When running tests locally, this error can surface.
```
---- test::agent_test::custom_resolver stdout ----
thread 'test::agent_test::custom_resolver' panicked at 'called `Result::unwrap()` on an `Err` value: Os { code: 22, kind: InvalidInput, message: "Invalid argument" }', src/stream.rs:60:13
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
```
The problem is that setting the timeouts might fail, and this is done
in a From trait where there is not possibility to "bubble" the
io::Error.
```
socket.set_read_timeout(None).unwrap();
socket.set_write_timeout(None).unwrap();
```
This commit moves the resetting of timers to an explicit `Stream::reset()` fn
that must be called every time we're unwrapping the inner stream.
Previously we had a special case for BadStatusRead that would happen
only when we got a ConnectionAborted error reading the status line.
However, sometimes we get ConnectionReset instead. Also the HTTP
spec says that idempotent requests may be retried anytime a connection
is closed prematurely.
The change treats as retryable any ConnectionAborted OR ConnectionReset
error while reading the status line and headers. It removes the special
case BadStatusRead error.
Fixes#165 (I think).
In a few places we relied on "localhost" as a default if a URL's host
was not set, but I think it's better to error out in these cases.
In general, there are a few places in Unit that assumed there is a
host as part of the URL. I've made that explicit by doing a check
at the beginning of `connect()`. I've also tried to plumb through
the semantics of "host is always present" by changing the parameter
types of some of the functions that use the hostname.
I considered a more thorough way to express this with types - for
instance implementing an `HttpUrl` struct that embeds a `Url`, and
exports most of the same methods, but guarantees that host is always
present. However, that was more invasive than this so I did a smaller
change to start.
This defines a new trait `Resolver`, which turns an address into a
Vec<SocketAddr>. It also provides an implementation of Resolver for
`Fn(&str)` so it's easy to define simple resolvers with a closure.
Fixes#82
Co-authored-by: Ulrik <ulrikm@spotify.com>
Instead of cloning most of `Request`'s fields individually when
creating a `Unit`, this PR switches to just cloning `Request` and
stuffing it in `Unit`, and changes references to `unit.[field]` to
`unit.req.[field]` where appropriate.
Fixes#155
`set_read_timeout` and `set_write_timeout` can cause `ErrorKind::WouldBlock` on unix-y platforms.
This PR normalizes those cases to `ErrorKind::TimedOut`. This will make it simpler higher up in the
stack to deal with timeouts.
The reference to time::Instant under feature = socks-proxy was
incorrectly scoped, and should have been just Instant. This breaks the
doc build and any builds that use feature = socks-proxy.
time_until_deadline had a time of check to time of use problem - the
deadline could pass between a call to checked_duration_since and the
evaluation of `deadline - now` (which panics if the result would be
negative). Resolve that by flipping the order of
checked_duration_since's arguments and using the result rather than
ignoring it.
Also there were three places that called deadline - now(), which could
panic. Replace those with time_until_deadline().
Adds some feature guards, and removes an unnecessary feature guard
around a call to connect_https (there's an implementation available for
non-TLS that returns UnknownScheme).
Also, remove unnecessary agent.state() method that was only available in
TLS builds. The state field is directly accessible within the crate, and
can be used in both TLS and non-TLS builds.
Co-authored-by: Martin Algesten <martin@algesten.se>
This deprecates timeout_read() and timeout_write() in favor of
timeout(). The new timeout method on Request takes a Duration instead
of a number of milliseconds, and is measured against overall request
time, not per-read time.
Once a request is started, the timeout is turned into a deadline
specific to that call. The deadline is used in conjunction with the
new DeadlineStream class, which sets a timeout on each read according
to the remaining time for the request. Once the request is done,
the DeadlineStream is unwrapped via .into::<Stream>() to become
an undecorated Stream again for return to the pool. Timeouts on the
stream are unset at this point.
Still to be done:
Add a setting on Agent for default timeout.
Change header-writing code to apply overall deadline rather than
per-write timeout.
Fixes#28.
Fix up cfg attributes to work on an xor basis.
Previously, the cfg(any()) attributes would cause issues when
both native-tls and tls features were enabled. Now, https functions
and enum variants will only be created when tls xor native-tls are
enabled. Additionally, a compile error has been added for when
both tls and native-tls features are enabled.
This builds on 753d61b. Before we send a request, we can do a 1-byte
nonblocking peek on the connection. If the server has closed the
connection, this will give us an EOF, and we can take the connection out
of the pool before sending any request on it. This will reduce the
likelihood that we send a non-retryable POST on an already-closed
connection.
The server could still potentially close the connection between when we
make this check and when we finish sending the request, but this should
handle the majority of cases.
If DNS resolves to multiple IPs but the service is only running on one
of them and it isn't teh first IP, a connection will fail.
This was detected via running vault that would only bind to IPv4 but
localhost was returning ::1 followed by 127.0.0.1.
After this fix, the service connects without problem.
