Commit Graph

169 Commits

Author SHA1 Message Date
Sean McArthur
e455fa2452 fix(client): fix connection leak when Response finishes before Request body
If the Response was received and the body finished while the Request
body was still streaming, the connection could get into a state where it
was never polled again, thus never re-inserting into the connection pool
or being dropped.

Closes #1717
2018-11-21 16:39:47 -08:00
Sean McArthur
9245e9409a fix(header): fix panic when parsing header names larger than 64kb 2018-11-06 14:37:13 -08:00
Sean McArthur
976a77a673 feat(client): add ALPN h2 support for client connectors
- Adds `Connected::negotiated_h2()` method to signal the connection must
  use HTTP2. `Connect` implementations should set this if using ALPN.

If a connection to a host is detected to have been upgraded via ALPN,
any other oustanding connect futures will be canceled, and the waiting
requests will make use of the single HTTP2 connection.

The `http2_only` builder configuration still works the same, not
requiring ALPN at all, and always using only a single connection.
2018-10-31 14:51:29 -07:00
Sean McArthur
5ca2905c84 test(server): server test cleanup 2018-10-26 12:50:27 -07:00
Sean McArthur
2d5af177c1 feat(client): add Resolve, used by HttpConnector
This introduces a `Resolve` trait to describe asynchronous DNS
resolution. The `HttpConnector` can be configured with a resolver,
allowing a user to still use all the functionality of the
`HttpConnector`, while customizing the DNS resolution.

To prevent a breaking change, the `HttpConnector` has its `Resolve`
generic set by default to `GaiResolver`. This is same as the existing
resolver, which uses `getaddrinfo` inside a thread pool.

Closes #1517
2018-10-18 12:10:15 -07:00
Sean McArthur
13d53e1d0c feat(client): adds HttpInfo to responses when HttpConnector is used
- Adds `client::connect::Connected::extra()`, which allows connectors to
  specify arbitrary custom information about a connected transport.

If a connector provides this extra value, it will be set in the
`Response` extensions.

Closes #1402
2018-10-16 14:40:50 -07:00
Sean McArthur
37ec724fd6 fix(http2): add Date header if not present for HTTP2 server responses 2018-10-08 17:52:08 -07:00
Sean McArthur
0cfa7231b2 test(client): update from deprecated export of ConnectFuture 2018-09-27 15:29:57 -07:00
Theodore Cipicchio
24f11a421d fix(http2): allow TE "trailers" request headers
The HTTP/2 spec allows TE headers in requests if the value is
"trailers". Other TE headers are still stripped.

Closes #1642
2018-08-27 10:32:53 -07:00
lambdasqd
1448e4067b fix(server): properly handle keep-alive for HTTP/1.0
Change behaviour of connection or server response when the request is
version 1.0 and the Connection: keep-alive header is not present.

1. If the response is also version 1.0, then connection is closed if the
server keep-alive header is not present.
2. If the response is version 1.1, then the keep-alive header is added
when downgrading to version 1.0.

Closes #1614
2018-08-15 12:10:03 -07:00
Sean McArthur
195fbb2a37 fix(server): coerce responses with HTTP2 version to HTTP/1.1 when protocol is 1.x 2018-08-10 13:08:08 -07:00
Sean McArthur
853266d873 test(server): add http1_only server test 2018-08-10 13:08:08 -07:00
Sean McArthur
6530a00a8e fix(http1): reduce closed connections when body is dropped
If a user makes use of `Body::is_end_stream` to optimize so as to not
need to do make a final poll just to receive `None`, previously the
connection would not have progressed its reading state to a finished
body, and so the connection would be closed.

Now, upon reading any chunk, the connection state will check if it
can know that the body would be finished, and progresses to a body
finished state sooner.

The integration tests were amplified by adding a naive hyper proxy
as a secondary test, which happens to make use of that optimization,
and thus caught the issue.
2018-07-23 09:39:19 -07:00
Sean McArthur
1f95f58837 chore(error): improve message about incomplete parsed messages 2018-06-25 17:29:40 -07:00
Sean McArthur
e61fe54093 test(server): fix unused result from block_on 2018-06-25 11:48:52 -07:00
Sean McArthur
6848ba63fb test(server): use current_thread instead of rt::run with http2 tests 2018-06-25 11:35:57 -07:00
Sean McArthur
2a3844acc3 feat(client): enable CONNECT requests through the Client
While the upgrades feature enabled HTTP upgrades in both and the server and client, and the goal was for `CONNECT` requests to work as well, only the server could use them for `CONNECT`. The `Client` had some specific code rejecting `CONNECT` requests, and this removes it and prepares the `Client` to handle them correctly.
2018-06-22 21:00:28 -07:00
Sean McArthur
e4ebf44823 chore(tests): change tests to use current_thread runtime 2018-06-18 12:30:56 -07:00
Josh Leeb-du Toit
9a28268b98 feat(http2): Add content_length() value to incoming h2 Body
- Add `Body::Kind::H2` to contain the content length of the body.
- Update `Body::content_length` to return the content length if `Body::Kind` is `H2`, instead of returning `None`.

