perf: lazy bootstrap options - first pass #21164
Conversation
littledivy
changed the title
littledivy
changed the title
|
What's the impact on startup time? |
|
/bench startup_nop_trivial |
startup_nop_trivial
start: id: server: |
|
Not significant. We'll see big improvement once we get rid of the |
| #[op2] | ||
| #[string] | ||
| pub fn op_bootstrap_user_agent(state: &mut OpState) -> String { | ||
| state.borrow::<BootstrapOptions>().user_agent.clone() | ||
| } |
There was a problem hiding this comment.
More for information than a request for a change:
Given that we have to clone strings to return them and that the ops are not fast, I think it would be reasonable if you want to return the entire BootstrapOptions as a #[serde] return since it's a one-off on first use. It might simplify the code overall and I don't think it'll have a measurable impact on boot time.
There was a problem hiding this comment.
Sure that's an option too. Let's do that afterwards maybe? I don't want to solidify this approach until we get to see some improvements after the removal of bootstrap call (proabably a few more PRs before that's possible)
There was a problem hiding this comment.
Sounds good... merge when you're ready!
I need to turn off the zero-decimal place rounding here on kijunsan.... |
|
/bench startup_nop_trivial |
startup_nop_trivial
start: id: server: |
Move most runtime options to be lazily loaded. Constant options will be covered in a different PR. Towards #21133
Move most runtime options to be lazily loaded. Constant options will be covered in a different PR. Towards denoland#21133
Move most runtime options to be lazily loaded. Constant options will be covered in a different PR.
Towards #21133