Memory metrics

[dlqqq]

Description

We would like some way of monitoring the memory usage of the server instance.

Right now, we don't have a way of auto-deleting YRoom instances without breaking the reference to the YRoom which may be held by consumers. Therefore the server's memory use will monotonically increase across its lifetime.

However, if we add metrics & discover that the server's memory use is small, we may be able to publish v0.1.0 before addressing memory cleanup.

[dlqqq]

This is pretty tricky. Spent some time today exploring the options.

I wasn't able to inspect the size of a pycrdt.Doc using the tracemalloc stdlib module. pycrdt depends on Yrs, which manages its own memory independent of the Python interpreter, so it doesn't seem like there's any way to use tracemalloc to track RTC memory use.

I did manage to produce some plots with memory-profiler, but the testing process is labor-intensive as it takes minutes to start jupyter-lab, run a test, and then save the plot to a file.

Here's the result of a single comparison test I ran with by printing 1 million lines in a notebook:

[dlqqq]

See also: y-crdt/y-crdt#458