2011/10/13 Peter Soetens <peter [..] ...>

> On Thu, Oct 13, 2011 at 11:05 PM, Willy Lambert <lambert [dot] willy [..] ...>
> wrote:
> >
> >
> > 2011/10/13 Peter Soetens <peter [..] ...>
> >>
> >> On Thu, Oct 13, 2011 at 11:01 PM, Willy Lambert <
> lambert [dot] willy [..] ...>
> >> wrote:
> >> >
> >> >
> >> > 2011/10/13 Peter Soetens <peter [..] ...>
> >> >>
> >> >> On Thu, Oct 13, 2011 at 12:45 AM, Willy Lambert
> >> >> <lambert [dot] willy [..] ...>
> >> >> wrote:
> >> >> > Hi,
> >> >> > As I where trying to convert my deployment to Lua, I am bumbimp
> into
> >> >> > this
> >> >> > abort code :
> >> >> > ...
> >> >> > 0.630 [CRITICAL][Thread] Could not allocate configuration semaphore
> >> >> > 'sem'
> >> >> > for
> >> >> > terminate called after throwing an instance of 'std::bad_alloc'
> >> >> > what(): std::bad_alloc
> >> >> > ./run.sh: line 30: 16239 Aborted rosrun ocl
> >> >> > deployer-$OROCOS_TARGET -s $ROOT_DEPLOYMENT_FILE
> >> >> > This append when I connect "too much" ports to the deployer (I mean
> I
> >> >> > tried
> >> >> > to add different ports, and it always fail with the same number of
> >> >> > ports).
> >> >> > Do I forgot some memory size configuration somewhere ?
> >> >> > I am using orocos 2.4, with a xenomai deployer that uses lua
> services
> >> >> > to
> >> >> > deploy the application. Note that the rtt_dot_service is loaded
> also
> >> >> > (I
> >> >> > don't know how he could be involved, but never know...)
> >> >>
> >> >> Afaikt, this is a kernel option of Xenomai. So reconfigure your
> kernel
> >> >> and check for the registry size. Also check /proc/xenomai for any
> >> >> clues...
> >> >>
> >> >
> >> > arg, I did resize the default section because we had this problem at
> >> > B.A,
> >> > maybe it is not enougth. So it seems to be a pure xenomai question,
> >> > should
> >> > I ask on their mailing list ?
> >>
> >> Maybe there is a way to create semaphores as such that they don't need
> >> this registry. I would recommend fixing _that_ in RTT's fosi instead
> >> or raising the registry limit all the time.
> >>
> >
> > For my information, what is generating this huge amount of semaphores
> > ? Which registry is it ?
>
> The Orocos Mutex maps to the Xenomai semaphore. Since each port is
> protected by a Mutex to do its connection management, this can add up.
> A connection does not add a mutex, unless the protocol it uses would
> do so (like the Locked connections).
>
> We have also two semaphores for each thread, and a semaphore for each
> execution engine, with a condition variable.
>
> None of these can be avoided and their lack would crash your current
> program if removed :-)
>

Oh, I am not searching to remove them :p Just trying to know what's
happening underneath

>
> Peter
>

On Oct 13, 2011, at 17:23 , Peter Soetens wrote:

> On Thu, Oct 13, 2011 at 11:05 PM, Willy Lambert <lambert [dot] willy [..] ...> wrote:
>>
>>
>> 2011/10/13 Peter Soetens <peter [..] ...>
>>>
>>> On Thu, Oct 13, 2011 at 11:01 PM, Willy Lambert <lambert [dot] willy [..] ...>
>>> wrote:
>>>>
>>>>
>>>> 2011/10/13 Peter Soetens <peter [..] ...>
>>>>>
>>>>> On Thu, Oct 13, 2011 at 12:45 AM, Willy Lambert
>>>>> <lambert [dot] willy [..] ...>
>>>>> wrote:
>>>>>> Hi,
>>>>>> As I where trying to convert my deployment to Lua, I am bumbimp into
>>>>>> this
>>>>>> abort code :
>>>>>> ...
>>>>>> 0.630 [CRITICAL][Thread] Could not allocate configuration semaphore
>>>>>> 'sem'
>>>>>> for
>>>>>> terminate called after throwing an instance of 'std::bad_alloc'
>>>>>> what(): std::bad_alloc
>>>>>> ./run.sh: line 30: 16239 Aborted rosrun ocl
>>>>>> deployer-$OROCOS_TARGET -s $ROOT_DEPLOYMENT_FILE
>>>>>> This append when I connect "too much" ports to the deployer (I mean I
>>>>>> tried
>>>>>> to add different ports, and it always fail with the same number of
>>>>>> ports).
>>>>>> Do I forgot some memory size configuration somewhere ?
>>>>>> I am using orocos 2.4, with a xenomai deployer that uses lua services
>>>>>> to
>>>>>> deploy the application. Note that the rtt_dot_service is loaded also
>>>>>> (I
>>>>>> don't know how he could be involved, but never know...)
>>>>>
>>>>> Afaikt, this is a kernel option of Xenomai. So reconfigure your kernel
>>>>> and check for the registry size. Also check /proc/xenomai for any
>>>>> clues...
>>>>>
>>>>
>>>> arg, I did resize the default section because we had this problem at
>>>> B.A,
>>>> maybe it is not enougth. So it seems to be a pure xenomai question,
>>>> should
>>>> I ask on their mailing list ?
>>>
>>> Maybe there is a way to create semaphores as such that they don't need
>>> this registry. I would recommend fixing _that_ in RTT's fosi instead
>>> or raising the registry limit all the time.
>>>
>>
>> For my information, what is generating this huge amount of semaphores
>> ? Which registry is it ?
>
> The Orocos Mutex maps to the Xenomai semaphore. Since each port is
> protected by a Mutex to do its connection management, this can add up.
> A connection does not add a mutex, unless the protocol it uses would
> do so (like the Locked connections).

Are you saying that every port always has a mutex for connection management, or that a port only has a connection management mutex if it is part of a locked connection?

I thought connections weren't real-time safe to make/break? From a stall the thread point of view IIRC, not multi-thread safe.

> We have also two semaphores for each thread, and a semaphore for each
> execution engine, with a condition variable.
>
> None of these can be avoided and their lack would crash your current
> program if removed :-)
>
> Peter