-
Notifications
You must be signed in to change notification settings - Fork 344
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
Upgrade to Wasmer Reborn (part 1) #504
Conversation
7eb1351
to
8685586
Compare
3274902
to
cc94242
Compare
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.
Impressive PR.
I wrote a lot of questions that came up trying to understand the new APIs and your usage of them better. However, the code looks solid. Happy for another pair of eyes on this, but from my end, I see nothing blocking merging this.
Also, very curious as to how the benchmarks change when this is merged. Would be great to have a before/after diagram (or some tracking of it)
@@ -22,12 +22,13 @@ incremental = false | |||
overflow-checks = true | |||
|
|||
[features] | |||
# Change this to [] if you don't need Windows support and want faster integration tests. |
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.
This only affects the tests, right?
Meaning, cargo test
works out of the box on all platforms, but linux/osx can opt into cargo test --no-default-features
for speed
The alternative would be default [], cargo test
does the right thing on osx/linux, and you would need cargo test --features cranelift
to run on windows at all.
I am leaning towards the second as a pattern. But it would be cool to see if that could be done automatically for both.
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.
Hmmm.. wasted some time reflecting on Cargo files.
There is eg:
[target.x86_64-pc-windows-gnu.dependencies]
byte = "0.2.4"
But I cannot find the equivalent for enabling target-specific feature flags.
Minor point, but it would good to set a proper template for all the contracts
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.
This is only relevant for integration tests. Unit tests are not affected.
The alternative would be default [], cargo test does the right thing on osx/linux, and you would need cargo test --features cranelift to run on windows at all.
Jupp, but it also requires different documentation. I went for the less optimized but more beginner friendly way. But open for change.
Hmmm.. wasted some time reflecting on Cargo files.
See also #649, which is probably what you are looking for
@@ -125,7 +126,7 @@ where | |||
} | |||
|
|||
// Get module from file system cache | |||
if let Some(module) = self.fs_cache.load(checksum)? { | |||
if let Some(module) = self.fs_cache.load(checksum, options.memory_limit)? { |
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.
This take an optional config memory limit (very nice), why not config above?
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.
This limit is not optional but required for every instanciation.
You can have a single cache and get instances with different memory limits out of it. Not that we need it but I ran into design trouble when I had a fixed limit in the cache.
@@ -476,7 +479,7 @@ mod test { | |||
} | |||
|
|||
#[test] | |||
#[cfg(feature = "default-singlepass")] | |||
#[cfg(feature = "metering")] |
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.
Okay, so this code is just left as unmigrated placeholders and fails if metering flag is enabled now?
(Which is good to push into a separate PR)
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.
Jupp, at some point I realized that adding a simple metering flag makes almost everything compile and pass.
@@ -0,0 +1,170 @@ | |||
use crate::backend::{Querier, Storage}; |
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.
This whole file is a TODO/placeholder, right? (I know you said metering was still pending upstream)
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.
Right, lots of noop and outdated code here
/// | ||
/// Delegated to base. | ||
fn memory_style(&self, memory: &MemoryType) -> MemoryStyle { | ||
let adjusted = self.adjust_memory(memory); |
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.
no need to validate here?
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.
We validate everything we need to when storing the code the first time. Then self.adjust_memory
just sets the maximum for the memory.
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.
Ah, now I know what you mean.
The result type is fixed by the Tunables
trait in Wasmer and I can't return errors. But everything that goes through here also calls one of the other functions.
Self { limit, base } | ||
} | ||
|
||
/// Takes in input memory type as requested by the guest and sets |
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.
Great comments and clear logic here
} | ||
} | ||
|
||
impl<T: Tunables> Tunables for LimitingTunables<T> { |
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.
I am unclear from here what is called when a new page is allocated in an existing dynamic memory.
Or no need to intercept that, as we ensure valid minimum/maximum when they create the memory table in the first place, so the internal logic can handle dynamic page allocation?
Also, no intercept on pages means we cannot charge gas per memory usage, right? (No need to do so, I just remember the idea was floating around and curious if possible with the current API)
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.
Right, we set a maximum once and Wasmer checks that the number of pages does not exceed the maximum internally.
Also, no intercept on pages means we cannot charge gas per memory usage, right?
Basically yes.
In the new metering we could implement special treatment for the memory grop op code. But then we also need to look at the minimum. So if we really really want, it can be done. But I'd say memory usage if free within the allowed range.
} | ||
|
||
/// Created a store with no compiler and the given memory limit (in pages) | ||
pub fn make_store_headless(memory_limit: Size) -> Store { |
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.
Ah, headless can run pre-compiled code, but not compile itself, which makes it lighter to start up when running many instances, right?
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.
Yes. This is also usefull for cross compilation, e.g. for IoT stuff. You compile one machine and run in a much more limited environment.
Closes #503
Closes #462
Closes #501
Closes #375
Closes #495
Closes #555