Motivation:
There are various places in OpenSslEngine wher we can do performance optimizations.
Modifications:
- Reduce JNI calls when possible
- Detect finished handshake as soon as possible
- Eliminate double calculations
- wrap multiple ByteBuffer if possible in a loop
Result:
Better performance
Motivation:
If SO_LINGER is used shutdownOutput() and close() syscalls will block until either all data was send or until the timeout exceed. This is a problem when we try to execute them on the EventLoop as this means the EventLoop may be blocked and so can not process any other I/O.
Modifications:
- Add AbstractUnsafe.closeExecutor() which returns null by default and use this Executor for close if not null.
- Override the closeExecutor() in NioSocketChannel and EpollSocketChannel and return GlobalEventExecutor.INSTANCE if getSoLinger() > 0
- use closeExecutor() in shutdownInput(...) in NioSocketChannel and EpollSocketChannel
Result:
No more blocking of the EventLoop if SO_LINGER is used and shutdownOutput() or close() is called.
Motivation:
Currently, using a MessageAggregator in the pipeline always results in the creation of an unpooled heap CompositeByteBuf. By using the ByteBufAllocator the CompositeByteBuf will use the implementation specified by the ByteBufAllocator.
Modifications:
Use the ChannelHandlerContext's ByteBufAllocator to create the CompositeByteBuf for message aggregation
Result:
The CompositeByteBuf is now configured based on the ByteBufAllocator's settings.
Motivation:
Some of the methods are frequently called and so should be inlined if possible.
Modifications:
Give the compiler a hint that we want to inline these methods.
Result:
Better performance if inlined.
Motivation:
Older linux kernels have problems handling a large value for epoll_wait(...) and so wait for ever.
Modifications:
Adjust timeout on the fly if a too big value is passed in.
Result:
Correctly works also on older kernels.
Motivation:
The writeSpinCount was ignored in the epoll transport and it just kept on trying writing. This could cause unnessary cpu spinning if a slow remote peer was reading the data very very slow.
Modification:
- Correctly take writeSpinCount into account when writing.
Result:
Less cpu spinning when writing to a slow remote peer.
Motivation:
isRoot() is an expensive operation. We should avoid calling it if
possible.
Modifications:
Move the isRoot() checks to the end of the 'if' block, so that isRoot()
is evaluated only when really necessary.
Result:
isRoot() is evaluated only when SO_BROADCAST is set and the bind address
is anylocal address.
Motivation:
Fix regression introduced by 585ce1593f, which missed to set EPOLLRDHUP for all stream channels.
Modifications:
Correctly set EPOLLRDHUP for all stream channels in the AbstractEpollStreamChannel constructor.
Result:
No more test failures in EpollDomain*Channel tests.
Related:
- 27a25e29f7
Motivation:
The commit mentioned above introduced a regression where
channelReadComplete() event is swallowed by a handler which was added
dynamically.
Modifications:
Do not suppress channelReadComplete() if the current handler's
channelRead() method was not invoked at all, so that a just-added
handler does not suppress channelReadComplete().
Result:
Regression is gone, and channelReadComplete() is invoked when necessary.
Motivation:
Before we used a long[] to store the ready events, this had a few problems and limitations:
- An extra loop was needed to translate between epoll_event and our long
- JNI may need to do extra memory copy if the JVM not supports pinning
- More branches
Modifications:
- Introduce a EpollEventArray which allows to directly write in a struct epoll_event* and pass it to epoll_wait.
Result:
Better speed when using native transport, as shown in the benchmark.
Before:
[xxx@xxx wrk]$ ./wrk -H 'Connection: keep-alive' -d 120 -c 256 -t 16 -s scripts/pipeline-many.lua http://xxx:8080/plaintext
Running 2m test @ http://xxx:8080/plaintext
16 threads and 256 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 14.56ms 8.64ms 117.15ms 80.58%
Req/Sec 286.17k 38.71k 421.48k 68.17%
546324329 requests in 2.00m, 73.78GB read
Requests/sec: 4553438.39
Transfer/sec: 629.66MB
After:
[xxx@xxx wrk]$ ./wrk -H 'Connection: keep-alive' -d 120 -c 256 -t 16 -s scripts/pipeline-many.lua http://xxx:8080/plaintext
Running 2m test @ http://xxx:8080/plaintext
16 threads and 256 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 14.12ms 8.69ms 100.40ms 83.08%
Req/Sec 294.79k 40.23k 472.70k 66.75%
555997226 requests in 2.00m, 75.08GB read
Requests/sec: 4634343.40
Transfer/sec: 640.85MB
Motivation:
Even if a handler called ctx.fireChannelReadComplete(), the next handler
should not get its channelReadComplete() invoked if fireChannelRead()
was not invoked before.
