pd: allow bundling the web frontend into the fullnode #3522
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
hdevalence
temporarily deployed
to
smoke-test
GitHub Actions
Inactive
hdevalence
temporarily deployed
to
smoke-test
GitHub Actions
Inactive
hdevalence
commented
Dec 17, 2023
crates/bin/pd/src/main.rs
Outdated
Comment on lines
461
to
464
| // TODO: will need a bit of finesse to make this work with auto-https, | ||
| // since auto-https support above works with the tonic `serve_with_incoming` | ||
| // method, but now we're dropping down to axum directly. just needs some realignment | ||
| // probably. |
There was a problem hiding this comment.
In the ~year since this code was written, there may be better options. https://docs.rs/tokio-rustls-acme/latest/tokio_rustls_acme/ seems promising
|
opened #3646 to track landing this feature. |
cratelyn
added a commit
that referenced
this pull request
Jan 24, 2024
```
/!\ this is a work in progress and will /!\
/!\ be force pushed until ready for review /!\
```
fixes #3627.
this reorganizes the logic in pd's startup code related to automatically
managed https functionality. in particular, this is done in such a
manner as to expose an axum `Router` that will permit later work
(#3646, #3522) to serve other additional routes from our grpc
endpoint.
Refs: #3627
Refs: #3646
Refs: #3522
cratelyn
added a commit
that referenced
this pull request
Jan 24, 2024
```
/!\ this is a work in progress and will /!\
/!\ be force pushed until ready for review /!\
```
## 👀 overview
fixes #3627.
this reorganizes the logic in pd's startup code related to automatically
managed https functionality.
## 🎨 background & motivation
this PR, besides cleaning up the `rustls-acme`-related auto-https logic, is
also interested in *creating a state-of-affairs that will dovetail into
pr #3522*. in particular, this expression to start the GRPC serve given a
bound listener...
```rust
tokio::task::Builder::new()
.name("grpc_server")
.spawn(grpc_server.serve_with_incoming(tls_incoming))
.expect("failed to spawn grpc server")
```
...should be adjusted so as to accept an `axum::Router`.
other tertiary requirements:
- when no `grpc_auto_https` flag has been configured, serve without using an ACME resolver
- ALPN permit http/1.1 for grpc-web backwards compatibility.
### ⚖️ `rustls-acme` and `tokio-rustls-acme`
quoth the #3627 description, citing an earlier comment:
> In the ~year since this code was written, there may be better options.
> `tokio-rustls-acme` seems promising
\- <#3522 (comment)>
for reference, the repositories for each live here, and here:
- <https://github.com/FlorianUekermann/rustls-acme>
- <https://github.com/n0-computer/tokio-rustls-acme>
after some comparison, i have come to the opinion that `rustls-acme` will still
be adequate for our purposes. the latter is a fork of the former, but active
development appears to have continued in the former, and i did not see any
particular "_must-have_" features for us in the latter.
## 🚰 dropping down to `axum`
(this part is lengthy...)
as stated above, we want to switch to an `axum::Router`. this means that
we won't be able to use the `AcmeConfig::incoming` function. the
`rustls-acme` library provides some "low-level" examples we can check
out:
- <https://github.com/FlorianUekermann/rustls-acme/blob/main/examples/low_level_tokio.rs>
- <https://github.com/FlorianUekermann/rustls-acme/blob/main/examples/low_level_axum.rs>
`low_level_axum` sounds promising, given that #3522 itself drops down
from tonic into axum. a catch is that the "axum" example is referring to
`axum-server`, a library that wraps `axum`, rather than `axum` itself.
more on that momentarily.
we also use `tonic` 0.10.2 in pd, and elsewhere in the penumbra
monorepo. tonic isn't using hyper 1.x yet. this was being worked on in
hyperium/tonic#1583, continued on in hyperium/tonic#1595, and tracked in
hyperium/tonic#1579. that work also depends upon hyperium/hyper#3461.
this is complicated by the fact that `axum-server` 0.6.0 depends on
`hyper` 1.1.0. this is particularly relevant because of
hyperium/hyper#3040. so, we must be mindful that there is now a
`tower::Service` and a `hyper::Service`. `hyper-util` provides
facilities to convert between the two, and other helper glue.
additionally, see the "_upgrading"_ migration guide for more information
on the differences between pre- and post-1.0 interfaces in `hyper`.
