Micro-optimization: set cache header on kernelspecs logos

[mlucool]

Problem

kernelspecs's often contain png logos. These logos are on the order of 3kb each. Since kernels can't share a logo, each kernel contains a copy of the logo it provides. This leads to 3*number of kernels of uncached content being loaded on each page. I have ~20 kernels today, so that's 60kb being sent per refresh. (Note: this was flagged in lighthouse).

Proposed Solution

Ideally, we'd be able to share generic logos, but since that is a bigger change, I propose we add long cache headers on logos served from kernelspecs/<kernel>/logo-64x64.png. Unlike the kernelspec itself, these are expected to very rarely change.

[kevin-bates]

Hi @mlucool - this seems like a great idea. I don't know what is involved to do this, but I thought I'd point out the handler responsible for returning the logos (resources) in case you, or anyone else, would like to contribute the changes:

jupyter_server/jupyter_server/kernelspecs/handlers.py

Lines 11 to 32 in 62607ce

	class KernelSpecResourceHandler(web.StaticFileHandler, JupyterHandler):
	SUPPORTED_METHODS = ("GET", "HEAD")
	auth_resource = AUTH_RESOURCE
	def initialize(self):
	web.StaticFileHandler.initialize(self, path="")
	@web.authenticated
	@authorized
	def get(self, kernel_name, path, include_body=True):
	ksm = self.kernel_spec_manager
	try:
	self.root = ksm.get_kernel_spec(kernel_name).resource_dir
	except KeyError as e:
	raise web.HTTPError(404, "Kernel spec %s not found" % kernel_name) from e
	self.log.debug("Serving kernel resource from: %s", self.root)
	return web.StaticFileHandler.get(self, path, include_body=include_body)
	@web.authenticated
	@authorized
	def head(self, kernel_name, path):
	return self.get(kernel_name, path, include_body=False)

[mlucool]

Humm, it seems the code you pointed to already uses a StaticFileHandler which sets an ETag. Lighthouse is upset that there is not Cache TTL, which should be easy enough to change.