High memory usage. I keep running out of memory.

[mzso]

Hi!
For a good while now I'm experienceing high memory usage, and associated sluggishness, yank.
I have a bunch of stuff, addons, userscripts and such.
I tried to get memory reports analyzed by AI, but most fail with such large files. Only Gemini managed to finish an analysis. It pointed at TST and ghost windows. It's statements are not near certain, but I guess it's not impossible that TST is related.
I have loads of unloaded tabs, that I never seem to get to. Is that likely to be a big factor?

It also says the the report only accounts for ~11GB of memory. But around 24GB was freed up when I closed FF. So I'm not sure what to make of that.

Any ideas how to figure out what's causing such leak?
I have memreports, but I don't think anyone wants to sift through giant memory reports. Maybe there's an LLM that's good at processing this stuff?

[irvinm]

Not a memory leak expert, but I usually start with looking at "about:memory". I see they are now showing some high level information (which is nice) which then you can drill down into. I'm on Windows and this 4.7GB of memory use roughly reflects what is being displaying in task manager. (32 tabs loaded of 713 in my session)

[mzso]

@irvinm Well, I don't. Unless I add up the private bytes in process explorer. But it's easy to see in Process Explorer how much system commit is freed up when I closed FF.
Right now I have this:

Which is 15,3 GB total according to Gemini. So that's a 5GB differece, but it's not near how bad it can get. It's also quite weird that I have so many processes. I have set dom.ipc.processCount;12 a long while ago. Maybe they removed the setting.

[irvinm]

I think you can also see the difference if you enable "Working Set" in "Process Explorer". I think there is usually a difference between the 2 numbers so that Firefox doesn't dominate physical memory vs. all of the other running processes. In my example (just my running state right now) the gap is pretty big.

At the end of the day, I doubt this is TST related. I would imagine if you disabled TST and restarted and them measured things after a while, you would see the same relative difference. If you take off the Firefox filter in process explorer, I am seeing tons of applications that have "private bytes" that are 50% to 500% larger than their working set. I'm sure windows pushes idle/background process memory out in favor of the active ones.

What do you think?

[mzso]

@irvinm
As far as I know the working set is a subset of private bytes. The part that is currently in the memory. So unless I'm mistaken the difference between them is not relevant in this regard. There's a bunch more memory allocated than what about memory shows. So it's natural for private bytes to be much larger, as long as you have loads of swap space. (On a totally different note, just to make matters fun, right now I have a runaway gpu process at 13GB...)

[irvinm]

I agree with you in terms of the definitions and that the difference isn't important.

I still think you should either try running your profile without TST enabled (for a while) to see what the numbers look like and/or maybe try a new profile to see if that changes things.

[piroor]

Anyway there are too many possible factors which can increase memory usage in TST:

• Half-transparent colors
• Half-transparent gradients
• Animations
• data: URI strings for favicons (this is the largest factor I know)
• JS objects kept on the memory unexpectedly due to long-live references

We need to make the bottle neck clear with step-by-step investigation, with huge help, for example: commenting out some codes and trying to run it via about:debugging.

[piroor]

FYI, my case:
It may be GC-blocked problem if you see shrunken memory usage with clicking the button "Minimize memory usage".

[mzso]

@piroor
Pressing minimize does reduce memory usage somewhat. But not near in proportion to the memory usage of the browser.
And apparently much of it is unlisted. According to chatgpt. There isn't even a summary, and too many items to add up manually.