Shouldn't we just use the total_size column in the watchdog instead (at least for now on Windows)?

resident_size is the RSS on Linux which are the active pages in physical memory, so as a concept it sort of maps to the Working Set of Windows.

I believe the issue talks about Windows, but given that Private Bytes also accounts for paged out memory, we should also change the behavior on the other platforms for consistency.

On Linux you might need to sum other counters like VmSwap or similar (I only had a brief look at this).

Also, this is slightly tangential, but now that I see, we don't have a "1:1" mapping between Linux and Windows on total_size too, because on Linux we use the virtual address space size (VmSize) which is not the same as Windows's Private Bytes, but indeed maps to the size of Virtual Address space (which also includes shared memory for instance).

I'm not totally sure on Windows how you would take that though.

Right, perhaps we need a new column "private_size". Then we can switch back to using Working Set for Windows. I am unsure how to get the virtual size to accurately set "total_size".

The private size on Linux would be a summation of Private Dirty values within smaps. (My best guess at the most accurate representation.)

Right, perhaps we need a new column "private_size". Then we can switch back to using Working Set for Windows. I am unsure how to get the virtual size to accurately set "total_size".

I'm in favor of adding such a column, but what I was trying to say is that the expectation and definition of resident_size is the memory that is kept in physical memory, so setting it to the value of Private Bytes would be incorrect, because it also contains paged out memory.

Also given that we already have such value in the total_size column, wouldn't it be better, for this specific watchdog fix, to actually change the watchdog so that it looks at total_size instead of resident_size, on Windows?

And I mean, specifically changing here:

to get total_size.

Yeap, I agree

Introducing a new private_size column is a bit complex at this point in time. Let's go with the half-measure band aid and update the watcher logic for now.