stackforge
/
novaimagebuilder
Code Issues Proposed changes

initial locking implementation 

This adds both thread level and process level locking to the cache index.
It also adds a "pending" state to cached object collections, allowing a
single instance of the CacheManager to do long-running downloads without
holding a lock on the index the entire time.

Change-Id: I6a0fcf0941f75fa5d78cc06eb78af6531d712162
This commit is contained in:
Ian McLeod 2013-10-30 09:25:28 -05:00
parent 17f95d3c7d
commit 65e06f4603
3 changed files with 150 additions and 38 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

112
novaimagebuilder/CacheManager.py
View File

@ -16,11 +16,15 @@
import logging
import json
import os
import os.path
import pycurl
import guestfs
import fcntl
import threading
import time
import StackEnvironment
from Singleton import Singleton
from StackEnvironment import StackEnvironment
class CacheManager(Singleton):
@ -40,11 +44,11 @@ class CacheManager(Singleton):
# TODO: Sane handling of a pending cache item
# TODO: Configurable
CACHE_ROOT = "/var/lib/novaimagebuilder/"
#INDEX_LOCK = lock()
INDEX_THREAD_LOCK = threading.Lock()
INDEX_FILE = "_cache_index"
def _singleton_init(self):
self.env = StackEnvironment()
self.env = StackEnvironment.StackEnvironment()
self.log = logging.getLogger('%s.%s' % (__name__, self.__class__.__name__))
self.index_filename = self.CACHE_ROOT + self.INDEX_FILE
if not os.path.isfile(self.index_filename):
@ -55,6 +59,8 @@ class CacheManager(Singleton):
index_file.close()
# This should be None except when we are actively working on it and hold a lock
self.index = None
self.index_file = None
self.locked = False
def lock_and_get_index(self):
"""
@ -64,30 +70,46 @@ class CacheManager(Singleton):
write_index_and_unlock() or unlock_index() depending upon whether or not the
index has been modified.
"""
#self.INDEX_LOCK.acquire()
index_file = open(self.index_filename)
self.index = json.load(index_file)
index_file.close()
# We acquire a thread lock under all circumstances
# This is the safest approach and should be relatively harmless if we are used
# as a module in a non-threaded Python program
self.INDEX_THREAD_LOCK.acquire()
# atomic create if not present
fd = os.open(self.index_filename, os.O_RDWR | os.O_CREAT)
# blocking
fcntl.flock(fd, fcntl.LOCK_EX)
self.index_file = os.fdopen(fd, "r+")
index = self.index_file.read()
if len(index) == 0:
# Empty - possibly because we created it earlier - create empty dict
self.index = { }
else:
self.index = json.loads(index)
def write_index_and_unlock(self):
"""
Write contents of self.index back to the persistent file and then unlock it
"""
index_file = open(self.index_filename, 'w')
json.dump(self.index , index_file)
index_file.close()
self.index_file.seek(0)
self.index_file.truncate()
json.dump(self.index , self.index_file)
# TODO: Double-check that this is safe
self.index_file.flush()
fcntl.flock(self.index_file, fcntl.LOCK_UN)
self.index_file.close()
self.index = None
#self.INDEX_LOCK.release()
self.INDEX_THREAD_LOCK.release()
def unlock_index(self):
"""
Release the cache index lock without updating the persistent file
"""
self.index = None
#self.INDEX_LOCK.release()
fcntl.flock(self.index_file, fcntl.LOCK_UN)
self.index_file.close()
self.index_file = None
self.INDEX_THREAD_LOCK.release()
# INDEX looks like
#
@ -95,6 +117,11 @@ class CacheManager(Singleton):
# "install_iso_kernel": { "local"
def _get_index_value(self, os_ver_arch, name, location):
"""
Utility function to retrieve the location of the named object for the given OS version and architecture.
Only use this if your thread has obtained the thread-global lock by using the
lock_and_get_index() function above
"""
if self.index is None:
raise Exception("Attempt made to read index values while a locked index is not present")
@ -114,6 +141,11 @@ class CacheManager(Singleton):
return self.index[os_ver_arch][name][location]
def _set_index_value(self, os_ver_arch, name, location, value):
"""
Utility function to set the location of the named object for the given OS version and architecture.
Only use this if your thread has obtained the thread-global lock by using the
lock_and_get_index() function above
"""
if self.index is None:
raise Exception("Attempt made to read index values while a locked index is not present")
@ -124,9 +156,8 @@ class CacheManager(Singleton):
self.index[os_ver_arch][name] = {}
# If the specific location is not specified, assume value is the entire dict
# or a string indicating the object is pending
if not location:
if type(value) is not dict:
raise Exception("When setting a value without a location, the value must be a dict")
self.index[os_ver_arch][name] = value
return
@ -150,19 +181,42 @@ class CacheManager(Singleton):
glance: Glance object UUID
cinder: Cinder object UUID
"""
# TODO: Gracefully deal with the situation where, for example, we are asked to save_local
# and find that the object is already cached but only exists in glance and/or cinder
# TODO: Allow for local-only caching
self.lock_and_get_index()
existing_cache = self._get_index_value(os_plugin.os_ver_arch(), object_type, None)
if existing_cache:
self.log.debug("Found object in cache")
self.unlock_index()
return existing_cache
# TODO: special case when object is ISO and sub-artifacts are not cached
pending_countdown = 360
while True:
self.lock_and_get_index()
existing_cache = self._get_index_value(os_plugin.os_ver_arch(), object_type, None)
if existing_cache == None:
# We are the first - mark as pending and then start to retreive
self._set_index_value(os_plugin.os_ver_arch(), object_type, None, "pending")
self.write_index_and_unlock()
break
if isinstance(existing_cache, dict):
self.log.debug("Found object in cache")
self.unlock_index()
return existing_cache
# TODO: special case when object is ISO and sub-artifacts are not cached
if existing_cache == "pending":
# Another thread or process is currently obtaining this object
# poll every 10 seconds until we get a dict, then return it
# TODO: A graceful event based solution
self.unlock_index()
if pending_countdown == 360:
self.log.debug("Object is being retrieved in another thread or process - Waiting")
pending_countdown -= 1
if pending_countdown == 0:
raise Exception("Waited one hour on pending cache fill for version (%s) - object (%s)- giving up" %
( os_plugin.os_ver_arch(), object_type ) )
sleep(10)
continue
# The object is not yet in the cache
# TODO: Some mechanism to indicate that retrieval is in progress
# additional calls to get the same object should block until this is done
self.unlock_index()
# We should never get here
raise Exception("Got unexpected non-string, non-dict, non-None value when reading cache")
# If we have gotten here the object is not yet in the cache
self.log.debug("Object not in cache")
# TODO: If not save_local and the plugin doesn't need the iso, direct download in glance
@ -245,4 +299,4 @@ class CacheManager(Singleton):
c.perform()
c.close()
finally:
os.close(fd)
os.close(fd)

