Motivation:
Now the ```resolveAll``` method of RoundRobinInetAddressResolver returns results without any rotation and shuffling. As a result, it doesn't force any round-robin for clients that get a result of ```resolveAll``` and use addresses from the result one by one for a connection establishing until success. This commit implements round-robin in RoundRobinInetAddressResolver#resolveAll. These improvements inspired by the discussion here: https://github.com/AsyncHttpClient/async-http-client/issues/1285
Modifications:
Rotate collection from internal ```resolveAll``` call by index, which is incremented every call to RoundRobinInetAddressResolver#resolveAll method.
Random replaced by an incrementing counter, which makes code cheaper and guarantees predictable address order in tests.
Result:
Improved ```RoundRobinInetAddressResolver``` is compatible with clients that use ```resolveAll``` result.
Motivation:
I had a need to know the user credentials of a connected unix domain socket.
Modifications:
Added a class to encapsulate user credentials (UID, GID, and the PID).
Augemented the Socket class to provide the JNI native interface to return this new class
Augemented the c code to call getSockOpts passing <a href=http://man7.org/linux/man-pages/man7/socket.7.html>SO_PEERCRED</a>
Then surfaced the ability to get user credentials in the EpollDomainSocketChannel
Result:
The EpollDomainSocketChannel now has a the following function signature:
public PeerCredentials peerCredentials() throws IOException allowing a caller to get the UID, GID, and PID of the linux process
connected to the unix domain socket.
Motivation:
To guard against the case that a user will enqueue a lot of empty or small buffers and so raise an OOME we need to also take the overhead of the ChannelOutboundBuffer / PendingWriteQueue into account when detect if a Channel is writable or not. This is related to #5856.
Modifications:
When calculate the memory for an message that is enqueued also add some extra bytes depending on the implementation.
Result:
Better guard against OOME.
Motivation:
ResourceLeakDetector shows two main problems, racy access and heavy lock contention.
Modifications:
This PR fixes this by doing two things:
1. Replace the sampling counter with a ThreadLocalRandom. This has two benefits.
First, it makes the sampling ration no longer have to be a power of two. Second,
it de-noises the continuous races that fight over this single value. Instead,
this change uses slightly more CPU to decide if it should sample by using TLR.
2. DefaultResourceLeaks need to be kept alive in order to catch leaks. The means
by which this happens is by a singular, doubly-linked list. This creates a
large amount of contention when allocating quickly. This is noticeable when
running on a multi core machine.
Instead, this uses a concurrent hash map to keep track of active resources
which has much better contention characteristics.
Results:
Better concurrent hygiene. Running the gRPC QPS benchmark showed RLD taking about
3 CPU seconds for every 1 wall second when runnign with 12 threads.
There are some minor perks to this as well. DefaultResourceLeak accounting is
moved to a central place which probably has better caching behavior.
Motivation:
Make small refactoring for recently merged PR #5867 to make the code more flexible and expose aggressive round robin as a NameResolver too with proper code reuse.
Modifications:
Round robin is a method of hostname resolving - so Round robin related code fully moved to RoundRobinInetAddressResolver implements NameResolver<InetAddress>, RoundRobinInetSocketAddressResolver is deleted as a separate class, instance with the same functionality could be created by calling #asAddressResolver.
Result:
New forced Round Robin code exposed not only as an AddressResolver but as a NameResolver too, more proper code and semantic reusing of InetNameResolver and InetSocketAddressResolver classes.
Motivation:
We need to ensure we not add the Transfer-Encoding header if the HttpMessage is EOF terminated.
Modifications:
Only add the Transfer-Encoding header if an Content-Length header is present.
Result:
Correctly handle HttpMessage that is EOF terminated.
Motivation:
The previously generated manifest causes a parse exception when loaded into an Apache Felix OSGI container.
Modifications:
Fix parameter delimiter and unbalanced quotes in manifest entry. Suffixed with asterisk so the bundle is resolved on other architectures as well even if native libs won't be loaded.
Result:
Bundle will load properly in OSGI containers.
Motivation:
A new version of centos was released we should verify against it when release.
Modifications:
Bump up version.
Result:
Release on latest centos version.
Motivation:
We want to reject the upgrade as quickly as possible, so that we can
support streamed responses.
Modifications:
Reject the upgrade as soon as we inspect the headers if they're wrong,
instead of waiting for the entire response body.
Result:
If a remote server doesn't know how to use the http upgrade and tries to
responsd with a streaming response that never ends, the client doesn't
buffer forever, but can instead pass it along. Fixes#5954
Motivation
It's possible to extend LocalChannel as well as LocalServerChannel but the LocalServerChannel's serve(peer) method is hardcoded to create only instances of LocalChannel.
Modifications
Add a protected factory method that returns by default new LocalChannel(...) but users may override it to customize it.
Result
It's possible to customize the LocalChannel instance on either end of the virtual connection.
Motivation:
When auto-read is disabled and no reads are issued by a user, ProxyHandler will stall the connection on the proxy handshake phase waiting for the first response from a server (that was never read).
Modifications:
Read if needed when very first handshake message is send by ProxyHandler.
Result:
Proxy handshake now succeeds no matter if auto-read disabled or enabled. Without the fix, the new test is failing on master.
Motivation:
PlatformDependent has a hash code algorithm which utilizes UNSAFE for performance reasons. This hash code algorithm must also be consistent with CharSequence objects that represent a collection of ASCII characters. In order to make the UNSAFE versions and CharSequence versions the endianness should be taken into account. However the big endian code was not correct in a few places.
Modifications:
- Correct bugs in PlatformDependent class related to big endian ASCII hash code computation
Result:
Fixes https://github.com/netty/netty/issues/5925
Motivation:
8cf90f0512 switch a duplicate opreration to a slice operation. Typically this would be fine but DNS supports a compression (https://www.ietf.org/rfc/rfc1035 4.1.4. Message compression) where the payload contains absolute indexes which refer back to previously referenced content. Using a slice will break the ability for the indexes in the payload to correctly self reference to the index of the originial payload, and thus decoding may fail.
Modifications:
- Use duplicate instead of slice so DNS message compression and index references are preserved.
Result:
Fixes DefaultDnsRecordDecoder regression
Motivation:
The SETTINGS_MAX_HEADER_LIST_SIZE limit, as enforced by the HPACK Encoder, should be a stream error and not apply to the whole connection.
Modifications:
Made the necessary changes for the exception to be of type StreamException.
Result:
A HEADERS frame exceeding the limit, only affects a specific stream.
Motivation:
Commit 908464f161 also introduced a change to guard against re-entrance but failed to correctly handle the debugData and promise.
Modifications:
Release debugData and correctly notify the promise.
Result:
No more buffer leak and promise is always notified.
Motivation:
In some ByteBuf implementations we not correctly implement getBytes(index, ByteBuffer).
Modifications:
Correct code to do what is defined in the javadocs and adding test.
Result:
Implementation works as described.
Motivation:
Since Java 7, X509TrustManager implementation is wrapped by a JDK class
called AbstractTrustManagerWrapper, which performs an additional
certificate validation for Socket or SSLEngine-backed connections.
This makes the TrustManager implementations provided by
InsecureTrustManagerFactory and FingerprintTrustManagerFactory not
insecure enough, where their certificate validation fails even when it
should pass.
Modifications:
- Add X509TrustManagerWrapper which adapts an X509TrustManager into an
X509ExtendedTrustManager
- Make SimpleTrustManagerFactory wrap an X509TrustManager with
X509TrustManagerWrapper is the provided TrustManager does not extend
X509ExtendedTrustManager
Result:
- InsecureTrustManagerFactory and FingerprintTrustManagerFactory are now
insecure as expected.
- Fixes#5910
Motivation:
8ba5b5f740 removed some ciphers from the default list, and SocketSslEchoTest had one of these ciphers hard coded in the test. The test will fail if the cihper is not supported by default.
Modifications:
SocketSslEchoTest should ensure a cipher is used which will be supported by the peer
Result:
Test result no longer depends upon default cipher list.
Motivation:
Our default cipher list has not been updated in a while. We current support some older ciphers not commonly in use and we don't support some newer ciphers which are more commonly used.
Modifications:
- Update the default list of ciphers for JDK and OpenSSL.
Result:
Default cipher list is more likely to connect to peers.
Fixes https://github.com/netty/netty/issues/5859
Motivation:
Some unit tests in SingleThreadEventLoopTest rely upon Thread.sleep for sequencing events between threads. This can be unreliable and result in spurious test failures if thread scheduling does not occur in a fair predictable manner.
Modifications:
- Reduce the reliance on Thread.sleep in SingleThreadEventLoopTest
Result:
Fixes https://github.com/netty/netty/issues/5851
Motivation:
The local transport is used to communicate in the same JVM so we should use heap buffers.
Modifications:
Use heapbuffers by default if not requested otherwise.
