Commit 658f2a08 authored by Chris McDonough's avatar Chris McDonough

dd Michael Dunstan's "explicit abort after error propagates into publisher"...

dd Michael Dunstan's "explicit abort after error propagates into publisher" patch as documented in http://zope.org/Collectors/Zope/789 and as required by recent changes to ZODB which prevent a connection from being cleanly closed if modifications are extant in that connection.

(Merge from 2.7 branch)
parent ef68907c
...@@ -136,35 +136,30 @@ def publish(request, module_name, after_list, debug=0, ...@@ -136,35 +136,30 @@ def publish(request, module_name, after_list, debug=0,
if parents: if parents:
parents=parents[0] parents=parents[0]
try: try:
response = err_hook(parents, request, try:
return err_hook(parents, request,
sys.exc_info()[0], sys.exc_info()[0],
sys.exc_info()[1], sys.exc_info()[1],
sys.exc_info()[2], sys.exc_info()[2],
) )
if transactions_manager: except Retry:
transactions_manager.abort() if not request.supports_retry():
return response return err_hook(parents, request,
except Retry:
if not request.supports_retry():
response = err_hook(parents, request,
sys.exc_info()[0], sys.exc_info()[0],
sys.exc_info()[1], sys.exc_info()[1],
sys.exc_info()[2], sys.exc_info()[2],
) )
if transactions_manager: finally:
transactions_manager.abort()
return response
if transactions_manager: if transactions_manager:
transactions_manager.abort() transactions_manager.abort()
newrequest=request.retry()
request.close() # Free resources held by the request. # Only reachable if Retry is raised and request supports retry.
try: newrequest=request.retry()
return publish(newrequest, module_name, after_list, debug) request.close() # Free resources held by the request.
finally: try:
newrequest.close() return publish(newrequest, module_name, after_list, debug)
finally:
newrequest.close()
else: else:
if transactions_manager: if transactions_manager:
......
from ZPublisher import Retry
from ZODB.POSException import ConflictError
class Tracer:
"""Trace used to record pathway taken through the publisher
machinery. And provide framework for spewing out exceptions at
just the right time.
"""
def __init__(self):
self.reset()
def reset(self):
self.tracedPath = []
self.exceptions = {}
def append(self, arg):
self.tracedPath.append(arg)
def showTracedPath(self):
for arg in self.tracedPath:
print arg
def possiblyRaiseException(self, context):
exceptions = tracer.exceptions.get(context, None)
if exceptions:
exception = exceptions[0]
exceptions.remove(exception)
exceptionShortName = str(exception).split('.')[-1]
self.append('raising %s from %s' % (exceptionShortName, context))
raise exception
tracer = Tracer()
class TransactionsManager:
"""Mock TransactionManager to replace
Zope.App.startup.TransactionsManager.
"""
def abort(self):
tracer.append('abort')
def begin(self):
tracer.append('begin')
def commit(self):
tracer.append('commit')
tracer.possiblyRaiseException('commit')
def recordMetaData(self, obj, request):
pass
zpublisher_transactions_manager = TransactionsManager()
def zpublisher_exception_hook(published, request, t, v, traceback):
"""Mock zpublisher_exception_hook to replace
Zope.App.startup.zpublisher_exception_hook
"""
if issubclass(t, ConflictError):
raise Retry(t, v, traceback)
if t is Retry:
v.reraise()
tracer.append('zpublisher_exception_hook')
tracer.possiblyRaiseException('zpublisher_exception_hook')
return 'zpublisher_exception_hook'
class Object:
"""Mock object for traversing to.
"""
def __call__(self):
tracer.append('__call__')
tracer.possiblyRaiseException('__call__')
return '__call__'
class Response:
"""Mock Response to replace ZPublisher.HTTPResponse.HTTPResponse.
"""
def setBody(self, a):
pass
class Request:
"""Mock Request to replace ZPublisher.HTTPRequest.HTTPRequest.
"""
args = ()
def __init__(self):
self.response = Response()
def processInputs(self):
pass
def get(self, a, b=''):
return ''
def __setitem__(self, name, value):
pass
def traverse(self, path, validated_hook):
return Object()
def close(self):
pass
retry_count = 0
retry_max_count = 3
def supports_retry(self):