Reference: #1556, #1557

Closes #1546
2018-06-18 11:50:12 -07:00
Sean McArthur
fea29b29e2 feat(http1): Add higher-level HTTP upgrade support to Client and Server (#1563)
- Adds `Body::on_upgrade()` that returns an `OnUpgrade` future.
- Adds `hyper::upgrade` module containing types for dealing with
  upgrades.
- Adds `server::conn::Connection::with_upgrades()` method to enable
  these upgrades when using lower-level API (because of a missing
  `Send` bound on the transport generic).
- Client connections are automatically enabled.
- Optimizes request parsing, to make up for extra work to look for
  upgrade requests.
  - Returns a smaller `DecodedLength` type instead of the fatter
    `Decoder`, which should also allow a couple fewer branches.
  - Removes the `Decode::Ignore` wrapper enum, and instead ignoring
    1xx responses is handled directly in the response parsing code.

Ref #1563 

Closes #1395
2018-06-14 13:39:29 -07:00
Laurențiu Nicola
386fc0d70b feat(http2): set Content-Length header on outgoing messages
Closes #1547
2018-06-10 15:22:30 -07:00
Josh Leeb-du Toit
f20afba57d feat(http2): Strip connection headers before sending
Automatically removes "connection" headers before sending over HTTP2.
These headers are illegal in HTTP2, and would otherwise cause errors.

Closes: #1551
2018-06-08 17:31:22 -07:00
Laurențiu Nicola
a0a0fcdd9b feat(body): make Body know about incoming Content-Length
When getting a `Body` from hyper, such as in a client response,
the method `Body::content_length()` now returns a value if the header
was present.

Closes #1545
2018-06-08 13:00:46 -07:00
Sean McArthur
d7ab016676 fix(server): correctly handle CONNECT requests
- In the higher-level `Server` API, since connection upgrades aren't yet
  supported, returning a 2xx response to a `CONNECT` request is a user
  error. A 500 response is written to the client, the connection is
  closed, and an error is reported back to the user.
- In the lower-level `server::Connection` API, where upgrades *are*
  supported, a 2xx response correctly marks the response as the final
  one, instead of trying to parse more requests afterwards.
2018-06-07 14:59:01 -07:00
Steven Fackler
a096799c1b feat(body): add Sender::abort
This allows a client or server to indicate that the body should be cut off
in an abnormal fashion so the server doesn't simply get a "valid" but
truncated body.
2018-06-05 17:09:31 -07:00
Sean McArthur
898e919504 perf(h1): optimize for when Body is only 1 chunk
- When the `Body` is created from a buffer of bytes (such as
  `Body::from("hello")`), we can skip some bookkeeping that is
  normally required for streaming bodies.
- Orthogonally, optimize encoding body chunks when the strategy
  is to flatten into the headers buf, by skipping the EncodedBuf
  enum.
2018-05-31 19:27:24 -07:00
Sean McArthur
26417fc24a perf(h1): improve parsing and encoding of http1 messages 2018-05-15 13:24:58 -07:00
estk
bc6af88a32 feat(server): support HTTP1 and HTTP2 automatically
If an HTTP/1 connection has a parse error, but it starts with the HTTP2 preface, converts the connection automatically into an HTTP2 server connection.

Closes #1486
2018-05-10 14:23:42 -07:00
Sean McArthur
18f4dd2406 chore(ci): remove flaky tests from failing CI 2018-05-09 16:02:49 -07:00
Sean McArthur
190a8501d6 docs(client): add module level docs for hyper::client 2018-05-02 13:32:46 -07:00
Sean McArthur
7a7453ba52 refactor(lib): change from futures-timer to tokio-timer 2018-04-30 19:11:05 -07:00
Sean McArthur
11c92ff467 chore(tests): set a short timeout on the integration clients pool 2018-04-26 12:37:31 -07:00
Matt Bilker
a02fec8c78 feat(client): add support for title case header names (#1497)
This introduces support for the HTTP/1 Client to write header names as title case when encoding
the request.