Modifications:
- Ensure channelReadComplete() is invoked only when the handler of the
current context actually produced a message, because otherwise there's
no point of triggering channelReadComplete().
i.e. channelReadComplete() must follow channelRead().
- Fix a bug where ctx.read() was not called if the handler of the
current context did not produce any message, making the connection
stall. Read the new comment for more information.
Result:
- channelReadComplete() is invoked only when it makes sense.
- No stale connection
Motivation:
At the moment we log priming read and handshake errors via info log level and still throw a SSLException that contains the error. We should only log with debug level to generate less noise.
Modifications:
Change logging to debug level.
Result:
Less noise .
Motivation:
The Http2DefaultFrameWriter copies all contents into a buffer (or uses a CompositeBuffer in 1 case) and then writes that buffer to the socket. There is an opportunity to avoid the copy operations and write directly to the socket.
Modifications:
- Http2DefaultFrameWriter should avoid copy operations where possible.
- The Http2FrameWriter interface should be clarified to indicate that ByteBuf objects will be released.
Result:
Hopefully less allocation/copy leads to memory and throughput performance benefit.
Motivation:
Because of a re-entrance bug in PendingWriteQueue it was possible to get the queue corrupted and also trigger an IllegalStateException caused by multiple recycling of the internal PendingWrite objects.
Modifications:
- Correctly guard against re-entrance
Result:
No more IllegalStateException possible
Motivation:
To use WebSocketClientHandshaker / WebSocketServerHandshaker it's currently a requirement of having a HttpObjectAggregator in the ChannelPipeline. This is not a big deal when a user only wants to server WebSockets but is a limitation if the server serves WebSockets and normal HTTP traffic.
Modifications:
Allow to use WebSocketClientHandshaker and WebSocketServerHandshaker without HttpObjectAggregator in the ChannelPipeline.
Result:
More flexibility
Motivation:
Netty uses edge-triggered epoll by default for performance reasons. The downside here is that a messagesPerRead limit can not be enforced correctly, as we need to consume everything from the channel when notified.
Modification:
- Allow to switch epoll modes before channel is registered
- Some refactoring to share more code
Result:
It's now possible to switch epoll mode.
Motivation:
Http2Stream has several methods that provide state information. We need
to simplify how state is used and consolidate as many of these fields as
possible.
Modifications:
Since we already have a concept of a stream being active or inactive,
I'm now separating the deactivation of a stream from the act of closing
it. The reason for this is the case of sending a frame with
endOfStream=true. In this case we want to close the stream immediately
in order to disallow further writing, but we don't want to mark the
stream as inactive until the write has completed since the inactive
event triggers the flow controller to cancel any pending writes on the
stream.
With deactivation separated out, we are able to eliminate most of the
additional state methods with the exception of `isResetSent`. This is
still required because we need to ignore inbound frames in this case (as
per the spec), since the remote endpoint may not yet know that the
stream has been closed.
Result:
Fixes#3382
Motiviation:
When using domain sockets on linux it is supported to recv and send file descriptors. This can be used to pass around for example sockets.
Modifications:
- Add support for recv and send file descriptors when using EpollDomainSocketChannel.
- Allow to obtain the file descriptor for an Epoll*Channel so it can be send via domain sockets.
Result:
recv and send of file descriptors is supported now.
Motivation:
Using Unix Domain Sockets can be very useful when communication should take place on the same host and has less overhead then using loopback. We should support this with the native epoll transport.
Modifications:
- Add support for Unix Domain Sockets.
- Adjust testsuite to be able to reuse tests.
Result:
Unix Domain Sockets are now support when using native epoll transport.
Motivation:
SonarQube (clinker.netty.io/sonar) reported a resource which may not have been properly closed in all situations in AbstractDiskHttpData.
Modifications:
- Ensure file channels are closed in the presence of exceptions.
- Correct instances where local channels were created but potentially not closed.
Result:
Less leaks. Less SonarQube vulnerabilities.
Motivation:
SonarQube (clinker.netty.io/sonar) reported a few 'critical' issues related to the OpenSslEngine.
Modifications:
- Remove potential for dereference of null variable.
- Remove duplicate null check and TODO cleanup.
Results:
Less potential for null dereference, cleaner code, and 1 less TODO.
Motivation:
Previously flow-controller had to know the implementation details of each frame type in order to write it correctly. That concern is more correctly handled by the encoder. By encapsulating the payload types to be flow-controlled it will be easier to add support for extension types later. This change also fixes#3353.
Modifications:
Add interface FlowControlled which is now delivered to flow-controller.
Implement this interface for HEADERS and DATA
Refactor and improve tests for flow-control.
Result:
Flow control semantics are more cleanly separated for data encoding and implementation is simpler overall.