Result:
No allocating of direct buffers by default when using local transport
Motivation:
Suppose the domain `foo.example.com` resolves to the following ip
addresses `10.0.0.1`, `10.0.0.2`, `10.0.0.3`. Round robin DNS works by
having each client probabilistically getting a different ordering of
the set of target IP’s, so connections from different clients (across
the world) would be split up across each of the addresses. Example: In
a `ChannelPool` to manage connections to `foo.example.com`, it may be
desirable for high QPS applications to spread the requests across all
available network addresses. Currently, Netty’s resolver would return
only the first address (`10.0.0.1`) to use. Let say we are making
dozens of connections. The name would be resolved to a single IP and
all of the connections would be made to `10.0.0.1`. The other two
addresses would not see any connections. (they may see it later if new
connections are made and `10.0.0.2` is the first in the list at that
time of a subsequent resolution). In these changes, I add support to
select a random one of the resolved addresses to use on each resolve
call, all while leveraging the existing caching and inflight request
detection. This way in my example, the connections would be make to
random selections of the resolved IP addresses.
Modifications:
I added another method `newAddressResolver` to
`DnsAddressResolverGroup` which can be overriden much like
`newNameResolver`. The current functionality which creates
`InetSocketAddressResolver` is still used. I added
`RoundRobinDnsAddressResolverGroup` which extends
DnsAddressResolverGroup and overrides the `newAddressResolver` method
to return a subclass of the `InetSocketAddressResolver`. This subclass
is called `RoundRobinInetSocketAddressResolver` and it contains logic
that takes a `resolve` request, does a `resolveAll` under the hood, and
returns a single element at random from the result of the `resolveAll`.
Result:
The existing functionality of `DnsAddressResolverGroup` is left
unchanged. All new functionality is in the
`RoundRobinInetSocketAddressResolver` which users will now have the
option to use.
Motivation:
If the user removes the SslHandler while still in the processing loop we will produce an IllegalReferenceCountException. We should stop looping when the handlerwas removed.
Modifications:
Ensure we stop looping when the handler is removed.
Result:
No more IllegalReferenceCountException.
Motivation:
The weight header with the default value is not set but it should be (rfc7540#5.3.5: …Pushed streams initially depend on their associated stream … are assigned a default weight of 16).
Modifications:
Add STREAM_WEIGHT header.
Result:
Correctly add headers.
Motivation:
When using java.nio.DatagramChannel we should not close the channel when a SocketException was thrown as we can still use the channel.
Modifications:
Not close the Channel when SocketException is thrown
Result:
More robust and correct handling of exceptions when using NioDatagramChannel.
Motivation:
If an exception is thrown while processing the ready channels in the EventLoop we should still run all tasks as this may allow to recover. For example a OutOfMemoryError may be thrown and runAllTasks() will free up memory again. Beside this we should also ensure we always allow to shutdown even if an exception was thrown.
Modifications:
- Call runAllTasks() in a finally block
- Ensure shutdown is always handles.
Result:
More robust EventLoop implementations for NIO and Epoll.
Motivation:
We should better first process OP_WRITE before OP_READ as this may allow us to free memory in a faster fashion for previous queued writes.
Modifications:
Process OP_WRITE before OP_READ
Result:
Free memory faster for queued writes.
Motivation:
The responsibility for retaining the settings values and enforcing the settings constraints is spread out in different areas of the code and may be initialized with different values than the default specified in the RFC. This should not be allowed by default and interfaces which are responsible for maintaining/enforcing settings state should clearly indicate the restrictions that they should only be set by the codec upon receipt of a SETTINGS ACK frame.
Modifications:
- Encoder, Decoder, and the Headers Encoder/Decoder no longer expose public constructors that allow the default settings to be changed.
- Http2HeadersDecoder#maxHeaderSize() exists to provide some bound when headers/continuation frames are being aggregated. However this is roughly the same as SETTINGS_MAX_HEADER_LIST_SIZE (besides the 32 byte octet for each header field) and can be used instead of attempting to keep the two independent values in sync.
- Encoding headers now enforces SETTINGS_MAX_HEADER_LIST_SIZE at the octect level. Previously the header encoder compared the number of header key/value pairs against SETTINGS_MAX_HEADER_LIST_SIZE instead of the number of octets (plus 32 bytes overhead).
- DefaultHttp2ConnectionDecoder#onData calls shouldIgnoreHeadersOrDataFrame but may swallow exceptions from this method. This means a STREAM_RST frame may not be sent when it should for an unknown stream and thus violate the RFC. The exception is no longer swallowed.
Result:
Default settings state is enforced and interfaces related to settings state are clarified.