- https://hyper.rs/guides/1/upgrading/
- <https://docs.rs/hyper-util/latest/hyper_util/service/struct.TowerToHyperService.html>
at a high level we need to take our `tonic::Router`, and turn that into
a `MakeService`, some object that can be used to _generate_ `Service`
instances to process a request.
this is additionally complicated by the fact that `axum-server` defines
its _own_ `MakeService` trait. that trait is sealed, blanket
implemented, and is a bound included in
`axum_server::server::Server::serve`.
```rust
// axum_server::serve
impl Server {
pub async fn serve<M>(self, mut make_service: M) -> io::Result<()>
where
M: MakeService<SocketAddr, Request<Incoming>>,
// ...
{ todo!() }
}
```
the `SocketAddr` parameter above is of importance.
`axum::Router::into_make_service` gives us `IntoMakeService<Router>`
- <https://docs.rs/axum/latest/axum/routing/struct.IntoMakeService.html>
```rust
impl<S, T> Service<T> for IntoMakeService<S>
where
S: Clone,
{
type Response = S;
type Error = Infallible;
type Future = IntoMakeServiceFuture<S>;
```
...which means that we are faced with this error:
```
error[E0277]: the trait bound `Router: tower_service::Service<http::request::Request<hyper::body::incoming::Incoming>>` is not satisfied
--> crates/bin/pd/src/auto_https.rs:62:20
|
62 | .serve(make_svc)
| ----- ^^^^^^^^ the trait `tower_service::Service<http::request::Request<hyper::body::incoming::Incoming>>` is not implemented for `Router`
| |
| required by a bound introduced by this call
|
= help: the trait `tower_service::Service<axum::http::Request<axum::body::Body>>` is implemented for `Router`
= help: for that trait implementation, expected `axum::http::Request<axum::body::Body>`, found `http::request::Request<hyper::body::incoming::Incoming>`
= note: required for `IntoMakeService<Router>` to implement `axum_server::service::MakeService<std::net::SocketAddr, http::request::Request<hyper::body::incoming::Incoming>>`
```
`axum::Router::into_make_service` does not play well with
`axum_server::Server`.
...and if we specify the new `hyper::body::incoming::Incoming` as the
body type for the axum router,
```
error[E0599]: the method `into_make_service` exists for struct `Router<(), Incoming>`, but its trait bounds were not satisfied
--> crates/bin/pd/src/auto_https.rs:57:31
|
57 | let make_svc = router.into_make_service();
| ^^^^^^^^^^^^^^^^^ method cannot be called on `Router<(), Incoming>` due to unsatisfied trait bounds
|
::: /home/katie/.cargo/registry/src/index.crates.io-6f17d22bba15001f/hyper-1.1.0/src/body/incoming.rs:47:1
|
47 | pub struct Incoming {
| ------------------- doesn't satisfy `hyper::body::Incoming: HttpBody`
|
= note: the following trait bounds were not satisfied:
`hyper::body::Incoming: HttpBody`
```
`HttpBody` was the pre-1.0 trait, that is now `Body` post-1.0. in short, there
are some incongruencies at play as different libraries catch up to the new
breaking changes in `hyper`.
## 🤨 ...so now what?
TODO(kate): write up course of action for tomorrow.
---
🟥 NB: i worry about breakage due to untested load-bearing behavior
during a migration like this. see "tertiary requirements", above.
consider this a PR to be suspect of when deploying a new testnet.
