Is it possible somehow to cook "inProccess" in the same thread (or maybe reuse the main one)?

What we are trying to solve:
We have not thread-safe call and probably make workitem dep. (in the same row) is the first think which is coming to mind but it can transform into a limitation because we don't need a predefined order. We just need a random serial execution.

If you're using Houdini 18.5, you can create an in process scheduler and set it as the scheduler override for the node(s) that have work items that need to run on the same thread. Then, using the in process scheduler or by adding job parms, you can set the "Run on Main Thread" option to force the work items to run on the main thread. Only one can run at a time when doing that, so I think that should achieve what you're looking for.

Is it possible to implement "Run on Main Thread" option in a custom scheduler for "inProccess" workitems?
Do you have publicly available source for inprocess scheduler?

Thanks

The in process scheduler is built into PDG itself -- it's written in C++, not in Python. It's currently not possible to implement the same functionality from a custom scheduler, but we can look into exposing it.

It will be really helpful to expose this functionality.

Thanks
Ostap

There's a new API method in the next daily build of H18.5 that should do what you need. It gives you a way to run a Python function on the main thread, and optionally wait for the function to complete. You can use it anywhere in Python as part of a PDG cook, so in a Python Script or a Python Processor's onCookTask callback, for example. The method will let you pass a function and a list of arguments, which will get queued up and run serially on the main thread. That means, for example, you could use it to open a Qt dialog from a Python Script and then wait for a response:


The docs for the new API method will also be in the build tomorrow, but to give a brief summary the pdg.Scheduler class has a new runOnMainThread method that can run an arbitrary Python function on the main thread. The first argument indicates whether you want to wait for the function to run -- if you need access to the returned value from the function, then you'll have to wait for example. The second argument is the Python function or callable object, and any additional arguments after that are passed to the function when it gets called.

This is similar to the existing hdefereval available in HOM, but it's managed by the PDG cook instead. That means if you cancel a PDG cook any unstarted main thread tasks will also be dropped from the queue. It will also work in a headless Houdini session, for example when cooking a TOP graph from hbatch or hython, as long as the headless graph cook is configured to block instead of cook in the background.

It's using the same internal mechanism that the built-in in process scheduler uses. Note that its still not possible to write a custom in process scheduler, which we do plan to add support for as well. That's a larger change though, and I think the new runOnMainThread method is enough to solve your use case on its own.

Thanks!