From 6da1dbc8498b4c3143221fd4e1b9f59374134fd3 Mon Sep 17 00:00:00 2001
From: Michael Still <mikal@stillhq.com>
Date: Fri, 14 Dec 2012 10:22:44 +1100
Subject: Retry NBD device allocation.

Mitigates bug 726244.

Change-Id: I2f09f3a69c3eb61ca3af90355d5861373b7ae277
---
 nova/virt/disk/mount/nbd.py | 27 +++++++++++++++++++++++----
 1 file changed, 23 insertions(+), 4 deletions(-)

(limited to 'nova/virt')

diff --git a/nova/virt/disk/mount/nbd.py b/nova/virt/disk/mount/nbd.py
index 90d858a4b..1b6cc0778 100644
--- a/nova/virt/disk/mount/nbd.py
+++ b/nova/virt/disk/mount/nbd.py
@@ -37,15 +37,13 @@ CONF = cfg.CONF
 CONF.register_opts(nbd_opts)
 
 NBD_DEVICE_RE = re.compile('nbd[0-9]+')
+MAX_NBD_WAIT = 30
 
 
 class NbdMount(api.Mount):
     """qemu-nbd support disk images."""
     mode = 'nbd'
 
-    # NOTE(padraig): The remaining issue with this code is that multiple
-    # workers on a system can race against each other.
-
     def _detect_nbd_devices(self):
         """Detect nbd device files."""
         return filter(NBD_DEVICE_RE.match, os.listdir('/sys/block/'))
@@ -78,7 +76,7 @@ class NbdMount(api.Mount):
             pid = int(f.readline())
         return pid
 
-    def get_dev(self):
+    def _inner_get_dev(self):
         device = self._allocate_nbd()
         if not device:
             return False
@@ -102,12 +100,33 @@ class NbdMount(api.Mount):
                 break
             time.sleep(1)
         else:
+            _out, err = utils.trycmd('qemu-nbd', '-d', device,
+                                     run_as_root=True)
+            if err:
+                LOG.warn(_('Detaching from erroneous nbd device returned '
+                           'error: %s'), err)
             self.error = _('nbd device %s did not show up') % device
             return False
 
+        self.error = ''
         self.linked = True
         return True
 
+    def get_dev(self):
+        """Retry requests for NBD devices."""
+        start_time = time.time()
+        device = self._inner_get_dev()
+        while not device:
+            LOG.info(_('nbd device allocation failed. Will retry in 2 '
+                       'seconds.'))
+            time.sleep(2)
+            if time.time() - start_time > MAX_NBD_WAIT:
+                LOG.warn(_('nbd device allocation failed after repeated '
+                           'retries.'))
+                return False
+            device = self._inner_get_dev()
+        return True
+
     def unget_dev(self):
         if not self.linked:
             return
-- 
cgit