Refs: #3627
Refs: #3646
Refs: #3522
cratelyn
added a commit
that referenced
this pull request
Jan 25, 2024
## 👀 overview fixes #3627. this reorganizes the logic in pd's startup code related to automatically managed https functionality. ## 🎨 background & motivation this PR, besides cleaning up the `rustls-acme`-related auto-https logic, is also interested in *creating a state-of-affairs that will dovetail into pr #3522*. in particular, this expression to start the GRPC serve given a bound listener... ```rust tokio::task::Builder::new() .name("grpc_server") .spawn(grpc_server.serve_with_incoming(tls_incoming)) .expect("failed to spawn grpc server") ``` ...should be adjusted so as to work with an `axum::Router`. ### ⚖️ `rustls-acme` and `tokio-rustls-acme` quoth the #3627 description, citing an earlier comment: > In the ~year since this code was written, there may be better options. > `tokio-rustls-acme` seems promising \- <#3522 (comment)> for reference, the repositories for each live here, and here: - <https://github.com/FlorianUekermann/rustls-acme> - <https://github.com/n0-computer/tokio-rustls-acme> after some comparison, i have come to the opinion that `rustls-acme` will still be adequate for our purposes. the latter is a fork of the former, but active development appears to have continued in the former, and i did not see any particular "_must-have_" features for us in the latter. ## 🎴 changes this commit moves some of the auto-https related code from the `main` entrypoint, into standalone functions in `pd::main`. some constants are defined, to keep control flow clear and to help facilitate the addition of future options e.g. a flag to control the LetsEncrypt environment to use. ## 🚰 dropping down to `axum`; a brief note on future upgrades as stated above, we want to switch to an `axum::Router`. this means that we won't be able to use the `AcmeConfig::incoming` function. the `rustls-acme` library provides some "low-level" examples this work is based upon - <https://github.com/FlorianUekermann/rustls-acme/blob/main/examples/low_level_tokio.rs> - <https://github.com/FlorianUekermann/rustls-acme/blob/main/examples/low_level_axum.rs> we also use `tonic` 0.10.2 in pd, and elsewhere in the penumbra monorepo. tonic isn't using hyper 1.x yet. this was being worked on in hyperium/tonic#1583, continued on in hyperium/tonic#1595, and tracked in hyperium/tonic#1579. that work also depends upon hyperium/hyper#3461. we will need to be wait for tonic to finish its migration over to hyper 1.0, see: hyperium/tonic#1579 (comment) this is understandable, but i make note of this situation as a signpost for our future selves when considering a migration to recent versions of axum-server, axum, rustls-acme, or hyper. for now, it's easiest to stay in lock-step with tonic, and we can revisit the upgrade path(s) at a future date. === Refs: #3627 Refs: #3646 Refs: #3522
conorsch
pushed a commit
that referenced
this pull request
Jan 26, 2024
## 👀 overview fixes #3627. this reorganizes the logic in pd's startup code related to automatically managed https functionality. ## 🎨 background & motivation this PR, besides cleaning up the `rustls-acme`-related auto-https logic, is also interested in *creating a state-of-affairs that will dovetail into pr #3522*. in particular, this expression to start the GRPC serve given a bound listener... ```rust tokio::task::Builder::new() .name("grpc_server") .spawn(grpc_server.serve_with_incoming(tls_incoming)) .expect("failed to spawn grpc server") ``` ...should be adjusted so as to work with an `axum::Router`. ### ⚖️ `rustls-acme` and `tokio-rustls-acme` quoth the #3627 description, citing an earlier comment: > In the ~year since this code was written, there may be better options. > `tokio-rustls-acme` seems promising \- <#3522 (comment)> for reference, the repositories for each live here, and here: - <https://github.com/FlorianUekermann/rustls-acme> - <https://github.com/n0-computer/tokio-rustls-acme> after some comparison, i have come to the opinion that `rustls-acme` will