LBAAS-843 - Add a mechanism to save failed load balancers for forensics

Change-Id: Iae6f9677801fba5d68cd42212111ef9c9b9cf22f

etc/libra.cfg

@@ -140,6 +140,12 @@ pid = /var/run/libra/libra_admin_api.pid
 #stats_enable = False
 #stats_purge_enable = False
 
+# These are the number of instances to save for forensic analysis on failure
+# A value of 0 disables saving any instances
+# Failed devices will have a status of "SAVED-OFFLINE" and "SAVED-ONLINE"
+#offline_failed_save = 0
+#online_failed_save = 0
+
 # The following are the seconds of each minute
 # that the timers will run.  The defaults should
 # not need to be changed..

libra/admin_api/__init__.py

@@ -139,6 +139,14 @@ cfg.CONF.register_opts(
         cfg.IntOpt('exists_timer_seconds',
                    default=55,
                    help='Second of each minute exists timer should run'),
-    ],
+        cfg.IntOpt('offline_failed_save',
+                   default=0,
+                   help='Number of failed offline instances to save '
+                        'for forensic analysis'),
+         cfg.IntOpt('online_failed_save',
+                   default=0,
+                   help='Number of failed online instances to save '
+                        'for forensic analysis'),
+     ],
     group=adminapi_group
 )

libra/admin_api/library/rebuild.py

@@ -17,6 +17,7 @@ from libra.common.api.lbaas import loadbalancers_devices, Vip, Counters
 from libra.common.api.lbaas import Device, LoadBalancer, db_session
 from libra.common.api.gearman_client import submit_job, submit_vip_job
 from libra.openstack.common import log
+from oslo.config import cfg
 
 
 LOG = log.getLogger(__name__)
@@ -25,6 +26,7 @@ LOG = log.getLogger(__name__)
 def rebuild_device(device_id):
     new_device_id = None
     new_device_name = None
+    ONLINE_FAILED_SAVE = cfg.CONF['admin_api'].online_failed_save
     with db_session() as session:
         new_device = session.query(Device).\
             filter(~Device.id.in_(
@@ -72,9 +74,14 @@ def rebuild_device(device_id):
         vip = session.query(Vip).filter(Vip.device == device_id).first()
         if vip:
             vip.device = new_device_id
+        saved_count = session.query(Device).\
+            filter(Device.status == 'SAVED-ONLINE').count()
         device = session.query(Device).\
             filter(Device.id == device_id).first()
-        device.status = 'DELETED'
+        if ONLINE_FAILED_SAVE > 0 and saved_count < ONLINE_FAILED_SAVE:
+            device.status = 'SAVED-ONLINE'
+        else:
+            device.status = 'DELETED'
         lbs = session.query(LoadBalancer).\
             join(LoadBalancer.devices).\
             filter(Device.id == new_device_id).all()

libra/admin_api/stats/drivers/database/driver.py

@@ -16,6 +16,7 @@ from libra.common.api.lbaas import Device, LoadBalancer, db_session
 from libra.common.api.lbaas import loadbalancers_devices
 from libra.admin_api.library.rebuild import rebuild_device
 from libra.openstack.common import log
+from oslo.config import cfg
 
 
 LOG = log.getLogger(__name__)
@@ -50,10 +51,19 @@ class DbDriver(AlertDriver):
             self._rebuild_device(device_id)
 
     def send_delete(self, message, device_id, device_ip, device_name):
+        OFFLINE_FAILED_SAVE = cfg.CONF['admin_api'].offline_failed_save
         with db_session() as session:
-            session.query(Device).\
-                filter(Device.id == device_id).\
-                update({"status": "DELETED"}, synchronize_session='fetch')
+            saved_count = session.query(Device).\
+                filter(Device.status == 'SAVED-OFFLINE').count()
+            if OFFLINE_FAILED_SAVE > 0 and saved_count < OFFLINE_FAILED_SAVE:
+                session.query(Device).\
+                    filter(Device.id == device_id).\
+                    update({"status": "SAVED-OFFLINE"},\
+                           synchronize_session='fetch')
+            else:
+                session.query(Device).\
+                    filter(Device.id == device_id).\
+                    update({"status": "DELETED"}, synchronize_session='fetch')
             session.commit()
 
     def send_node_change(self, message, lbid, degraded):