Fix record extraction and do not request only deleted records

Commit 0058b6c introduced a bug in how
we retrieve the records, since we introduced the "deleted" filter and
this was filtering out all the instances that are not deleted.

Moreover, the way we were retrieving the results was wrong as only
servers with a changed status within the requested period were
considered for the record building.

We cannot use 'changes-since' in the servers.list() API query, as it
will only include changes that have chanted its status after that date.
Also, we cannot just get all the usages and then query the servers.get()
server by server, as deleted servers are not returned by ths
servers.get() call. What we do now is the following.

    1.- List all the deleted servers that changed after the start date
    2.- Build the records for the period [start, end]
    3.- Get all the usages
    4.- Iter over the usages and:
    4.1.- get information for non deleted servers
    4.2.- do nothing with deleted servers, as we collected in in step (2)

However, usage was eventually correctly reported, as sooner or later one
server may have a changed status and we will generate the record.

In order to get accurate results, accounting records must be regenerated
at sites, possibly month by month, like:

    caso-extract  --extract-from 2017-01-01 --extract-to 2017-01-31
    caso-extract  --extract-from 2017-02-01 --extract-to 2017-02-28
    (...)

Closes #47

caso/extract/nova.py

@@ -119,10 +119,29 @@ def extract_for_project(self, project, extract_from, extract_to):
             flavors[flavor.id] = flavor.to_dict()
             flavors[flavor.id]["extra"] = flavor.get_keys()
 
+        images = {image.id: image for image in glance.images.list()}
+        records = {}
+
+        vo = self.voms_map.get(project)
+
+        # We cannot use 'changes-since' in the servers.list() API query, as it
+        # will only include changes that have changed its status after that
+        # date. However we cannot just get all the usages and then query
+        # server by server, as deleted servers are not returned  by this
+        # servers.get() call. What we do is the following:
+        #
+        # 1.- List all the deleted servers that changed after the start date
+        # 2.- Build the records for the period [start, end]
+        # 3.- Get all the usages
+        # 4.- Iter over the usages and:
+        # 4.1.- get information for non deleted servers
+        # 4.2.- do nothing with deleted servers, as we collected in in step (2)
+
+        # 1.- List all the deleted servers from that period.
         servers = []
         limit = 200
         marker = None
-        # Iter over results until we do not have more to get
+        # Use a marker and iter over results until we do not have more to get
         while True:
             aux = nova.servers.list(
                 search_opts={"changes-since": extract_from,
@@ -138,35 +157,41 @@ def extract_for_project(self, project, extract_from, extract_to):
 
         servers = sorted(servers, key=operator.attrgetter("created"))
 
-        # We are going to query for the sever usages below. It will return the
-        # usages for the specified period. However, usages should be absolute
-        # values, not deltas. Therefore we need to change the start time for
-        # the query with that of the oldest server.
-        if servers:
-            start = dateutil.parser.parse(servers[0].created)
-            start = start.replace(tzinfo=None)
-        else:
-            start = extract_from
-
-        aux = nova.usage.get(project_id, start, extract_to)
-        usages = getattr(aux, "server_usages", [])
-
-        images = {image.id: image for image in glance.images.list()}
-        records = {}
-
-        vo = self.voms_map.get(project)
-
+        # 2.- Build the records for the period. Drop servers outside the period
+        # (we do this manually as we cannot limit the query to a period, only
+        # changes after start date).
         for server in servers:
             server_start = dateutil.parser.parse(server.created)
             server_start = server_start.replace(tzinfo=None)
-            if server_start > extract_to:
+            # Some servers may be deleted before 'extract_from' but updated
+            # afterwards
+            server_end = server.__getattr__('OS-SRV-USG:terminated_at')
+            if server_end:
+                server_end = dateutil.parser.parse(server_end)
+                server_end = server_end.replace(tzinfo=None)
+            if (server_start > extract_to or
+                    (server_end and server_end < extract_from)):
                 continue
             records[server.id] = self.build_record(server, vo, images,
                                                    flavors, users)
 
+        # 3.- Get all the usages for the period
+        start = extract_from
+        aux = nova.usage.get(project_id, start, extract_to)
+        usages = getattr(aux, "server_usages", [])
+
+        # 4.- Iter over the results and
         for usage in usages:
+            # 4.1 and 4.2 Get the server if it is not yet there
             if usage["instance_id"] not in records:
-                continue
+                server = nova.servers.get(usage["instance_id"])
+
+                server_start = dateutil.parser.parse(server.created)
+                server_start = server_start.replace(tzinfo=None)
+                if server_start > extract_to:
+                    continue
+                records[server.id] = self.build_record(server, vo, images,
+                                                       flavors, users)
             instance_id = usage["instance_id"]
             records[instance_id].memory = usage["memory_mb"]
             records[instance_id].cpu_count = usage["vcpus"]
@@ -185,10 +210,10 @@ def extract_for_project(self, project, extract_from, extract_to):
                 ended = None
 
             # Wall and CPU durations are absolute values, not deltas for the
-            # reporting period. The nova API only gives use the "changes-since"
-            # usages, therefore we need to calculate the wall duration by
-            # ourselves, then multiply by the nr of CPUs to get the CPU
-            # duration.
+            # reporting period. The nova API only gives use the usages for the
+            # requested period, therefore we need to calculate the wall
+            # duration by ourselves, then multiply by the nr of CPUs to get the
+            # CPU duration.
 
             # If the machine has not ended, report consumption until
             # extract_to, otherwise get its consuption by substracting ended -

caso/tests/extract/test_nova.py

@@ -0,0 +1,28 @@
+# -*- coding: utf-8 -*-
+
+# Licensed under the Apache License, Version 2.0 (the "License"); you may
+# not use this file except in compliance with the License. You may obtain
+# a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+# License for the specific language governing permissions and limitations
+# under the License.
+
+from caso.extract import nova
+from caso.tests import base
+
+
+class TestCasoManager(base.TestCase):
+    def setUp(self):
+        super(TestCasoManager, self).setUp()
+        self.flags(mapping_file="etc/caso/voms.json.sample")
+        self.extractor = nova.OpenStackExtractor()
+
+    def tearDown(self):
+        self.reset_flags()
+
+        super(TestCasoManager, self).tearDown()