9353: Don't retry all BaseHTTPError, only retry InvalidInstanceID.NotFound.

[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 36bfc96213b9888df3a9afb7bd9c37fd6b76d4e1..b853f00a6728693cce4b855021e18bb35c869087 100644 (file)
--- a/services/nodemanager/arvnodeman/computenode/driver/gce.py
+++ b/services/nodemanager/arvnodeman/computenode/driver/gce.py
@@ -34,13 +34,10 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
         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')
@@ -63,9 +60,12 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
             self.create_kwargs['ex_metadata']['sshKeys'] = (
                 'root:' + ssh_file.read().strip())
 
-    def arvados_create_kwargs(self, arvados_node):
-        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)
         disks = [
             {'autoDelete': True,
              'boot': True,
@@ -100,12 +100,20 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
                 })
         return result
 
+
     def list_nodes(self):
         # The GCE libcloud driver only supports filtering node lists by zone.
         # Do our own filtering based on tag list.
-        return [node for node in
-                super(ComputeNodeDriver, self).list_nodes()
-                if self.node_tags.issubset(node.extra.get('tags', []))]
+        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]
+        return nodelist
 
     @classmethod
     def _find_metadata(cls, metadata_items, key):
@@ -158,3 +166,7 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
                     node.extra['metadata']['items'], 'booted_at'))
         except KeyError:
             return 0
+
+    @classmethod
+    def node_id(cls, node):
+        return node.id