-
Notifications
You must be signed in to change notification settings - Fork 29.8k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
process: js fast path for cached bindings #18365
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Rest LGTM
lib/internal/bootstrap_node.js
Outdated
|
||
process.binding = function binding(module) { | ||
if (!module || typeof module !== 'string') | ||
return; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It’s unfortunate, but process.binding
is used in the wild too often for us to be able to just return undefined if the requested module is undefined. We should cast it to a string just as the C++ version did.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Also, when would core code call it with something that is not a string? This guard should probably just go.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
If you're adding caching in JS land, then I'd get rid of env->binding_cache_object()
+env->internal_binding_cache_object()
and strip down node::Binding()
and node::InternalBinding()
to their bare essentials.
lib/internal/bootstrap_node.js
Outdated
|
||
process.binding = function binding(module) { | ||
if (!module || typeof module !== 'string') | ||
return; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Also, when would core code call it with something that is not a string? This guard should probably just go.
@TimothyGu @bnoordhuis Thanks for the feedback. I've updated the PR based on your suggestions. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM modulo comment. Binding()
and InternalBinding()
have quite some shared logic you could factor out.
lib/internal/bootstrap_node.js
Outdated
if (typeof bindingObj[module] === 'object') | ||
return bindingObj[module]; | ||
bindingObj[module] = getLinkedBinding(module); | ||
return bindingObj[module]; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Minor anti-pattern, looking up twice. Do this instead:
let mod = bindingObj[module];
if (typeof mod !== 'object')
mod = bindingObj[module] = getBinding(module);
return mod;
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Good catch, don't know what I was thinking :) Updated.
f45097e
to
ea60d2c
Compare
Ok, I've added one final change in terms of moving |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks nice.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Nice, this also gives me a good idea where to monkey-patch the deprecate bindings....
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Nice catch.
Landed in bb27bfe |
Currently, both process.binding and internalBinding have to call into C++ regardless of whether the module has been cached or not. This creates significant overhead to all binding calls and unfortunately the rest of the codebase doesn't really optimize the amount of times that bindings are required (as an example: 12 files require the async_wrap binding). Changing all the usage of this function throughout the codebase would introduce a lot of churn (and is kind of a hassle) so instead this PR introduces a JS fast path for both functions for cases where the binding has already been cached. While micro benchmarks are not super meaningful here (it's not like we call binding that often...), this does speed up the cached call by 400%. In addition, move moduleLoadList creation and management entirely into JS-land as it requires less code and is more efficient. PR-URL: #18365 Reviewed-By: Tiancheng "Timothy" Gu <timothygu99@gmail.com> Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl> Reviewed-By: Joyee Cheung <joyeec9h3@gmail.com> Reviewed-By: Anna Henningsen <anna@addaleax.net> Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Jeremiah Senkpiel <fishrock123@rocketmail.com>
Currently, both process.binding and internalBinding have to call into C++ regardless of whether the module has been cached or not. This creates significant overhead to all binding calls and unfortunately the rest of the codebase doesn't really optimize the amount of times that bindings are required (as an example: 12 files require the async_wrap binding). Changing all the usage of this function throughout the codebase would introduce a lot of churn (and is kind of a hassle) so instead this PR introduces a JS fast path for both functions for cases where the binding has already been cached. While micro benchmarks are not super meaningful here (it's not like we call binding that often...), this does speed up the cached call by 400%. In addition, move moduleLoadList creation and management entirely into JS-land as it requires less code and is more efficient. PR-URL: #18365 Reviewed-By: Tiancheng "Timothy" Gu <timothygu99@gmail.com> Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl> Reviewed-By: Joyee Cheung <joyeec9h3@gmail.com> Reviewed-By: Anna Henningsen <anna@addaleax.net> Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Jeremiah Senkpiel <fishrock123@rocketmail.com>
