PerformanceData Postprocessor Error

[singhgp4321]

I am trying to obtain the jacobian compute metrics for a simulation in Bison app using the PerformanceData postprocessor and keep getting the following error:

Unknown section_name: FEProblem::computeJacobianInternal in PerfGraph::getTime()
If you are attempting to retrieve the root use "Root".

If I change the data_type to calls in the postprocessor block, it shows the same error in PerfGraph::getNumCalls().

I tried to debug it. Looking at the top frame when it is about to error out: it tries to find section_name (holding the string FEProblem::computeJacobianInternal) inside the container _section_time. The _section_time container seems to have many objects except FEProblem::computeJacobianInternal.

When I remove this postprocessor and turn on the perf_graph I see the computeJacobianInternal related information in the generated performance log table - indicating the jacobian metrics are being determined successfully but postprocessor is not able to retrieve that information.

I am wondering why is the PerformanceData postprocessor failing to retrieve the jacobian compute information.

Thanks,
Gyan

[aeslaughter]

@friedmud Can you help with this?

[singhgp4321]

Just wondering if someone could help with this. Let me know if you need any information.

[loganharbour]

PerformanceData is deprecrated. You should use PerfGraphData instead.

I believe this is what you're looking for:

[Postprocessors]
  [jacobian_time]
    type = PerfGraphData
    section_name = 'FEProblem::computeJacobianInternal'
    data_type = TOTAL
  []
[]

See here for more information: https://mooseframework.inl.gov/source/postprocessors/PerfGraphData.html

[singhgp4321]

Actually I am using the PerfGraphData and not the PerformanceData object. I got confused in naming due to this page (https://mooseframework.inl.gov/source/postprocessors/PerformanceData.html).

[loganharbour]

This is interesting. I added the above to test/tests/kernels/simple_diffusion/simple_diffusion.i and got the intended results.

Can you please open an issue within the BISON repository (do not do it here) with the input that you are running? Respond here when you do and I'll take a look there.

[loganharbour]

Alright, found the issue. Postprocessors by default run on TIMESTEP_END. If the section that you are executing has not been called yet before TIMESTEP_END, it doesn't exist it'll error out that it can't find the section.

For the time being, your solution is to set execute_on to something on your own that occurs after the section has been called, which in your case is almost anything but TIMESTEP_END. Something like NONLINEAR or FINAL.

This isn't a very friendly error. We actually register the section very early on (well before INITIAL), therefore we do know that it exists. I'll work on a change that makes this more meaningful. I'd prefer that it just return 0 in all cases if it has been registered but has not been called yet.

[loganharbour]

#16704 will resolve this issue in the future - valid sections will return zero if they have not ran yet.

[singhgp4321]

Thanks Logan! Appreciate it very much!