Forking this from Issue 717636 .
We have a user which is constantly ending up with a GPU process which has a high (800 M - 1.5 GB) private memory footprint.
See the traces attached to https://bugs.chromium.org/p/chromium/issues/detail?id=717636#c21 .
Soon after restart, after opening a bunch of tabs, its gpu process jumps to 800 MB.
More interestingly, in the "before_restart" trace, half of the memory (640 MB) is swapped, which somehow suggest that most of that has not been used for a long time (hence probably a leak).
From the memory-infra traces, the memory seem to come from "malloc", so not something we currently account in our gpu/cc/discardable probes. So I think our only hope here is --enable-heap-profiling.

The user has a lot of tabs (hundreds) so I think that just using the chrome-wide --enable-heap-profiling will knock out chrome when it will try to load 100+ tabs.
I wonder if we can do something like: --gpu-launcher=script_thas_passes_enable_heap_profiling_flag_to_the_gpu_process_only.sh
which would enable heap profiling only for the gpu process.

Let's try to get this working. If we get a workflow that works we can get back to the user and ask him to do that.
Can anybody here help me to setup the experiment above?

Machine details of the user:
cpu-brand: " Intel(R) Xeon(R) CPU E5-1650 0 @ 3.20GHz",
cpu-family: 6,
cpu-model: 45,
cpu-stepping: 7,
gpu-devid: 4090,
gpu-driver: "367.57",
gpu-gl-renderer: "Quadro K600/PCIe/SSE2",
gpu-gl-vendor: "NVIDIA Corporation",
gpu-psver: "4.50",
gpu-venid: 4318,
gpu-vsver: "4.50",
highres-ticks: true,
network-type: "Ethernet",
num-cpus: 12,
os-arch: "x86_64",
os-name: "Linux",
os-version: "4.4.0-66-generic",
physical-memory: 32098,
product-version: "Chrome/58.0.3029.96",