Closes #1492
2018-04-24 16:41:02 -07:00
Evan Simmons
aac250f29d fix(server): panic on max_buf_size too small 2018-04-19 12:27:11 -07:00
Sean McArthur
a4eb612bd5 tests(http2): add bodies to parallel h2 tests 2018-04-18 16:34:44 -07:00
Sean McArthur
2dc6202fe7 feat(service): introduce hyper-specific Service
This introduces the `hyper::service` module, which replaces
`tokio-service`.

Since the trait is specific to hyper, its associated
types have been adjusted. It didn't make sense to need to define
`Service<Request=http::Request>`, since we already know the context is
HTTP. Instead, the request and response bodies are associated types now,
and slightly stricter bounds have been placed on `Error`.

The helpers `service_fn` and `service_fn_ok` should be sufficient for
now to ease creating `Service`s.

The `NewService` trait now allows service creation to also be
asynchronous.

These traits are similar to `tower` in nature, and possibly will be
replaced completely by it in the future. For now, hyper defining its own
allows the traits to have better context, and prevents breaking changes
in `tower` from affecting hyper.

Closes #1461

BREAKING CHANGE: The `Service` trait has changed: it has some changed
  associated types, and `call` is now bound to `&mut self`.

  The `NewService` trait has changed: it has some changed associated
  types, and `new_service` now returns a `Future`.

  `Client` no longer implements `Service` for now.

  `hyper::server::conn::Serve` now returns `Connecting` instead of
  `Connection`s, since `new_service` can now return a `Future`. The
  `Connecting` is a future wrapping the new service future, returning
  a `Connection` afterwards. In many cases, `Future::flatten` can be
  used.
2018-04-17 17:09:15 -07:00
Sean McArthur
c4974500ab feat(server): re-design Server as higher-level API
The `hyper::Server` is now a proper higher-level API for running HTTP
servers. There is a related `hyper::server::Builder` type, to construct
a `Server`. All other types (`Http`, `Serve`, etc) were moved into the
"lower-level" `hyper::server::conn` module.

The `Server` is a `Future` representing a listening HTTP server. Options
needed to build one are set on the `Builder`.

As `Server` is just a `Future`, it no longer owns a thread-blocking
executor, and can thus be run next to other servers, clients, or
what-have-you.

Closes #1322
Closes #1263

BREAKING CHANGE: The `Server` is no longer created from `Http::bind`,
  nor is it `run`. It is a `Future` that must be polled by an
  `Executor`.

  The `hyper::server::Http` type has move to
  `hyper::server::conn::Http`.
2018-04-16 14:29:19 -07:00
Sean McArthur
c119097fd0 feat(http2): add HTTP/2 support for Client and Server 2018-04-13 14:23:47 -07:00
Sean McArthur
fe1578acf6 feat(client): update construction of Clients
- `Client::new()` no longer needs a `Handle`, and instead makes use of
  tokio's implicit default.
- Changed `Client::configure()` to `Client::builder()`.
- `Builder` is a by-ref builder, since all configuration is now
  cloneable pieces.

BREAKING CHANGE: `Client:new(&handle)` and `Client::configure()` are now
  `Client::new()` and `Client::builder()`.
2018-04-10 17:30:10 -07:00
Sean McArthur
dfdca25c00 feat(body): rename Entity to Payload
Closes #1464
2018-04-10 15:55:23 -07:00
Sean McArthur
5d3c472228 feat(error): revamp hyper::Error type
**The `Error` is now an opaque struct**, which allows for more variants to
be added freely, and the internal representation to change without being
breaking changes.

For inspecting an `Error`, there are several `is_*` methods to check for
certain classes of errors, such as `Error::is_parse()`. The `cause` can
also be inspected, like before. This likely seems like a downgrade, but
more inspection can be added as needed!

The `Error` now knows about more states, which gives much more context
around when a certain error occurs. This is also expressed in the
description and `fmt` messages.

**Most places where a user would provide an error to hyper can now pass
any error type** (`E: Into<Box<std::error::Error>>`). This error is passed
back in relevant places, and can be useful for logging. This should make
it much clearer about what error a user should provide to hyper: any it
feels is relevant!

Closes #1128
Closes #1130
Closes #1431
Closes #1338

BREAKING CHANGE: `Error` is no longer an enum to pattern match over, or
  to construct. Code will need to be updated accordingly.

  For body streams or `Service`s, inference might be unable to determine
  what error type you mean to return. Starting in Rust 1.26, you could
  just label that as `!` if you never return an error.
2018-04-10 14:29:34 -07:00
Sean McArthur
c210524e94 chore(tests): fix tokio runtime deprecations 2018-04-10 12:56:55 -07:00
Sean McArthur
625e4daaa1 Revert "refactor(lib): convert to futures 0.2.0-beta (#1470)"
This reverts commit a12f7beed9.

