From f3845095c7f8802c4dd40d7679e14831d7f70b59 Mon Sep 17 00:00:00 2001
From: Nikos Skalkotos <skalkoto@grnet.gr>
Date: Mon, 31 Dec 2012 00:37:07 +0200
Subject: [PATCH] Retry cleanup commands if they fail

Add a new try_fail_repeat function that wraps cleanup system commands
and retries them if they fail. There are cases where commands like
dmsetup, umount and losetup -d may fail with device is busy. In most
of the cases, sleeping for halve a second and retrying does the job.
---
 image_creator/bundle_volume.py |   14 +++++++++-----
 image_creator/disk.py          |   13 +++++--------
 image_creator/util.py          |   20 +++++++++++++++++++-
 3 files changed, 33 insertions(+), 14 deletions(-)

diff --git a/image_creator/bundle_volume.py b/image_creator/bundle_volume.py
index 27839fa..bd418ce 100644
--- a/image_creator/bundle_volume.py
+++ b/image_creator/bundle_volume.py
@@ -34,7 +34,6 @@
 import os
 import re
 import tempfile
-import time
 from collections import namedtuple
 
 import parted
@@ -42,6 +41,7 @@ import parted
 from image_creator.rsync import Rsync
 from image_creator.util import get_command
 from image_creator.util import FatalError
+from image_creator.util import try_fail_repeat
 
 findfs = get_command('findfs')
 dd = get_command('dd')
@@ -66,9 +66,11 @@ MKFS_OPTS = {
     }
 
 
-class BundleVolume():
+class BundleVolume(object):
+    """This class can be used to create an image out of the running system"""
 
     def __init__(self, out, meta):
+        """Create an instance of the BundleVolume class."""
         self.out = out
         self.meta = meta
 
@@ -256,8 +258,7 @@ class BundleVolume():
         if not os.path.exists(dev):
             return
 
-        dmsetup('remove', dev.split('/dev/mapper/')[1])
-        time.sleep(0.1)
+        try_fail_repeat(dmsetup, 'remove', dev.split('/dev/mapper/')[1])
 
     def _mount(self, target, devs):
 
@@ -276,7 +277,7 @@ class BundleVolume():
 
         mpoints.sort()
         for mpoint in reversed(mpoints):
-            umount(mpoint)
+            try_fail_repeat(umount, mpoint)
 
     def _to_exclude(self):
         excluded = ['/tmp', '/var/tmp']
@@ -391,6 +392,9 @@ class BundleVolume():
             losetup('-d', loop)
 
     def create_image(self, image):
+        """Given an image filename, this method will create an image out of the
+        running system.
+        """
 
         size = self.disk.device.getLength() * self.disk.device.sectorSize
 
diff --git a/image_creator/disk.py b/image_creator/disk.py
index 8a06f78..bb90816 100644
--- a/image_creator/disk.py
+++ b/image_creator/disk.py
@@ -33,6 +33,7 @@
 
 from image_creator.util import get_command
 from image_creator.util import FatalError
+from image_creator.util import try_fail_repeat
 from image_creator.gpt import GPTPartitionTable
 from image_creator.bundle_volume import BundleVolume
 
@@ -43,7 +44,6 @@ import uuid
 import re
 import sys
 import guestfs
-import time
 from sendfile import sendfile
 
 
@@ -63,7 +63,8 @@ class Disk(object):
 
     def __init__(self, source, output):
         """Create a new Disk instance out of a source media. The source
-        media can be an image file, a block device or a directory."""
+        media can be an image file, a block device or a directory.
+        """
         self._cleanup_jobs = []
         self._devices = []
         self.source = source
@@ -76,7 +77,7 @@ class Disk(object):
     def _losetup(self, fname):
         loop = losetup('-f', '--show', fname)
         loop = loop.strip()  # remove the new-line char
-        self._add_cleanup(losetup, '-d', loop)
+        self._add_cleanup(try_fail_repeat, losetup, '-d', loop)
         return loop
 
     def _dir_to_disk(self):
@@ -144,11 +145,7 @@ class Disk(object):
             os.write(tablefd, "0 %d snapshot %s %s n 8" %
                               (int(size), sourcedev, cowdev))
             dmsetup('create', snapshot, table)
-            self._add_cleanup(dmsetup, 'remove', snapshot)
-            # Sometimes dmsetup remove fails with Device or resource busy,
-            # although everything is cleaned up and the snapshot is not
-            # used by anyone. Add a 2 seconds delay to be on the safe side.
-            self._add_cleanup(time.sleep, 2)
+            self._add_cleanup(try_fail_repeat, dmsetup, 'remove', snapshot)
 
         finally:
             os.unlink(table)
diff --git a/image_creator/util.py b/image_creator/util.py
index 61cc2a6..1cd97ff 100644
--- a/image_creator/util.py
+++ b/image_creator/util.py
@@ -34,6 +34,7 @@
 import sys
 import sh
 import hashlib
+import time
 
 
 class FatalError(Exception):
@@ -54,12 +55,29 @@ def get_command(command):
         return find_sbin_command(command, e)
 
 
+def try_fail_repeat(command, *args):
+
+    times = (0.1, 0.5, 1, 2)
+    i = iter(times)
+    while True:
+        try:
+            command(*args)
+            return
+        except sh.ErrorReturnCode:
+            try:
+                wait = i.next()
+            except StopIteration:
+                break
+            time.sleep(wait)
+
+    raise FatalError("Command: `%s %s' failed" % (command, " ".join(args)))
+
+
 class MD5:
     def __init__(self, output):
         self.out = output
 
     def compute(self, filename, size):
-
         MB = 2 ** 20
         BLOCKSIZE = 4 * MB  # 4MB
 
-- 
1.7.10.4