7286: Add comments clarifying arvados_node_missing() and broken(). Also bump

[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 4ef4e10bc90486710d6258012d44dedc715e7613..b1ec5e6abc95e923e1b6583d3c33e49a3247368d 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):
@@ -49,8 +51,12 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
         }
 
     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)
@@ -76,6 +82,12 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
                 super(ComputeNodeDriver, self).list_nodes()
                 if node.extra["tags"].get("arvados-class") == self.tags["arvados-class"]]
 
+    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):
         return node.extra["tags"].get("hostname")