A call to `sweep()`, although protected by the lock for concurrent calls to `reserve()`, may be concurrent with `remove(Entry)`.
`remove(Entry)` in turn calls `entries.remove(Object)`, so that the concurrent iteration in `sweep()` over `entries` fails with an `ArrayIndexOutOfBoundsException`.
Now using the bulk `entries.removeIf(Predicate)` method in `sweep()`, so that sweeping is atomic with respect to `entries.remove(Object)`.
Fixed other occurrences of manual iteration over CopyOnWriteArrayList that may be concurrent with removals.
Signed-off-by: Simone Bordet <simone.bordet@gmail.com>
Now `SslSessionData` is stored as a field, rather than in the `SSLSession` as an attribute.
This implies a little more cost to create the `SslSessionData` per connection rather than per `SSLSession`, but it should be negligible.
Now `SslSessionData` cannot be retrieved as a `SSLSession` attribute, but we have explicit APIs to retrieve it, so it should not be a problem.
Signed-off-by: Simone Bordet <simone.bordet@gmail.com>
Release request buffer before handling when there is no content
---------
Signed-off-by: Simone Bordet <simone.bordet@gmail.com>
Co-authored-by: Simone Bordet <simone.bordet@gmail.com>
* Defaulted the number of selectors to 1 in HttpClientTransportOverHTTP, to align with ClientConnector.
* Improved virtual thread documentation.
Signed-off-by: Simone Bordet <simone.bordet@gmail.com>
Fixed HttpReceiverOverHTTP2.read() to return a failed chunk if the stream has been reset.
This closes the window where a RST_STREAM frame may be received, drain the DATA frames in HTTP2Stream, and the application reading exactly at that point.
Signed-off-by: Simone Bordet <simone.bordet@gmail.com>
In this PR, the responsibility to release the buffers is in 2 methods: onFillable() (called when network data is available, and to process the next request) and parseAndFillForContent() (called from Request.read()).
Signed-off-by: Simone Bordet <simone.bordet@gmail.com>
Co-authored-by: gregw <gregw@webtide.com>
* Reworked HttpReceiverOverHTTP state machine, in particular:
** Introduced a boolean parameter to parseAndFill() and parse(), that specifies whether to notify the application demand callback.
This is necessary because reads may happen from any threads, and must not notify the application demand callback.
Only when there is no data, and fill interest is set, then the application demand callback must be notified.
** Removed action field to avoid lambda allocation.
** Now the application is called directly from the parse() method.
** Reading -1 from the network drives the parser by calling again parse(), rather than the parser directly.
This allows to have a central place to notify the response success event.
* Fixed FastCGI similarly to HTTP/1.1.
* Removed leftover of the multiplex implementation.
* Fixed test flakyness in `NetworkTrafficListenerTest`: consume the request content before sending the response.
* Follow up after #10880: only abort the request if there is request content in `AuthenticationProtocolHandler` and `RedirectProtocolHandler`.
This avoids the rare case where the response arrives before the request thread has modified the request state, even if the request has been fully sent over the network, causing the request to be failed even if it should not.
* added `SerializedInvoker` assertions about current thread invoking.
* Name all SerializedInvoker instances for better troubleshooting.
Signed-off-by: Simone Bordet <simone.bordet@gmail.com>
Signed-off-by: Ludovic Orban <lorban@bitronix.be>
Co-authored-by: Ludovic Orban <lorban@bitronix.be>
Now QoSHandler resumes requests using Request.getComponents().getExecutor().
This Executor is configured to be the virtual thread executor, if present, otherwise the Server Executor.
Removed warn() from VirtualThreads.isVirtualThread(), as it was too verbose.
Signed-off-by: Simone Bordet <simone.bordet@gmail.com>
Now HpackException.SessionException is treated specially in HTTP2Flusher.
It is caught, it fails all the entries, and then tries to send a GOAWAY, which will be the only frame allowed into the HTTP2FLusher at that point.
Signed-off-by: Simone Bordet <simone.bordet@gmail.com>
The test uncovered a larger problem detailed in the issue: the Handler Callback should be non-blocking.
Since all implementations of HttpStream are non-blocking, overridden HttpStream.getInvocationType() to return NON_BLOCKING.
This guarantees that even in case of all server threads blocked, blocked/pending writes can be completed.
Signed-off-by: Simone Bordet <simone.bordet@gmail.com>
Implemented Connection.getSslSessionData(), where the Connection can be obtained from the Request:
request.getConnection().getSslSessionData().
Updated documentation.
Signed-off-by: Simone Bordet <simone.bordet@gmail.com>
Simone Bordet
Jan Bartel
Greg Wilkins
Olivier Lamy
Ludovic Orban
Joakim Erdfelt
Lachlan
Lachlan Roberts
Dmitry Kaukov