Does this make sense for LTS? |
@MylesBorins I think it makes sense to backport this to v8.x, I am trying to backport #19112 to v8.x but a test fails because it depends on this to get a complete fix. @apapirovski are you working on a backport? I can help backporting this one in order to get #19112 backported. |
Currently, both process.binding and internalBinding have to call into C++ regardless of whether the module has been cached or not. This creates significant overhead to all binding calls and unfortunately the rest of the codebase doesn't really optimize the amount of times that bindings are required (as an example: 12 files require the async_wrap binding). Changing all the usage of this function throughout the codebase would introduce a lot of churn (and is kind of a hassle) so instead this PR introduces a JS fast path for both functions for cases where the binding has already been cached. While micro benchmarks are not super meaningful here (it's not like we call binding that often...), this does speed up the cached call by 400%. In addition, move moduleLoadList creation and management entirely into JS-land as it requires less code and is more efficient. PR-URL: nodejs#18365 Reviewed-By: Tiancheng "Timothy" Gu <timothygu99@gmail.com> Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl> Reviewed-By: Joyee Cheung <joyeec9h3@gmail.com> Reviewed-By: Anna Henningsen <anna@addaleax.net> Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Jeremiah Senkpiel <fishrock123@rocketmail.com>
@joyeecheung if you could backport to 8.x that would be great. should we set |
Currently, both process.binding and internalBinding have to call into C++ regardless of whether the module has been cached or not. This creates significant overhead to all binding calls and unfortunately the rest of the codebase doesn't really optimize the amount of times that bindings are required (as an example: 12 files require the async_wrap binding). Changing all the usage of this function throughout the codebase would introduce a lot of churn (and is kind of a hassle) so instead this PR introduces a JS fast path for both functions for cases where the binding has already been cached. While micro benchmarks are not super meaningful here (it's not like we call binding that often...), this does speed up the cached call by 400%. In addition, move moduleLoadList creation and management entirely into JS-land as it requires less code and is more efficient. PR-URL: nodejs#18365 Reviewed-By: Tiancheng "Timothy" Gu <timothygu99@gmail.com> Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl> Reviewed-By: Joyee Cheung <joyeec9h3@gmail.com> Reviewed-By: Anna Henningsen <anna@addaleax.net> Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Jeremiah Senkpiel <fishrock123@rocketmail.com>
ping @joyeecheung regarding a 8.x backport |
Currently, both process.binding and internalBinding have to call into C++ regardless of whether the module has been cached or not. This creates significant overhead to all binding calls and unfortunately the rest of the codebase doesn't really optimize the amount of times that bindings are required (as an example: 12 files require the async_wrap binding). Changing all the usage of this function throughout the codebase would introduce a lot of churn (and is kind of a hassle) so instead this PR introduces a JS fast path for both functions for cases where the binding has already been cached. While micro benchmarks are not super meaningful here (it's not like we call binding that often...), this does speed up the cached call by 400%. In addition, move moduleLoadList creation and management entirely into JS-land as it requires less code and is more efficient. PR-URL: nodejs#18365 Reviewed-By: Tiancheng "Timothy" Gu <timothygu99@gmail.com> Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl> Reviewed-By: Joyee Cheung <joyeec9h3@gmail.com> Reviewed-By: Anna Henningsen <anna@addaleax.net> Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Jeremiah Senkpiel <fishrock123@rocketmail.com>
Currently, both
process.binding
andinternalBinding
have to call into C++ regardless of whether the module has been cached or not. This creates significant overhead to all binding calls and unfortunately the rest of the codebase doesn't really optimize the amount of times that bindings are required (as an example: 12 files require the async_wrap binding).Changing all the usage of this function throughout the codebase would introduce a lot of churn (and is kind of a hassle) so instead this PR introduces a JS fast path for both functions for cases where the binding has already been cached. While micro benchmarks are not super meaningful here (it's not like we call
binding
that often...), this does speed up the cached call by 400%.Checklist
make -j4 test
(UNIX), orvcbuild test
(Windows) passesAffected core subsystem(s)
process, src