4
tests/MockStackEnvironment.py
View File

@ -41,7 +41,7 @@ class MockStackEnvironment(Singleton):
IMAGE_STATUS_LIST = ('QUEUED', 'SAVING', 'ACTIVE', 'KILLED', 'DELETED', 'PENDING_DELETE')
def _singleton_init(self):
super(MockStackEnvironment, self)._singleton_init()
super(StackEnvironment, self)._singleton_init()
self.log = logging.getLogger('%s.%s' % (__name__, self.__class__.__name__))
# Attributes controlling Mock behavior
self.cinder = False
@ -110,4 +110,4 @@ class MockStackEnvironment(Singleton):
def launch_instance(self, root_disk=None, install_iso=None, secondary_iso=None, floppy=None, aki=None, ari=None,
cmdline=None, userdata=None):
return MockNovaInstance(object(), self)
return MockNovaInstance(object(), self)

72
tests/testcache.py
View File

@ -16,15 +16,24 @@
# limitations under the License.
import sys
sys.path.append("../novaimagebuilder")
from MockStackEnvironment import MockStackEnvironment as StackEnvironment
from novaimagebuilder.CacheManager import CacheManager
sys.path.append("../")
import MockStackEnvironment
sys.modules['StackEnvironment'] = sys.modules.pop('MockStackEnvironment')
sys.modules['StackEnvironment'].StackEnvironment = sys.modules['StackEnvironment'].MockStackEnvironment
import StackEnvironment
import novaimagebuilder.CacheManager
novaimagebuilder.CacheManager.StackEnvironment = StackEnvironment
import logging
import threading
import multiprocessing
logging.basicConfig(level=logging.DEBUG,
format='%(asctime)s %(levelname)s %(name)s thread(%(threadName)s) Message: %(message)s')
se = StackEnvironment.StackEnvironment()
class MockOSPlugin(object):
def __init__(self, os_ver_arch = "fedora19-x86_64", wants_iso = True ):
@ -40,8 +49,57 @@ class MockOSPlugin(object):
print "---- the following should do a glance and cinder upload"
mosp = MockOSPlugin(os_ver_arch = "fedora18-x86_64", wants_iso = False)
mse = StackEnvironment("username","password","tenant","auth_url")
cm = CacheManager(mse)
#mse = StackEnvironment("username","password","tenant","auth_url")
mse = StackEnvironment.StackEnvironment()
cm = novaimagebuilder.CacheManager.CacheManager()
cm.retrieve_and_cache_object("install-iso", mosp, "http://repos.fedorapeople.org/repos/aeolus/imagefactory/testing/repos/rhel/imagefactory.repo",
True)
# Create our bogus entry in the cache index and set it to 0
cm.lock_and_get_index()
cm._set_index_value("testobjOS", "testobjname", "testloc", "0")
cm.write_index_and_unlock()
class UpdateThread():
def __call__(self):
#print "about to run 20 updates"
for i in range(0,20):
cm.lock_and_get_index()
#print "--------- three lines below"
#print "In the lock - 1 next line should always show value"
value = cm._get_index_value("testobjOS", "testobjname", "testloc")
#print "In the lock - 2 value %s" % (value)
newvalue = int(value) + 1
cm._set_index_value("testobjOS", "testobjname", "testloc", str(newvalue))
#print "In the lock - 3 did update - leaving"
#print "--------- three lines above"
cm.write_index_and_unlock()
class MultiThreadProcess():
def __call__(self):
#print "Here I run 20 threads"
threads = [ ]
for i in range (0,20):
thread = threading.Thread(group=None, target=UpdateThread())
threads.append(thread)
thread.run()
# Fork 20 copies of myself
processes = [ ]
for i in range(0,20):
proc = multiprocessing.Process(group=None, target=MultiThreadProcess())
processes.append(proc)
proc.start()
for proc in processes:
proc.join()
cm.lock_and_get_index()
value = cm._get_index_value("testobjOS", "testobjname", "testloc")
cm.unlock_index()
print "Final value should be 8000 and is %s" % (value)
# Have each process create 20 threads
# Have each
#cm.retrieve_and_cache_object("install-iso2", mosp, "http://repos.fedorapeople.org/repos/aeolus/imagefactory/testing/repos/rhel/imagefactory.repo",
# True)