Motivation:
`HttpResponseDecoder` and `HttpRequestDecoder` in the event when the max configured sizes for HTTP initial line, headers or content is breached, sends a `DefaultHttpResponse` and `DefaultHttpRequest` respectively. After this `HttpObjectDecoder` gets into `BAD_MESSAGE` state and ignores any other data received on this connection.
The combination of the above two behaviors, means that the decoded response/request are not complete (absence of sending `LastHTTPContent`). So, any code, waiting for a complete message will have to additionally check for decoder result to follow the correct semantics of HTTP.
If `HttpResponseDecoder` and `HttpRequestDecoder` creates a Full* invalid message then the request/response is a complete HTTP message and hence obeys the HTTP contract.
Modification:
Modified `HttpRequestDecoder`, `HttpResponseDecoder`, `RtspRequestDecoder` and `RtspResponseDecoder` to return Full* messages from `createInvalidMessage()`
Result:
Fixes the wrong behavior of sending incomplete messages from these codecs
In testEncodingSingleCookieV0():
Let's assume we encoded a cookie with MaxAge=50 when currentTimeMillis
is 10999.
Because the encoder will not encode the millisecond part for Expires,
the timeMillis value of the encoded Expires field will be 60000. (If we
did not dropped the millisecond part, it would be 60999.)
Encoding a cookie will take some time, so currentTimeMillis will
increase slightly, such as to 11001.
diff = (60000 - 11001) / 1000 = 48999 / 1000 = 48
maxAge - diff = 50 - 48 = 2
Due to losing millisecond part twice, we end up with the precision
problem illustrated above, and thus we should increase the tolerance
from 1 second to 2 seconds.
/cc @slandelle
Motivation:
SslHandler adds a pending write with an empty buffer and a VoidChannelPromise when a user flush and not pending writes are currently stored. This may produce an IllegalStateException later if the user try to add a ChannelFutureListener to the promise in the next ChannelOutboundHandler.
Modifications:
Replace ctx.voidPromise() with ctx.newPromise()
Result:
No more IllegalStateException possible
Motivation:
At the moment the max number of events that can be handled per epoll wakup was set during construction.
Modifications:
- Automatically increase the max number of events to handle
Result:
Better performance when a lot of events need to be handled without adjusting the code.
Motivation:
The current way how the guard against overflow when generating the nextId() is pretty slow once an overflow happened.
Modifications:
Once a possible overflow is detected all ids used by the EpollEventLoop are scrubed and re-assigned to the registered Channels. This way we only need to do extra work each time an overflow is detected.
Result:
More consistent performance even after the first overflow was detected.
Motivation:
SSLEngine specifies that IllegalArgumentException must be thrown if a null argument is given when using wrap(...) or unwrap(...).
Modifications:
Replace NullPointerException with IllegalArgumentException to match the javadocs.
Result:
Match the javadocs.
Motivation:
We failed to correctly calculate the endOffset when wrap multiple ByteBuffer and so not wrapped everything when an offset > 0 is used.
Modifications:
Correctly calculate endOffset.
Result:
All ByteBuffers are correctly wrapped when offset > 0.
Motivation:
As the ByteBuf is not set to null after release it we may try to release it again in handleReadException()
Modifications:
- set ByteBuf to null to avoid another byteBuf.release() to be called in handleReadException()
Result:
No IllegalReferenceCountException anymore
Motivation:
On Linux, you can gather various metrics using getsockopt(..., TCP_INFO,
...).
Modifications:
Add EpollSocketChannel.tcpInfo() which returns EpollTcpInfo that exposes
all metrics exposed via getsockopt(..., TCP_INFO, ...)
Result:
TCP_INFO support implemented
Motivation:
The JdkZlibDecoder and JZlibDecoder call isReadable and readableBytes in the same method. There is an opportunity to reduce the number of methods calls to just use readableBytes. JdkZlibDecoder reads from a ByteBuf with an absolute index instead of using readerIndex()
Modifications:
- Use readableBytes where isReadable was used
- Correct absolute ByteBuf index to be relative to readerIndex()
Result:
Less method calls duplicating work and preventing an index out of bounds exception.
Motivation:
In the native transport we use getpeername to obtain the remote address from the file descriptor. This may fail for various reasons in which case NULL is returned.
Modifications:
- Check for null when try to obtain remote / local address
Result:
No more NPE
Motivation:
Internet Explorer doesn't honor Set-Cookie header Max-Age attribute. It only honors the Expires one.
Modification:
Always generate an Expires attribute along the Max-Age one.
Result:
Internet Explorer compatible expiring cookies. Close#1466.
Motivation:
A downstream consumer of Netty failed as emitting zero-length http2 data frames in a unit test resulted in assertion errors in Http2LocalFlowController. Since zero-length frames are valid, an assertion that http2 data frame length must be positive is invalid.
