perf: Freeze GC before forking Gunicorn workers

[ankush]

Both Gunicorn and RQ use forking to spawn workers. In an ideal world, the fork should be sharing most of the memory if there are no writes made to data because of Copy on Write, however, python's GC is not CoW friendly and writes to data even if user-code doesn't. Specifically, the generational GC which stores and mutates every python object: PyGC_Head

This can be avoided by doing gc.freeze() pre-fork to freeze all current objects and move them permanently to the last generation of generational gc.

Reference:

• https://docs.python.org/3/library/gc.html#gc.freeze
• https://devguide.python.org/internals/garbage-collector/#memory-layout-and-object-structure
• Support for gc.freeze() for apps that use preloading benoitc/gunicorn#1640
• https://en.wikipedia.org/wiki/Copy-on-write
• https://instagram-engineering.com/copy-on-write-friendly-python-garbage-collection-ad6ed5233ddf

This is behind an environment variable for now, enable by setting FRAPPE_TUNE_GC to any truthy value.

Preliminary numbers

Metric	before	after	reduction
24 Worker total PSS	1791.518	1236.89	31%
1 RQ Worker + 1 Horse PSS	76.2	68.4	10%

PS: measuring benefit of this is kinda hard as memory usage depends on usage, without freeze most shared memory would become unshared over time.

TODO:

• Put it behind some environmental variable for now (?)
• Apply the same logic to background jobs. Not possible, we right now create duplicate processes instead of forking from one master process. RQ is inefficient for this, though something like workerpool or https://github.com/ankush/rq_orchestrator/ can solve this.
• Tweak gc params. We reload entire process after some requests, no need to aggressively GC considering that.
• Check behaviour with reloaders (or just disable in dev mode)
• test with autorestart
• Run in prod and gather some data on memory usage over time.
  • review what should be imported pre-fork, since sharing is now beneficial.

Should close #18927

Continues #21473, #21467

[codecov]

Codecov Report

Merging #21474 (29d28a4) into develop (278b6fc) will increase coverage by 0.00%.
The diff coverage is 68.42%.

❗ Current head 29d28a4 differs from pull request most recent head 150c36c. Consider uploading reports for the commit 150c36c to get more accurate results

Additional details and impacted files

@@           Coverage Diff            @@
##           develop   #21474   +/-   ##
========================================
  Coverage    63.85%   63.85%           
========================================
  Files          766      766           
  Lines        69436    69440    +4     
  Branches      6276     6276           
========================================
+ Hits         44337    44343    +6     
+ Misses       21514    21512    -2     
  Partials      3585     3585           

Flag	Coverage Δ
server	68.82% <68.42%> (-0.01%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.