More than 10x slowdown with 1.0.0rc0.

[s-banach]

Simple test script:

from time import perf_counter

import plotly.express as px

fig = px.scatter(x=[1, 2, 3], y=[1, 2, 3])

start = perf_counter()
for _ in range(10):
    fig.to_image(format="svg", engine="kaleido")
end = perf_counter()
print(f"{end - start:.02f}")

Time with 0.1.0.post1: 0.95 seconds
Time with 1.0.0rc0: 10.78 seconds

[s-banach]

I'm assuming it does something like open a new copy of chrome for each image.
In that case, please include guidance on how to make a persistent instance of chrome to generate multiple images.
Even better, a pool of chrome instances that can be used to render a list of figures in parallel.

[ayjayt]

Yeah, that's the issue. I want to upgrade the API soon and deprecate this one.

@s-banach, are you comfortable using asyncio for the speedup?

[s-banach]

Yes!

[ayjayt]

Alright, give me an hour or two and I"ll write and test a quickstart- it should make its way in to the README as well....

[ayjayt]

Hey, @s-banach, some news:

Basically the v1.0.0rc0 is the old API for backwards comparability with the new engine-> not great.
kaleido2 will be new api + new engine, I'm going to do it this weekend.

That said, I pushed a branch (https://github.com/plotly/Kaleido/tree/andrew/factor_for_faster) which allows you to run code like this:

  from choreographer import 
  import kaleido
  async with Browser() as browser:
    spec = scope.make_spec(fig, format="png") # serializes figure
    binary = await kaleido.to_image_w_browser(spec, browser)
    # something to write the binary to a file, async better

I used a blocking image write but still got 2.5X - it only uses one browser.

You can also wait for kaleido2 if you don't want to use a branch and this and it'll have a nice API, it's going to be a weekend sprint.

Sorry for the delay!