X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/1afb514035257fa921f251ac1556a19c71b608f0..06cf38c24519fa21556686545c768429b5af50dc:/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 127ca20471..167d8b3210 100644
--- a/services/nodemanager/arvnodeman/computenode/driver/azure.py
+++ b/services/nodemanager/arvnodeman/computenode/driver/azure.py
@@ -2,11 +2,13 @@
 
 from __future__ import absolute_import, print_function
 
+import pipes
 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 +17,7 @@ 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):
@@ -35,7 +38,7 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
             auth_kwargs, list_kwargs, create_kwargs,
             driver_class)
 
-    def arvados_create_kwargs(self, arvados_node):
+    def arvados_create_kwargs(self, size, arvados_node):
         cluster_id, _, node_id = arvados_node['uuid'].split('-')
         name = 'compute-{}-{}'.format(node_id, cluster_id)
         tags = {
@@ -43,25 +46,29 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
             'arv-ping-url': self._make_ping_url(arvados_node)
         }
         tags.update(self.tags)
+
+        customdata = """#!/bin/sh
+mkdir -p    /var/tmp/arv-node-data/meta-data
+echo %s > /var/tmp/arv-node-data/arv-ping-url
+echo %s > /var/tmp/arv-node-data/meta-data/instance-id
+echo %s > /var/tmp/arv-node-data/meta-data/instance-type
+""" % (pipes.quote(tags['arv-ping-url']),
+       pipes.quote(name),
+       pipes.quote(size.id))
+
         return {
             'name': name,
             'ex_tags': tags,
+            'ex_customdata': customdata
         }
 
     def sync_node(self, cloud_node, arvados_node):
-        self.real.ex_create_tags(cloud_node,
-                                 {'hostname': arvados_node_fqdn(arvados_node)})
-        self.real.ex_run_command(cloud_node,
-                                 """bash -c '
-                                 mkdir -p /var/tmp/arv-node-data/meta-data
-                                 echo "%s" > /var/tmp/arv-node-data/arv-ping-url
-                                 echo "%s" > /var/tmp/arv-node-data/meta-data/instance-id
-                                 echo "%s" > /var/tmp/arv-node-data/meta-data/instance-type
-                                 echo "%s" > /var/tmp/arv-node-data/meta-data/local-ipv4
-                                 '""" % (self._make_ping_url(arvados_node),
-                                         cloud_node.id,
-                                         cloud_node.extra["properties"]["hardwareProfile"]["vmSize"],
-                                         cloud_node.private_ips[0]))
+        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)
@@ -69,9 +76,20 @@ 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
-                super(ComputeNodeDriver, self).list_nodes()
+        nodes = [node for node in
+                super(ComputeNodeDriver, self).list_nodes(ex_fetch_nic=False)
                 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):
@@ -80,3 +98,7 @@ class ComputeNodeDriver(BaseComputeNodeDriver):
     @classmethod
     def node_start_time(cls, node):
         return arvados_timestamp(node.extra["tags"].get("booted_at"))
+
+    @classmethod
+    def node_id(cls, node):
+        return node.name