[ovirt-devel] FD leak while initializing vdsm

Martin Sivak msivak at redhat.com
Wed May 25 10:30:22 UTC 2016 

For example all the SocketServer or xmlrpclib
>> abstractions close the sockets automatically.
>
> $ grep __del__ /usr/lib64/python2.7/SocketServer.py
> $ grep __del__ /usr/lib64/python2.7/xmlrpclib.py
> $ grep __del__ /usr/lib64/python2.7/SimpleXMLRPCServer.py

I did not say they are using  __del__. I was merely pointing out that
not cleaning after yourself violates the rule of the least surprise in
Python. Because (almost) all higher level objects in the standard
library avoid the need for explicit close call.

> I disagree, __del__ cannot be implemented properly on python 2.
> It is possible to close stuff automatically only in a c extension.

Yes, you need to be careful and you might get into trouble because of
exception handling, but the simple cases where you catch everything
and delete self should be working just fine.

> Trying to implement will lead to endless troubles.

Any examples? (except __init__ not called and exception handling).


Martin


On Wed, May 25, 2016 at 12:21 PM, Nir Soffer <nsoffer at redhat.com> wrote:
> On Wed, May 25, 2016 at 12:18 PM, Martin Sivak <msivak at redhat.com> wrote:
>>> but it should
>>> not try to implement __del__ - we had lot of trouble with such code in vdsm.
>>
>> Del is perfectly fine if done properly. Python objects should really
>> clean after themselves. It is common to abstract the user from the low
>> level networking code. For example all the SocketServer or xmlrpclib
>> abstractions close the sockets automatically.
>
> $ grep __del__ /usr/lib64/python2.7/SocketServer.py
> $ grep __del__ /usr/lib64/python2.7/xmlrpclib.py
> $ grep __del__ /usr/lib64/python2.7/SimpleXMLRPCServer.py
>
>> The same is true about
>> file objects (closed when garbage collected).
>
> This is implemented as C extension, not using __del__.
>
>>
>> Context manager can (and should if possible) be used to handle
>> exceptions properly and to allow better control, but the default
>> behavior should still close the resources during __del__ too.
>
> I disagree, __del__ cannot be implemented properly on python 2.
> It is possible to close stuff automatically only in a c extension.
> Trying to implement will lead to endless troubles.
>
> Nir
>
>>
>> Martin
>>
>>
>> On Mon, May 23, 2016 at 2:57 PM, Nir Soffer <nsoffer at redhat.com> wrote:
>>> On Mon, May 23, 2016 at 3:46 PM, Martin Sivak <msivak at redhat.com> wrote:
>>>>> Explicit close is required.
>>>>>
>>>>> You should ensure that all resources are released using try finally.
>>>>
>>>> Not according to the code Simone found. The close is called by the
>>>> __del__ method explicitly. And Python is no C, we should really behave
>>>> "nice" by default.
>>>
>>> No, the code you found is misguided attempt and should be deleted from vdsm.
>>>
>>> The jsonrpc client can provide a context manger to help you close it,
>>> but it should
>>> not try to implement __del__ - we had lot of trouble with such code in vdsm.
>>>
>>> You must close the client explicitly when you are done.
>>>
>>> Nir
>>>
>>>>
>>>> Martin
>>>>
>>>> On Mon, May 23, 2016 at 2:40 PM, Nir Soffer <nsoffer at redhat.com> wrote:
>>>>> On Mon, May 23, 2016 at 11:15 AM, Martin Sivak <msivak at redhat.com> wrote:
>>>>>> Hi,
>>>>>>
>>>>>>> I know that we reconnect several times during hosted engine process. Do we
