                'hostname': arvados_node_fqdn(arvados_node),
-                })
+            'arvados_node_size': size.id,
+            'arv-ping-url': self._make_ping_url(arvados_node),
+            'booted_at': time.strftime(ARVADOS_TIMEFMT, time.gmtime()),
+            'hostname': arvados_node_fqdn(arvados_node),
+        })
         return result
 
     def list_nodes(self):
@@ -107,12 +114,14 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
         nodelist = [node for node in
                     super(ComputeNodeDriver, self).list_nodes()
                     if self.node_tags.issubset(node.extra.get('tags', []))]
-        # As of 0.18, the libcloud GCE driver sets node.size to the size's name.
-        # It's supposed to be the actual size object.  Check that it's not,
-        # 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]
+        for node in nodelist:
+            # As of 0.18, the libcloud GCE driver sets node.size to the size's name.
+            # It's supposed to be the actual size object.  Check that it's not,
+            # and monkeypatch the results when that's the case.
+            if not hasattr(node.size, 'id'):
+                node.size = self.sizes()[node.size]
+            # Get arvados-assigned cloud size id
+            node.extra['arvados_node_size'] = node.extra.get('metadata', {}).get('arvados_node_size')
         return nodelist
 
     @classmethod
@@ -135,6 +144,10 @@ 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
@@ -146,12 +159,8 @@ 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):
@@ -166,3 +175,7 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
                     node.extra['metadata']['items'], 'booted_at'))
         except KeyError:
             return 0
+
+    @classmethod
+    def node_id(cls, node):
+        return node.id