Modifications:
Assertions of data length in Http2LocalFlowController now permit zero.
Result:
Those running netty with assertions on can now emit zero length http2 data frames.
Motivation:
HTTP/2 codec was implemented in master branch.
Since, master is not yet stable and will be some time before it gets released, backporting it to 4.1, enables people to use the codec with a stable netty version.
Modification:
The code has been copied from master branch as is, with minor modifications to suit the `ChannelHandler` API in 4.x.
Apart from that change, there are two backward incompatible API changes included, namely,
- Added an abstract method:
`public abstract Map.Entry<CharSequence, CharSequence> forEachEntry(EntryVisitor<CharSequence> visitor)
throws Exception;`
to `HttpHeaders` and implemented the same in `DefaultHttpHeaders` as a delegate to the internal `TextHeader` instance.
- Added a method:
`FullHttpMessage copy(ByteBuf newContent);`
in `FullHttpMessage` with the implementations copied from relevant places in the master branch.
- Added missing abstract method related to setting/adding short values to `HttpHeaders`
Result:
HTTP/2 codec can be used with netty 4.1
Motivation:
HttpContentDecoder had the following issues:
- For chunked content, the decoder set invalid "Content-Length" header
with length of the first decoded chunk.
- Decoding of FullHttpRequests put both the original conent and decoded
content into output. As result, using HttpObjectAggregator before the
decoder lead to errors.
- Requests with "Expect: 100-continue" header were not acknowleged:
the decoder didn't pass the header message down the handler's chain
until content is received. If client expected "100 Continue" response,
deadlock happened.
Modification:
- Invalid "Content-Length" header is removed; handlers down the chain can either
rely on LastHttpContent message or ask HttpObjectAggregator to add the header.
- FullHttpRequest is split into HttpRequest and HttpContent (decoded) parts.
- Header (HttpRequest) part of request is sent down the chain as soon as it's received.
Result:
The issues are fixed, unittest is added.
Motivation:
Pull request for RFC6265 support had some unused flag first in ClientCookieDecoder.
Modification:
Remove unused flag first.
Result:
Cleaner code.
Motivation:
There are two member variables (addAllVisitor, setAllVisitor) which are likely not to be used in the majority of use cases.
Modifications:
Remove these member variables and rely on a method to return a new object when needed.
Result:
Two less member variables for each DefaultHeaders instance.
Motivation:
The Headers interface had two member variables (addAllVisitor, setAllVisitor) which are not necessarily always needed but are always instantiated. This may result in excess memory being used.
Modifications:
- addAllVisitor will be accessed via a method addAllVisitor() which will use lazy initialization.
- setAllVisitor will be accessed via a method addAllVisitor() which will use lazy initialization.
Result:
Potential memory savings by using lazy initialization.
Motivation:
Rfc6265Client/ServerCookieEncoder is a better replacement of the old
Client/ServerCookieEncoder, and thus there's no point of keeping both.
Modifications:
- Remove the old Client/ServerCookieEncoder
- Remove the 'Rfc6265' prefix from the new cookie encoder/decoder
classes
- Deprecate CookieDecoder
Result:
We have much better cookie encoder/decoder implementation now.
Motivation:
Currently Netty supports a weird implementation of RFC 2965.
First, this RFC has been deprecated by RFC 6265 and nobody on the
internet use this format.
Then, there's a confusion between client side and server side encoding
and decoding.
Typically, clients should only send name=value pairs.
This PR introduces RFC 6265 support, but keeps on supporting RFC 2965 in
the sense that old unused fields are simply ignored, and Cookie fields
won't be populated. Deprecated fields are comment, commentUrl, version,
discard and ports.
It also provides a mechanism for safe server-client-server roundtrip, as
User-Agents are not supposed to interpret cookie values but return them
as-is (e.g. if Set-Cookie contained a quoted value, it should be sent
back in the Cookie header in quoted form too).
Also, there are performance gains to be obtained by not allocating the
attribute name Strings, as we only want to match them to find which POJO
field to populate.
Modifications:
- New RFC6265ClientCookieEncoder/Decoder and
RFC6265ServerCookieEncoder/Decoder pairs that live alongside old
CookieEncoder/Decoder pair to not break backward compatibility.
- New Cookie.rawValue field, used for lossless server-client-server
roundtrip.
Result:
RFC 6265 support.
Clean separation of client and server side.
Decoder performance gain:
Benchmark Mode Samples Score Error
Units
parseOldClientDecoder thrpt 20 2070169,228 ± 105044,970
ops/s
parseRFC6265ClientDecoder thrpt 20 2954015,476 ± 126670,633
ops/s
This commit closes#3221 and #1406.