return self.retry_count < self.retry_max_count
def retry(self):
self.retry_count += 1
r = self.__class__()
r.retry_count = self.retry_count
return r
module_name = __name__
after_list = [None]
def testPublisher():
"""
Tests to ensure that the ZPublisher correctly manages the ZODB
transaction boundaries.
>>> from ZPublisher.Publish import publish
ZPublisher will commit the transaction after it has made a
rendering of the object.
>>> tracer.reset()
>>> request = Request()
>>> response = publish(request, module_name, after_list)
>>> tracer.showTracedPath()
begin
__call__
commit
If ZPublisher sees an exception when rendering the requested
object then it will try rendering an error message. The
transaction is eventually aborted after rendering the error
message. (Note that this handling of the transaction boundaries is
different to how Zope3 does things. Zope3 aborts the transaction
before rendering the error message.)
>>> tracer.reset()
>>> tracer.exceptions['__call__'] = [ValueError]
>>> request = Request()
>>> response = publish(request, module_name, after_list)
>>> tracer.showTracedPath()
begin
__call__
raising ValueError from __call__
zpublisher_exception_hook
abort
If there is a futher exception raised while trying to render the
error then ZPublisher is still required to abort the
transaction. And the exception propagates out of publish.
>>> tracer.reset()
>>> tracer.exceptions['__call__'] = [ValueError]
>>> tracer.exceptions['zpublisher_exception_hook'] = [ValueError]
>>> request = Request()
>>> response = publish(request, module_name, after_list)
Traceback (most recent call last):
...
ValueError
>>> tracer.showTracedPath()
begin
__call__
raising ValueError from __call__
zpublisher_exception_hook
raising ValueError from zpublisher_exception_hook
abort
ZPublisher can also deal with database ConflictErrors. The original
transaction is aborted and a second is made in which the request
is attempted again. (There is a fair amount of collaboration to
implement the retry functionality. Relies on Request and
zpublisher_exception_hook also doing the right thing.)
>>> tracer.reset()
>>> tracer.exceptions['__call__'] = [ConflictError]
>>> request = Request()
>>> response = publish(request, module_name, after_list)
>>> tracer.showTracedPath()
begin
__call__
raising ConflictError from __call__
abort
begin
__call__
commit
Same behaviour if there is a conflict when attempting to commit
the transaction. (Again this relies on collaboration from
zpublisher_exception_hook.)
>>> tracer.reset()
>>> tracer.exceptions['commit'] = [ConflictError]
>>> request = Request()
>>> response = publish(request, module_name, after_list)
>>> tracer.showTracedPath()
begin
__call__
commit
raising ConflictError from commit
abort
begin
__call__
commit
ZPublisher will retry the request several times. After 3 retries it
gives up and the exception propogates out.
>>> tracer.reset()
>>> tracer.exceptions['__call__'] = [ConflictError, ConflictError,
... ConflictError, ConflictError]
>>> request = Request()
>>> response = publish(request, module_name, after_list)
Traceback (most recent call last):
...
ConflictError: database conflict error
>>> tracer.showTracedPath()
begin
__call__
raising ConflictError from __call__
abort
begin
__call__
raising ConflictError from __call__
abort
begin
__call__
raising ConflictError from __call__
abort
begin
__call__
raising ConflictError from __call__
abort
However ZPublisher does not retry ConflictErrors that are raised
while trying to render an error message.
>>> tracer.reset()
>>> tracer.exceptions['__call__'] = [ValueError]
>>> tracer.exceptions['zpublisher_exception_hook'] = [ConflictError]
>>> request = Request()
>>> response = publish(request, module_name, after_list)
Traceback (most recent call last):
...
ConflictError: database conflict error
>>> tracer.showTracedPath()
begin
__call__
raising ValueError from __call__
zpublisher_exception_hook
raising ConflictError from zpublisher_exception_hook
abort
"""
pass
import doctest
def test_suite():
return doctest.DocTestSuite()
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment