Motivation:
NIO now supports a pluggable select strategy, but EPOLL currently doesn't support this. We should strive for feature parity for EPOLL.
Modifications:
- Add SelectStrategy to EPOLL transport.
Result:
EPOLL transport supports SelectStategy.
Motivation:
Under high throughput/low latency workloads, selector wakeups are
degrading performance when the incoming operations are triggered
from outside of the event loop. This is a common scenario for
"client" applications where the originating input is coming from
application threads rather from the socket attached inside the
event loops.
As a result, it can be desirable to defer the blocking select
so that incoming tasks (write/flush) do not need to wakeup
the selector.
Modifications:
This changeset adds the notion of a generic SelectStrategy which,
based on its contract, allows the implementation to optionally
defer the blocking select based on some custom criteria.
The default implementation resembles the original behaviour, that
is if tasks are in the queue `selectNow()` and move on, and if no
tasks need to be processed go into the blocking select and wait
for wakeup.
The strategy can be customized per `NioEventLoopGroup` in the
constructor.
Result:
High performance client applications are now given the chance to
customize for how long the actual selector blocking should be
deferred by employing a custom select strategy.
Motivation:
While backport the change from 4.1 to 4.0 I somehow missed to port one line and so broke the example.
Modifications:
Add needed handler which was missing due bad merge commit.
Result:
Example works again in 4.0
Motivation:
We need to ensure we run all pending tasks before doing any flush in writeOutbound(...) to ensure all pending tasks are run first. Also we should remove the assert of the future and just add a listener to it so it is processed later if needed. This is true as a user may schedule a write for later execution.
Modifications:
- Remove assert of future in writeOutbound(...)
- Correctly run pending tasks before doing the flush and also before doing the close of the channel.
- Add unit tests to proof the defect is fixed.
Result:
Correclty handle the situation of delayed writes.
Motivation:
We need to break out of the read loop for two reasons:
- If the input was shutdown in between (which may be the case when the user did it in the
fireChannelRead(...) method we should not try to read again to not produce any
miss-leading exceptions.
- If the user closes the channel we need to ensure we not try to read from it again as
the filedescriptor may be re-used already by the OS if the system is handling a lot of
concurrent connections and so needs a lot of filedescriptors. If not do this we risk
reading data from a filedescriptor that belongs to another socket then the socket that
was "wrapped" by this Channel implementation.
Modification:
Break the reading loop if the input was shutdown from within the channelRead(...) method.
Result:
No more meaningless exceptions and no risk to read data from wrong socket after the original was closed.
Conflicts:
transport-native-epoll/src/main/java/io/netty/channel/epoll/AbstractEpollStreamChannel.java
Motivation:
There are some use cases when a client may only be willing to read from a channel once
its previous write is finished (eg: serial dispatchers in Finagle). In this case, a
connection with SslHandler installed and ctx.channel().config().isAutoRead() == false
will stall in 100% of cases no matter what order of "channel active", "write", "flush"
events was.
The use case is following (how Finagle serial dispatchers work):
1. Client writeAndFlushes and waits on a write-promise to perform read() once it's satisfied.
2. A write-promise will only be satisfied once SslHandler finishes with handshaking and
sends the unencrypted queued message.
3. The handshaking process itself requires a number of read()s done by a client but the
SslHandler doesn't request them explicitly assuming that either auto-read is enabled
or client requested at least one read() already.
4. At this point a client will stall with NEED_UNWRAP status returned from underlying engine.
Modifiations:
Always request a read() on NEED_UNWRAP returned from engine if
a) it's handshaking and
b) auto read is disabled and
c) it wasn't requested already.
Result:
SslHandler is now completely tolerant of whether or not auto-read is enabled and client
is explicitly reading a channel.
Motivation:
We should throw a more helpful exception when a non PKCS#8 key is used by the user.
Modifications:
Change exception message to give a hint what is wrong.
Result:
Easier for user to understand whats wrong with their used key.
Motivation:
8dbf5d02e5 modified the shutdown code for Socket but did not correctly calculate the change in shutdown state and only applying this change. This is significant because if sockets are being opening and closed quickly and the underlying FD happens to be reused we need to take care that we don't unintentionally change the state of the new FD by acting on an object which represents the old incarnation of that FD.
Modifications:
- Calculate the shutdown change, and only apply what has changed, or exit if no change.
Result:
Socket.shutdown can not inadvertently affect the state of another logical FD.
Motivation:
Often the user uses EmbeddedChannel within unit tests where the only "important" thing is to know if any pending messages were in the buffer and then release these.
We should provide methods for this so the user not need to manually loop through these and release.
Modifications:
Add methods to easily handle releasing of messages.
Result:
Less boiler-plate code for the user to write.
Motivation:
See https://github.com/netty/netty/issues/3411.
Backport perf improvements on 4.0 and make AsyncHttpClient DNS modules
backports easier to maintain.
Modifications:
Cherry-picked b7415a3307
Result:
Reuse a thread local ArrayList to avoid allocations.
Motivation:
We should upgrade to latest netty-tcnative version.
Modifications:
Upgrade to version 1.1.33.Fork15
Result:
Latest netty-tcnative version is used.
Motivation:
The current slow path of FastThreadLocal is much slower than JDK ThreadLocal. See #4418
Modifications:
- Add FastThreadLocalSlowPathBenchmark for the flow path of FastThreadLocal
- Add final to speed up the slow path of FastThreadLocal
Result:
The slow path of FastThreadLocal is improved.
Motivation:
The code of transport-native-epoll missed some things in terms of static keywords, @deprecated annotations and other minor things.
Modifications:
- Add missing @deprecated annotation
- Not using FQCN in javadocs
- Add static keyword where possible
- Use final fields when possible
- Remove throws IOException from method where it is not needed.
Result:
Cleaner code.
Conflicts:
transport-native-epoll/src/main/java/io/netty/channel/epoll/AbstractEpollChannel.java
transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollDatagramChannel.java
transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollRecvByteAllocatorHandle.java
Motivation:
SSLContext.buildTrustManagerFactory(...) builds a KeyStore to
initialize the TrustManagerFactory from an array of X509Certificates,
assuming that array is a chain and that each certificate will have a
unique Subject Distinguised Name.
However, the collection of certificates used as trust anchors is generally
not a chain (it is an unordered collection), and it is legitimate for it
to contain multiple certificates with the same Subject DN.
The existing code uses the Subject DN as the alias name when filling in
the `KeyStore`, thereby overwriting other certificates with the same
Subject DN in this collection, so some certificates may be discarded.
In addition, the code related to building trust managers can take an array of
X509Certificate instances to use as trust anchors. The variable name is
usually trustCertChain, and the documentation refers to them as a "chain".
However, while it makes sense to talk about a "chain" from a keymanager
point of view, these certificates are just an unordered collection in a
trust manager. (There is no chaining requirement, having the Subject DN
matching its predecessor's Issuer DN.)
This can create confusion to for users not used with PKI concepts.
Modifications:
SSLContext.buildTrustManagerFactory(...) now uses a distinct alias for each
array (simply using a counter, since this name is never used for reference
later). This patch also includes a unit test with CA certificates using the
same Subject DN.
Also renamed trustCertChain into trustCertCollection, and changed the
references to "chain" in the Javadoc.
Result:
Each loaded certificate now has a unique identifier when loaded, so it is
now possible to use multiple certificates with the same Subject DN as
trust anchors.
Hopefully, renaming the parameter should also reduce confusion around PKI
concepts.
Motivation:
DefaultCookie constructor performs a name validation that doesn’t match
RFC6265. Moreover, such validation is already performed in strict
encoders and decoders.
Modifications:
Drop DefaultCookie name validation, rely on encoders and decoders.
Result:
no more duplicate broken validation
Motivation:
The method setBytes creates temporary heap buffer when source buffer is read-only.
But this temporary buffer is not used correctly and may lead to data corruption.
This problem occurs when target buffer is pooled and temporary buffer
arrayOffset() is not zero.
Modifications:
Use correct arrayOffset when calling PlatformDependent.copyMemory.
Unit test was added to test this case.
Result:
Setting buffer content works correctly when target is pooled buffer and source
is read-only ByteBuffer.
Motivation:
We need to ensure we call ctx.flush() before closing the actual channel when an handshake failure took place. If we miss to do so we may not send all pending data to the remote peer which also include SSL alerts.
Modifications:
Ensure we call ctx.flush() before ctx.close() on a handshake error.
Result:
All pending data (including SSL alerts) are written to the remote peer on a handshake error.
Motivation:
We need to ensure we do all checks inside of the try / catch block so we free native memory that was allocated in the constructor of the super class in a timely manner.
Modifications:
Move all checks inside of the try block.
Result:
Correctly release native memory (and not depend on the finalizer) when a check in the constructors fails
Motivation:
In commit acbca192bd we changed to have our native operations which either gall getsockopt or setsockopt throw IOExceptions (to be more specific we throw a ClosedChannelException in some cases). Unfortunally I missed to also do the same for getSoError() and missed to add throws IOException to the native methods.
Modifications:
- Correctly throw IOException from getSoError()
- Add throws IOException to native methods where it was missed.
Result:
Correct declaration of getSoError() and other native methods.
Conflicts:
transport-native-epoll/src/main/java/io/netty/channel/epoll/AbstractEpollChannel.java
Motivation:
A user may use a private key which is encrypted with an empty password. Because of this we should only handle a null password in a special way.
Modifications:
- Correctly handle private key that is encrypted with empty password.
- Make OpenSsl*Context implementions consistent in terms of initialization in the constructor.
Result:
Correctly support private key that is encrypted with empty password.
Motivation:
We also need to add synchronization when access fields to ensure we see the latest updates.
Modifications:
Add synchronization when read fields that are written concurrently.
Result:
Ensure correct visibility of updated.
Motivation:
If SO_LINGER is set to 0 the EPOLL transport will send a FIN followed by a RST. This is not consistent with the behavior of the NIO transport. This variation in behavior can cause protocol violations in streaming protocols (e.g. HTTP) where a FIN may be interpreted as a valid end to a data stream, but RST may be treated as the data is corrupted and should be discarded.
https://github.com/netty/netty/issues/4170 Claims the behavior of NIO always issues a shutdown when close occurs. I could not find any evidence of this in Netty's NIO transport nor in the JDK's SocketChannel.close() implementation.
Modifications:
- AbstractEpollChannel should be consistent with the NIO transport and not force a shutdown on every close
- FileDescriptor to keep state in a consistent manner with the JDK and not allow a shutdown after a close
- Unit tests for NIO and EPOLL to ensure consistent behavior
Result:
EPOLL is capable of sending just a RST to terminate a connection.
Motivation:
We had some double spacing in the methods which should be removed to keep things consistent.
Modifications:
Remove redundant spaces.
Result:
Cleaner / consistent coding style.
Motivation:
My previous commit b88a980482 introduced a flawed unit test,
that executes an assertion in a different thread than the test thread.
If this assertion fails, the test doesn't fail.
Modifications:
Replace the assertion by a proper workaround.
Result:
More correct unit test
Motivation:
To be consistent with the JDK we should ensure our native methods throw a ClosedChannelException if the Channel was previously closed. This will then be wrapped in a ChannelException as usual. For all other errors we continue to just throw a ChannelException directly.
Modifications:
Ensure getsockopt and setsockopt will throw a ClosedChannelException if the channel was closed before, on other errors we throw a ChannelException as before diretly.
Result:
Consistent with the NIO Channel implementations.
Motivation:
Circular assignment of arenas to thread caches can lead to less than optimal
mappings in cases where threads are (frequently) shutdown and started.
Example Scenario:
There are a total of 2 arenas. The first two threads performing an allocation
would lead to the following mapping:
Thread 0 -> Arena 0
Thread 1 -> Arena 1
Now, assume Thread 1 is shut down and another Thread 2 is started. The current
circular assignment algorithm would lead to the following mapping:
Thread 0 -> Arena 0
Thread 2 -> Arena 0
Ideally, we want Thread 2 to use Arena 1 though.
Presumably, this is not much of an issue for most Netty applications that do all
the allocations inside the eventloop, as eventloop threads are seldomly shut down
and restarted. However, applications that only use the netty-buffer package
or implement their own threading model outside the eventloop might suffer from
increased contention. For example, gRPC Java when using the blocking stub
performs some allocations outside the eventloop and within its own thread pool
that is dynamically sized depending on system load.
