Hi Pavel,
I believe that I have found a solution by shuffling the __wait_for_events()
invoke. This is the patch (also available as attachment):
--- C:/Python27/Lib/site-packages/fdb-1.4.3-py2.7.egg/fdb/fbcore.py Mo Nov
17 08:27:57 2014
+++ C:/Python27/Lib/site-packages/fdb-1.4.3-py2.7.egg/fdb/fbcore.py Mo Nov
17 08:29:08 2014
@@ -1729,9 +1729,10 @@ class EventBlock(object):
self.buf_length = api.isc_event_block(ctypes.pointer(self.event_buf),
ctypes.pointer(self.result_buf),
*[b(x) for x in event_names])
- self.__wait_for_events()
def __lt__(self,other):
return self.event_id.value < other.event_id.value
+ def begin(self):
+ self.__wait_for_events()
def __wait_for_events(self):
api.isc_que_events(self._isc_status,self._db_handle,self.event_id,
self.buf_length,self.event_buf,
@@ -1841,7 +1842,9 @@ class EventConduit(object):
for block_events in self.__blocks:
- self.__event_blocks.append(EventBlock(self.__queue,
self._db_handle, block_events))
+ event_block = EventBlock(self.__queue, self._db_handle,
block_events)
+ self.__event_blocks.append(event_block)
+ event_block.begin()
With this, even after 50000+ events raised and processed and 2400+ event
conduits created and disposed, gc.garbage doesn't contain EventBlock items
anymore. *HOORAY!* :-)
While inspecting gc.garbage I have further noticed that it does contain several
fdb.fbcore.PreparedStatement items. According to
https://docs.python.org/2/library/gc.html#gc.garbage :
"By default, this list contains only objects with __del__() methods. [...] Note
that these objects are kept alive even so by virtue of being in the garbage
list, so they should be removed from garbage too. For example, after breaking
cycles, do del gc.garbage[:] to empty the list. It’s generally better to avoid
the issue by not creating cycles containing objects with __del__() methods, and
garbage can be examined in that case to verify that no such cycles are being
created."
To me some obvious questions arise:
1] Is this behavior expected?
2] If [1], does fdb clean up gc.garbage from time to time?
3] If not [1], should fdb.fbcore.PreparedStatement better not override __del__
such that gc can do its job?
4] If not [2] and not [3], should fdb clean up gc.garbage?
Please note also that the memory foodprint of the fdb.fbcore.PreparedStatement
items in question is minimal because all containing data in
fdb.fbcore.PreparedStatement is freed. Due to this this issue is not
problematic in "normal" environments, but might become critical in an
environment that runs for a long time under heavy load.
Cheers,
Dominik
Von: firebird-python@yahoogroups.com [mailto:firebird-python@yahoogroups.com]
Gesendet: Freitag, 14. November 2014 15:03
An: firebird-python@yahoogroups.com
Betreff: Re: AW: AW: AW: [firebird-python] FDB v1.4.2
Hi,
Can you try to add a call to self.close() before exception is raised in
__wait_for_events ? Does it helps?
if db_api_error(self._isc_status):
self.close()
raise exception_from_status(DatabaseError, self._isc_status, "Error
while waiting for events:")
best regards
Pavel
Dne 14.11.2014 v 14:08 Dominik Psenner dominik.psen...@topcontrol.it
[firebird-python] napsal(a):
> I refreshed the package and it looks like the memory leak is almost gone. I
> can observe that the number of EventBlock’s left in memory are equal to the
> count of errors raised with the invoke of begin() (namely one of: unknown ISC
> error 0, error reading from connection, error writing to connection).
>
> So there’s still a troublesome execution path left and it took some time
> until I had wrapped my head around it. According to the stack trace the
> exception is thrown in the constructor of EventBlock:
>
> File "dbEvents.py", line 103, in run
> event_listener.begin()
> File "C:\Python27\lib\site-packages\fdb-1.4.3-py2.7.egg\fdb\fbcore.py", line
> 1844, in begin
> self.__event_blocks.append(EventBlock(self.__queue, self._db_handle,
> block_events))
> File "C:\Python27\lib\site-packages\fdb-1.4.3-py2.7.egg\fdb\fbcore.py", line
> 1732, in __init__
> self.__wait_for_events()
> File "C:\Python27\lib\site-packages\fdb-1.4.3-py2.7.egg\fdb\fbcore.py", line
> 1741, in __wait_for_events
> "Error while waiting for events:")
> DatabaseError: ('Error while waiting for events:\n- SQLCODE: -902\n- Error
> reading data from the connection.', -902, 335544726)
>
> That would be right here where the EventBlock should be appended to the list
> of event blocks:
>
> for block_events in self.__blocks:
> self.__event_blocks.append(EventBlock(self.__queue, self._db_handle,
> block_events))
>
> But like this an initialized and broken EventBlock is never appended to the
> __event_blocks list and therefore EventBlock.close() is never invoked. ☹
>
> Von: firebird-python@yahoogroups.com [mailto:firebird-python@yahoogroups.com]
> Gesendet: Freitag, 14. November 2014 13:38
> An: firebird-python@yahoogroups.com
> Betreff: Re: AW: AW: [firebird-python] FDB v1.4.2
>
>
>
> Hi,
>
> Fixed. This time I've just refreshed the 1.4.3 package, as I don't want
> to make another point release in single day. If anyone already got
> unpatched 1.4.3 (unlikely that anyone beside you get it so quickly),
> then it will replaced by next version 1.5 (full Firebird 3.0 support)
> scheduled for release before end of year.
>
> best regards
> Pavel Cisar
> IBPhoenix
>
> Dne 14.11.2014 v 12:24 Dominik Psenner
> dominik.psen...@topcontrol.it<mailto:dominik.psen...@topcontrol.it>
> [firebird-python] napsal(a):
>> Ah, you were faster than I could write. Sorry for me being so persistent.
>> Unfortunately I am unable to post a file in the bug tracker now that the
>> issue is resolved, therefore I write you here.
>>
>> The good thing is, 1.4.3 improved the situation. One cyclic reference has
>> gone away but there’s still one cyclic reference left to be broken. The
>> attached objgraph shows that the one left might be the callback function
>> which is stored in the EventBlock upon creation. You could break that by
>> cleaning up the callback attribute on close(). Doing that should resolve
>> this issue for good.
>>
>> Von: firebird-python@yahoogroups.com<mailto:firebird-python@yahoogroups.com>
>> [mailto:firebird-python@yahoogroups.com]
>> Gesendet: Freitag, 14. November 2014 12:06
>> An: firebird-python@yahoogroups.com<mailto:firebird-python@yahoogroups.com>
>> Betreff: AW: AW: [firebird-python] FDB v1.4.2
>>
>>
>> Awesome! Thanks for the fast responses, you’re doing a good job. ☺ If you
>> feel like sending me a patch that I can apply before you push 1.4.3 out of
>> the door, I’ll gladly run my test scripts on a patched 1.4.2.
>>
>> Cheers
>>
>> Von:
>> firebird-python@yahoogroups.com<mailto:firebird-python@yahoogroups.com<mailto:firebird-python@yahoogroups.com%3cmailto:firebird-python@yahoogroups.com>>
>> [mailto:firebird-python@yahoogroups.com]
>> Gesendet: Freitag, 14. November 2014 11:50
>> An:
>> firebird-python@yahoogroups.com<mailto:firebird-python@yahoogroups.com<mailto:firebird-python@yahoogroups.com%3cmailto:firebird-python@yahoogroups.com>>
>> Betreff: Re: AW: [firebird-python] FDB v1.4.2
>>
>>
>>
>> Hi,
>>
>> Fixed in 1.4.3, see PyPI. Thanks for pointing this out.
>>
>> best regards
>> Pavel Cisar
>> IBPhoenix
>>
>> Dne 14.11.2014 v 11:42 Dominik Psenner
>> dominik.psen...@topcontrol.it<mailto:dominik.psen...@topcontrol.it<mailto:dominik.psen...@topcontrol.it%3cmailto:dominik.psen...@topcontrol.it>>
>> [firebird-python] napsal(a):
>>> Hi Pavel,
>>>
>>> great to hear that you have got this release out of the door so quickly.
>>> However, I’m afraid that http://tracker.firebirdsql.org/browse/PYFB-43 is
>>> not yet resolved. As posted in the issue I was just now able to reproduce
>>> the memory leak with 1.4.2. Thus, people, beware from invoking
>>> event_conduit() in a loop!
>>>
>>> Are there any plans for fixing the memory leak?
>>>
>>> Best regards,
>>> Dominik
>>>
>>> Von:
>>> firebird-python@yahoogroups.com<mailto:firebird-python@yahoogroups.com<mailto:firebird-python@yahoogroups.com%3cmailto:firebird-python@yahoogroups.com>>
>>> [mailto:firebird-python@yahoogroups.com]
>>> Gesendet: Freitag, 14. November 2014 10:36
>>> An:
>>> firebird-python@yahoogroups.com<mailto:firebird-python@yahoogroups.com<mailto:firebird-python@yahoogroups.com%3cmailto:firebird-python@yahoogroups.com>>
>>> Betreff: [firebird-python] FDB v1.4.2
>>>
>>>
>>>
>>> Hi,
>>>
>>> FDB Python driver for Firebird v1.4.2 is available for download . This
>>> mostly bugfix version introduces a backward incompatible change to
>>> events API, so read the Release Notes before deployment.
>>>
>>> https://pypi.python.org/pypi/fdb
>>>
>>> best regards
>>> Pavel Cisar
>>> IBPhoenix
>>>
>>
>