>>>>>>> close client when it is not used anymore?
>>>>>>
>>>>>> No, we are not closing it according to Simone, shouldn't it be
>>>>>> released automatically? We are using Python after all.. explicit close
>>>>>> is not exactly common there.
>>>>>
>>>>> Explicit close is required.
>>>>>
>>>>> You should ensure that all resources are released using try finally.
>>>>>
>>>>> Nir
>>>>>
>>>>>>
>>>>>> Martin
>>>>>>
>>>>>> On Mon, May 23, 2016 at 8:55 AM, Piotr Kliczewski <pkliczew at redhat.com> wrote:
>>>>>>> Sandro,
>>>>>>>
>>>>>>> I know that we reconnect several times during hosted engine process. Do we
>>>>>>> close client when it is not used anymore?
>>>>>>>
>>>>>>> Please provide lsof for the process and the log.
>>>>>>>
>>>>>>> Thanks,
>>>>>>> Piotr
>>>>>>>
>>>>>>> On Mon, May 23, 2016 at 8:42 AM, Sandro Bonazzola <sbonazzo at redhat.com>
>>>>>>> wrote:
>>>>>>>>
>>>>>>>> MainThread::WARNING::2016-05-23
>>>>>>>> 07:09:38,629::hosted_engine::480::ovirt_hosted_engine_ha.agent.hosted_engine.HostedEngine::(start_monitoring)
>>>>>>>> Unexpected error
>>>>>>>> Traceback (most recent call last):
>>>>>>>>   File
>>>>>>>> "/usr/lib/python2.7/site-packages/ovirt_hosted_engine_ha/agent/hosted_engine.py",
>>>>>>>> line 444, in start_monitoring
>>>>>>>>     self._initialize_vdsm()
>>>>>>>>   File
>>>>>>>> "/usr/lib/python2.7/site-packages/ovirt_hosted_engine_ha/agent/hosted_engine.py",
>>>>>>>> line 635, in _initialize_vdsm
>>>>>>>>     timeout=envconstants.VDSCLI_SSL_TIMEOUT
>>>>>>>>   File
>>>>>>>> "/usr/lib/python2.7/site-packages/ovirt_hosted_engine_ha/lib/util.py", line
>>>>>>>> 187, in connect_vdsm_json_rpc
>>>>>>>>     requestQueue=requestQueue,
>>>>>>>>   File "/usr/lib/python2.7/site-packages/vdsm/jsonrpcvdscli.py", line 222,
>>>>>>>> in connect
>>>>>>>>     responseQueue)
>>>>>>>>   File "/usr/lib/python2.7/site-packages/vdsm/jsonrpcvdscli.py", line 212,
>>>>>>>> in _create
>>>>>>>>     lazy_start=False)
>>>>>>>>   File "/usr/lib/python2.7/site-packages/yajsonrpc/stompreactor.py", line
>>>>>>>> 576, in StandAloneRpcClient
>>>>>>>>     reactor = Reactor()
>>>>>>>>   File "/usr/lib/python2.7/site-packages/yajsonrpc/betterAsyncore.py",
>>>>>>>> line 200, in __init__
>>>>>>>>     self._wakeupEvent = AsyncoreEvent(self._map)
>>>>>>>>   File "/usr/lib/python2.7/site-packages/yajsonrpc/betterAsyncore.py",
>>>>>>>> line 164, in __init__
>>>>>>>>     map=map
>>>>>>>>   File "/usr/lib64/python2.7/asyncore.py", line 650, in __init__
>>>>>>>>     self.set_file(fd)
>>>>>>>>   File "/usr/lib64/python2.7/asyncore.py", line 657, in set_file
>>>>>>>>     self.socket = file_wrapper(fd)
>>>>>>>>   File "/usr/lib64/python2.7/asyncore.py", line 616, in __init__
>>>>>>>>     self.fd = os.dup(fd)
>>>>>>>> OSError: [Errno 24] Too many open files
>>>>>>>>
>>>>>>>> Simone, Rafael, Piotr, Martin, can you please investigate?
>>>>>>>>
>>>>>>>> vdsm-yajsonrpc-4.18.0-16.git51df339.el7.centos.noarch
>>>>>>>>
>>>>>>>> ovirt-hosted-engine-ha-2.0.0-0.2.master.20160520143206.20160520143149.gita012f18.el7.noarch
>>>>>>>>
>>>>>>>> --
>>>>>>>> Sandro Bonazzola
>>>>>>>> Better technology. Faster innovation. Powered by community collaboration.
>>>>>>>> See how it works at redhat.com
>>>>>>>
>>>>>>>
>>>>>> _______________________________________________
>>>>>> Devel mailing list
>>>>>> Devel at ovirt.org
>>>>>> http://lists.ovirt.org/mailman/listinfo/devel

More information about the Devel mailing list