Обсуждение: BackendPidGetProc doesn't return PGPROC for background worker?

On 2015-05-15 AM 05:01, Pavel Stehule wrote:
> 
> I am trying to start bgworker from bgworker and create communication
> between these process. I have a code based on test_shm_mq. This code fails
> because BackendPidGetProc doesn't find related bgworker process, although
> the registrant process is living
> 

One reason for this may be that the worker was not started with the flag
BGWORKER_SHMEM_ACCESS which is necessary to perform InitProcess() that would
initialize a PGPROC entry for it. But if you'd used the same method for
initializing workers as test_shm_mq_setup(), then it should have.

Thanks,
Amit

On 2015-05-15 AM 10:39, Amit Langote wrote:
> On 2015-05-15 AM 05:01, Pavel Stehule wrote:
>>
>> I am trying to start bgworker from bgworker and create communication
>> between these process. I have a code based on test_shm_mq. This code fails
>> because BackendPidGetProc doesn't find related bgworker process, although
>> the registrant process is living
>>
> 
> One reason for this may be that the worker was not started with the flag
> BGWORKER_SHMEM_ACCESS which is necessary to perform InitProcess() that would
> initialize a PGPROC entry for it. But if you'd used the same method for
> initializing workers as test_shm_mq_setup(), then it should have.
> 

It seems in addition, a BackgroundWorkerInitializeConnection() is also
necessary for the PGPROC entry of the worker to be visible to others. I do not
see that done anywhere in test_shm_mq(); so perhaps that's missing?

Thanks,
Amit

On 2015-05-15 AM 10:59, Amit Langote wrote:
> On 2015-05-15 AM 10:39, Amit Langote wrote:
>> On 2015-05-15 AM 05:01, Pavel Stehule wrote:
>>>
>>> I am trying to start bgworker from bgworker and create communication
>>> between these process. I have a code based on test_shm_mq. This code fails
>>> because BackendPidGetProc doesn't find related bgworker process, although
>>> the registrant process is living
>>>
>>
>> One reason for this may be that the worker was not started with the flag
>> BGWORKER_SHMEM_ACCESS which is necessary to perform InitProcess() that would
>> initialize a PGPROC entry for it. But if you'd used the same method for
>> initializing workers as test_shm_mq_setup(), then it should have.
>>
> 
> It seems in addition, a BackgroundWorkerInitializeConnection() is also
> necessary for the PGPROC entry of the worker to be visible to others. I do not
> see that done anywhere in test_shm_mq(); so perhaps that's missing?
> 

And these conditions apply to the bgworker that started another bgworker, that
is the registrant bgworker. I think such a pattern does not exist in existing
code. That is, normally all workers are started by a user backend that has a
valid shared PGPROC entry.

Thanks,
Amit

On 2015-05-15 PM 02:13, Pavel Stehule wrote:
> 2015-05-15 3:39 GMT+02:00 Amit Langote <Langote_Amit_f8@lab.ntt.co.jp>:
>> One reason for this may be that the worker was not started with the flag
>> BGWORKER_SHMEM_ACCESS which is necessary to perform InitProcess() that
>> would
>> initialize a PGPROC entry for it. But if you'd used the same method for
>> initializing workers as test_shm_mq_setup(), then it should have.
>>
> 
> I have it
> 

See my other (the last) email. You said you created a bgworker from bgworker,
so if the parent bgworker did not perform
BackgroundWorkerInitializeConnection, then it would not be found in ProcArray.

Thanks,
Amit