Much sadness 😢.
2018-04-10 12:56:55 -07:00
Sam Rijs
a12f7beed9 refactor(lib): convert to futures 0.2.0-beta (#1470) 2018-03-29 13:32:44 -07:00
Sam Reis
27b8db3af8 feat(lib): convert to use tokio 0.1
BREAKING CHANGE: All uses of `Handle` now need to be new-tokio `Handle`.

Co-authored-by: Sean McArthur <sean@seanmonstar.com>
2018-03-19 11:43:47 -07:00
Sean McArthur
8c52c2dfd3 feat(client): redesign the Connect trait
The original `Connect` trait had some limitations:

- There was no way to provide more details to the connector about how to
  connect, other than the `Uri`.
- There was no way for the connector to return any extra information
  about the connected transport.
- The `Error` was forced to be an `std::io::Error`.
- The transport and future had `'static` requirements.

As hyper gains HTTP/2 support, some of these things needed to be
changed. We want to allow the user to configure whether they hope to
us ALPN to start an HTTP/2 connection, and the connector needs to be
able to return back to hyper if it did so.

The new `Connect` trait is meant to solve this.

- The `connect` method now receives a `Destination` type, instead of a
  `Uri`. This allows us to include additional data about how to connect.
- The `Future` returned from `connect` now must be a tuple of the
  transport, and a `Connected` metadata value. The `Connected` includes
  possibly extra data about what happened when connecting.

BREAKING CHANGE: Custom connectors should now implement `Connect`
  directly, instead of `Service`.

  Calls to `connect` no longer take `Uri`s, but `Destination`. There
  are `scheme`, `host`, and `port` methods to query relevant
  information.

  The returned future must be a tuple of the transport and `Connected`.
  If no relevant extra information is needed, simply return
  `Connected::new()`.

Closes #1428
2018-03-19 11:43:47 -07:00
Sean McArthur
fbc449e49c feat(body): introduce an Entity trait to represent bodies
This dedicated `Entity` trait replaces the previous `Stream<Item=impl
AsRef<[u8]>, Error=hyper::Error>`. This allows for several improvements
immediately, and prepares for HTTP2 support.

- The `Entity::is_end_stream` makes up for change away from
  `Option<Body>`, which was previously used to know if the body should be
  empty. Since `Request` and `Response` now require a body to be set,
  this method can be used to tell hyper that the body is actually empty.

  It also provides the possibility of slight optimizations when polling
  for data, by allowing to check `is_end_stream` before polling again.
  This can allow a consumer to know that a body stream has ended without
  polling for `None` afterwards.

- The `Entity::content_length` method allows a body to automatically
  declare a size, in case a user doesn't set a `Content-Length` or
  `Transfer-Encoding` header.

- It's now possible to send and receive trailers, though this will be
  for HTTP2 connections only.

By being a trait owned by hyper, new methods can be added later as new
features are wanted (with default implementations).

The `hyper::Body` type now implements `Entity` instead of `Stream`,
provides a better channel option, and is easier to use with custom
streams via `Body::wrap_stream`.

BREAKING CHANGE: All code that was assuming the body was a `Stream` must
  be adjusted to use an `Entity` instead.

  Using `hyper::Body` as a `Stream` can call `Body::into_stream`
  to get a stream wrapper.

  Passing a custom `impl Stream` will need to either implement
  `Entity`, or as an easier option, switch to `Body::wrap_stream`.

  `Body::pair` has been replaced with `Body::channel`, which returns a
  `hyper::body::Sender` instead of a `futures::sync::mpsc::Sender`.

Closes #1438
2018-03-19 11:43:47 -07:00
Sean McArthur
3cd48b45fb feat(lib): replace types with those from http crate
BREAKING CHANGE: `Method`, `Request`, `Response`, `StatusCode`,
  `Version`, and `Uri` have been replaced with types from the `http`
  crate. The `hyper::header` module is gone for now.

  Removed `Client::get`, since it needed to construct a `Request<B>`
  with an empty body. Just use `Client::request` instead.

  Removed `compat` cargo feature, and `compat` related API.
2018-03-19 11:43:47 -07:00
Sean McArthur
91b9700862 refactor(client): only spawn pooled in executor if not ready when response recieved 2018-03-13 18:46:52 -07:00
Sean McArthur
d58aa73246 feat(server): add upgrade support to lower-level Connection API (#1459)
Closes #1323
2018-03-09 10:05:27 -08:00