Merge branch 'master' into 7454-azure-custom-data

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

diff --git a/services/nodemanager/arvnodeman/computenode/driver/azure.py b/services/nodemanager/arvnodeman/computenode/driver/azure.py
index f4686b976f125bb6d7c41084f65dc2c59f401ce1..20988c2660d8c687e5541c4b41270becafa49fab 100644 (file)
--- a/services/nodemanager/arvnodeman/computenode/driver/azure.py
+++ b/services/nodemanager/arvnodeman/computenode/driver/azure.py
@@ -7,6 +7,7 @@ import time
 import libcloud.compute.base as cloud_base
 import libcloud.compute.providers as cloud_provider
 import libcloud.compute.types as cloud_types
+from libcloud.common.exceptions import BaseHTTPError
 
 from . import BaseComputeNodeDriver
 from .. import arvados_node_fqdn, arvados_timestamp, ARVADOS_TIMEFMT
@@ -15,6 +16,7 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
 
     DEFAULT_DRIVER = cloud_provider.get_driver(cloud_types.Provider.AZURE_ARM)
     SEARCH_CACHE = {}
+    CLOUD_ERRORS = BaseComputeNodeDriver.CLOUD_ERRORS + (BaseHTTPError,)
 
     def __init__(self, auth_kwargs, list_kwargs, create_kwargs,
                  driver_class=DEFAULT_DRIVER):
@@ -27,6 +29,10 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
         self.tags = {key[4:]: value
                      for key, value in create_kwargs.iteritems()
                      if key.startswith('tag_')}
+        # filter out tags from create_kwargs
+        create_kwargs = {key: value
+                         for key, value in create_kwargs.iteritems()
+                         if not key.startswith('tag_')}
         super(ComputeNodeDriver, self).__init__(
             auth_kwargs, list_kwargs, create_kwargs,
             driver_class)
@@ -42,11 +48,16 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
         return {
             'name': name,
             'ex_tags': tags,
+            'ex_customdata': tags['arv-ping-url']
         }
 
     def sync_node(self, cloud_node, arvados_node):
-        self.real.ex_create_tags(cloud_node,
-                                 {'hostname': arvados_node_fqdn(arvados_node)})
+        try:
+            self.real.ex_create_tags(cloud_node,
+                                     {'hostname': arvados_node_fqdn(arvados_node)})
+            return True
+        except BaseHTTPError as b:
+            return False
 
     def _init_image(self, urn):
         return "image", self.get_image(urn)
@@ -54,9 +65,20 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
     def list_nodes(self):
         # Azure only supports filtering node lists by resource group.
         # Do our own filtering based on tag.
-        return [node for node in
+        nodes = [node for node in
                 super(ComputeNodeDriver, self).list_nodes()
                 if node.extra["tags"].get("arvados-class") == self.tags["arvados-class"]]
+        for n in nodes:
+            # Need to populate Node.size
+            if not n.size:
+                n.size = self.sizes[n.extra["properties"]["hardwareProfile"]["vmSize"]]
+        return nodes
+
+    def broken(self, cloud_node):
+        """Return true if libcloud has indicated the node is in a "broken" state."""
+        # UNKNOWN means the node state is unrecognized, which in practice means some combination
+        # of failure that the Azure libcloud driver doesn't know how to interpret.
+        return (cloud_node.state in (cloud_types.NodeState.ERROR, cloud_types.NodeState.UNKNOWN))
 
     @classmethod
     def node_fqdn(cls, node):