Postprocessing script to release Python memory?

[divergentti]

Kubuntu 24.04 LTS + Automatic 1.9.4 seems to hold memory of Python3.10 process if some script crashes, like Tiled Diffusion or ControlNet scripts. This may happen quite often in Linux due to CUDA Out Of Memory or if ContolNet preprosessor is not satisfied to parameters etc.

Typical Python3.10 process memory consumption just after startup using SDXL checkpoint is about 8GB. Upscaling with Tiled Diffusion / VAE could increase memory comsumption sometimes up to 32GB (max RAM) and stay there until startup process webui.sh is killed. Changing checkpoints, or txt2img etc. do not release memory usage.

I created a simple script for a carbage collection from Scripts selection:

import gc
import modules.scripts as scripts
import gradio as gr

class FreeMemScript(scripts.Script):
    def title(self):
        return "Free Memory Script"

    def ui(self, is_img2img):
        button = gr.Button("Run Free Memory Script")
        # Link the button to the `freemem` function
        button.click(fn=self.freemem, inputs=[], outputs=[])
        return [button]

    def freemem(self):
        def getCurrentMemoryUsage():
            with open('/proc/self/status') as f:
                memusage = f.read().split('VmRSS:')[1].split('\n')[0][:-3]
            return int(memusage.strip())

        print("Free memory before gc: %s Kb" % getCurrentMemoryUsage())
        gc.collect()
        print("Free memory after gc: %s Kb " % getCurrentMemoryUsage())

    def run(self, p, *args):
        # Run method can remain empty if no additional processing is needed
        return p

# Register the script
script = FreeMemScript()

---

... but it seems that gc.collect() does not release memory. Is there some other ways to release hung memory?