Modifications:
Implement a linear scan algorithm that assigns a new thread cache to the arena
that currently backs the fewest thread caches.
Result:
Closer to ideal mappings between thread caches and arenas. In order to always
get an ideal mapping, we would have to re-balance the mapping whenever a thread
dies. However, that's difficult because of deallocation.
Motivation:
We want to allow the use of an uber jar that contains shared dynamic libraries for all platforms (including fedora).
Modifications:
Modified OpenSsl to try and load the fedora library if the OS is Linux and the platform specified library fails before using the default lib.
Result:
True uber support.
Motivation:
PromiseAggregator's API allows for the aggregate promise to complete before the user is done adding promises. In order to support this use case the API structure would need to change in a breaking manner.
Modifications:
- Deprecate PromiseAggregator and subclasses
- Introduce PromiseCombiner which corrects these issues
Result:
PromiseCombiner corrects the deficiencies in PromiseAggregator.
Motivation:
We should always first notify the promise before trigger an event through the pipeline to be consistent.
Modifications:
Ensure we notify the promise before fire event.
Result:
Consistent behavior
Motivation:
EpollServerSocketConfig.isFreebind() throws an exception when called.
Modifications:
Use the correct getsockopt arguments.
Result:
No more exception when call EpollServerSocketConfig.isFreebind()
Motivation:
TCP_MD5 is only supported by SocketChannels so remove it from EpollServerChannelConfig which is generic.
Modifications:
Remove invalid code.
Result:
Remove invalid / dead code.
Motivation:
A custom SecurityManager may prevent calling File.exists() and so throw a SecurityException in the static init block of NetUtil.
Modifications:
Correctly catch the exception and so allow to static init NetUtil.
Result:
Allow static init method of NetUtil to work even with custom SecurityManager.
Motivation:
The statistic counters PoolArena.(allocationsTiny|allocationsSmall) are
not protected by a per arena lock, but by a per size class lock. Thus,
two concurrent allocations of different size (class) could lead to a
race and ultimately to wrong statistics.
Modifications:
Use a thread-safe LongCounter instead of a plain long data type.
Result:
Fewer data races.
Motivation:
PendingWriteQueue should guard against re-entrant writes once
removeAndFailAll() is run.
Modifications:
removeAndFailAll() should repeat until the queue is finally empty.
Result:
assertEmpty() will always hold.
Motivation:
The current HttpPostMultipartRequestDecoder can decode multipart/form-data parts with a Content-Type that specifies a charset. When this charset is invalid the Charset.forName() throws an unchecked UnsupportedCharsetException. This exception is not catched by the decoder. It should actually be rethrown as an ErrorDataDecoderException, because the developer using the API would expect this validation failure to be reported as such.
Modifications:
Add a catch block for UnsupportedCharsetException and rethrow it as an ErrorDataDecoderException.
Result:
UnsupportedCharsetException are now rethrown as ErrorDataDecoderException.
Motivation:
When using the native transport have support for TCP_DEFER_ACCEPT or / and TCP_QUICKACK can be useful.
Modifications:
- Add support for TCP_DEFER_ACCEPT and TCP_QUICKACK
- Ad unit tests
Result:
TCP_DEFER_ACCEPT and TCP_QUICKACK are supported now.
Conflicts:
transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollChannelOption.java
Motivation:
We should guard users from using Unsafe methods from outside the EventLoop if not designed to do so.
Modifications:
Add asserts
Result:
Easier for users to detect miss-use.
Motivation:
DefaultChannelHandlerContext calls a utility method validatePromise which may throw if the arguments are not valid. If this method throws then the message will not be released.
Modifications:
- If an exception is thrown the message should be released
Result:
No more leak in DefaultChannelHandlerContext.write
Motivation:
See #3321
Modifications:
1. Add CharsetUtil.encoder/decoder() methods
2. Deprecate CharsetUtil.getEncoder/getDecoder() methods
Result:
Users can use new CharsetUtil.encoder/decoder() to specify error actions
Motivation:
For on tests we expected a ConnectTimeoutException but used the default timeout of 10 seconds. This slows down testing.
Modifications:
Use connect timeout of 1 second in unit test.
Result:
Faster execution of unit test.