still be adequate for our purposes. the latter is a fork of the former, but active development appears to have continued in the former, and i did not see any particular "_must-have_" features for us in the latter. ## 🎴 changes this commit moves some of the auto-https related code from the `main` entrypoint, into standalone functions in `pd::main`. some constants are defined, to keep control flow clear and to help facilitate the addition of future options e.g. a flag to control the LetsEncrypt environment to use. ## 🚰 dropping down to `axum`; a brief note on future upgrades as stated above, we want to switch to an `axum::Router`. this means that we won't be able to use the `AcmeConfig::incoming` function. the `rustls-acme` library provides some "low-level" examples this work is based upon - <https://github.com/FlorianUekermann/rustls-acme/blob/main/examples/low_level_tokio.rs> - <https://github.com/FlorianUekermann/rustls-acme/blob/main/examples/low_level_axum.rs> we also use `tonic` 0.10.2 in pd, and elsewhere in the penumbra monorepo. tonic isn't using hyper 1.x yet. this was being worked on in hyperium/tonic#1583, continued on in hyperium/tonic#1595, and tracked in hyperium/tonic#1579. that work also depends upon hyperium/hyper#3461. we will need to be wait for tonic to finish its migration over to hyper 1.0, see: hyperium/tonic#1579 (comment) this is understandable, but i make note of this situation as a signpost for our future selves when considering a migration to recent versions of axum-server, axum, rustls-acme, or hyper. for now, it's easiest to stay in lock-step with tonic, and we can revisit the upgrade path(s) at a future date. === Refs: #3627 Refs: #3646 Refs: #3522
cratelyn
added
A-node
|
Status page completed and ready to be added to pd: penumbra-zone/web#592 |
|
Great! I have a rebase of this branch in progress, I'll pick it up next. |
|
so excited to see this work moving forward! 🚀 |
hdevalence
force-pushed
the
bundle-frontend-into-pd
branch
from
8042942 to
b74ca3a
Compare
hdevalence
had a problem deploying
to
smoke-test
GitHub Actions
Failure
hdevalence
force-pushed
the
bundle-frontend-into-pd
branch
from
b74ca3a to
1ab2e6a
Compare
hdevalence
had a problem deploying
to
smoke-test
GitHub Actions
Error
|
Here's a video of this code in action: Screen.Recording.2024-02-25.at.11.29.10.AM.movNote that sending a transaction fails because the web extension hasn't been updated to the new proving parameters. However, it otherwise seems to work! |
This is an experiment towards bundling the web frontend code into the fullnode,
with the goal that users who run `pd` nodes can self-host a complete Penumbra
frontend, on their custom domain with auto-HTTPS, with just one command: `pd start`.
To do this, we
- add a new `zipserve` module in `pd`, which takes a `&'static [u8]` with archive data and produces an `axum::Router`;
- bundle an `assets/{minifront,node_status}.zip` (tracked with git-lfs) into the `pd` crate;
- use `axum::Router::merge` to combine the GRPC and frontend endpoints.
hdevalence
force-pushed
the
bundle-frontend-into-pd
branch
from
1ab2e6a to
b95bd84
Compare
hdevalence
temporarily deployed
to
smoke-test
GitHub Actions
Inactive
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This is an experiment towards bundling the web frontend code into the fullnode, with the goal that users who run
pdnodes can self-host a complete Penumbra frontend, on their custom domain with auto-HTTPS, with just one command:pd start.To do this, we
zipservemodule inpd, which takes a&'static [u8]with archive data and produces anaxum::Router;assets/frontend.zip(tracked with git-lfs) into thepdcrate;axum::Router::mergeto combine the GRPC and frontend endpoints.To make this work for real, we'd need to:
/appwhen served inpd, to allow a status page, but/when atapp.testnet.penumbra.zoneor other deployments);assets/frontend.zipfrom thewebrepo periodically and check the result into the repo (under LFS).These seem pretty doable, and it'd be really cool to be able to have easily self-hosted frontends!