Motivation:
The HTTP/2 child channel API does not allow to create local/outbound HTTP/2 streams.
Modifications:
Add a Http2StreamChannelBootstrap that allows to create outbound streams.
Result:
The HTTP/2 child channel API now supports outbound streams.
Motivation:
HttpToHttp2ConnectionHandler.write(ctx, msg, promise) ignores HttpConversionUtil.ExtensionHeaderNames.STREAM_DEPENDENCY_ID header in outbound message. Beside this InboundHttp2ToHttpAdapter also not add the STREAM_DEPENDENCY_ID and STREAM_WEIGHT headers.
Modifications:
Respect STREAM_DEPENDENCY_ID and STREAM_WEIGHT
Result:
Correctly respect STREAM_DEPENDENCY_ID and STREAM_WEIGHT.
Motivation:
Add test-case for doing mutal auth with a certificate chain that holds more then one certificate.
Modifications:
Add test case
Result:
more tests.
Motivation:
the build doesnt seem to enforce this, so they piled up
Modifications:
removed unused import lines
Result:
less unused imports
Signed-off-by: radai-rosenblatt <radai.rosenblatt@gmail.com>
Motivation:
The Javadocs of HttpUtil.getContentLength(HttpMessage, long) and its int overload state that the provided default value is returned if the Content-Length value is not a number. NumberFormatException is thrown instead.
Modifications:
Correctly handle when the value is not a number.
Result:
API works as stated in javadocs.
Motivation:
It'd be usually good to use the latest library version.
Modification:
Bumped JMH to the latest version as of today.
Result:
Now we use JMH version 1.14.1 for our benchmark.
Motivation:
When using the AF_UNIX PROXY protocol, the reader index was not set correctly after consuming the message bytes of the original header ByteBuf. This caused no immediate harm because after the codepath there is no consumer of the ByteBuf in the current implementation. It’s a bug nevertheless, because consumers of the ByteBuf for extensions (like TLVs, which are allowed by the PROXY protocol spec) would consume a ByteBuf that has a wrong readerIndex when using AF_UNIX instead of e.g. IPv4 (which has correct behaviour)
Modifications:
Increase the reader index of the ByteBuf after it was read
Result:
Correct and consistent behaviour of the AF_UNIX codepath
Motivation:
Http2EventAdapter implements the Http2FrameListener interface but implements the #onUnknownFrame(...) method without the interface's throws Http2Exception.
Modifications:
Add throws Http2Exception.
Result:
More correct method signature.
Motivation:
ResourceLeakDetector reports leak for first call to open(obj) as its leakCheckCnt starts with value 0 and increment subsequently. with value of leakCheckCnt =0, it always returns ResourceLeak. Our application calls ResourceLeakDetector.open(obj) to validate Leak and it fails at very first call even though there is no leak in application.
Modifications:
ResourceLeakDetector.leakCheckCnt value will not be 0 while deriving leak and it will not return incorrect value of ResourceLeak.
Result:
Fix false leak report on first call on ResourceLeakDetector.
the implicit #fireChannelReadComplete() in EmbeddedChannel#writeInbound().
Motivation
We use EmbeddedChannels to implement a ProxyChannel of some sorts that shovels
messages between a source and a destination Channel. The latter are real network
channels (such as Epoll) and they may or may not be managed in a ChannelPool. We
could fuse both ends directly together but the EmbeddedChannel provides a nice
disposable section of a ChannelPipeline that can be used to instrument the messages
that are passing through the proxy portion.
The ideal flow looks abount like this:
source#channelRead() -> proxy#writeOutbound() -> destination#write()
source#channelReadComplete() -> proxy#flushOutbound() -> destination#flush()
destination#channelRead() -> proxy#writeInbound() -> source#write()
destination#channelReadComplete() -> proxy#flushInbound() -> source#flush()
The problem is that #writeOutbound() and #writeInbound() emit surplus #flush()
and #fireChannelReadComplete() events which in turn yield to surplus #flush()
calls on both ends of the pipeline.
Modifications
Introduce a new set of write methods that reain the same sematics as the #write()
method and #flushOutbound() and #flushInbound().
Result
It's possible to implement the above ideal flow.
Fix for EmbeddedChannel#ensureOpen() and Unit Tests for it
Some PR stuff.
Motivation:
The StreamBufferingEncoder is missing documentation of what happens
to buffered frames when it's closed.
Modifications:
Added this missing piece of documentation.
Result:
Improved documentation.
Motivation:
Currently FlushConsolidationHandler only consolidates if a read loop is
active for a Channel, otherwise each writeAndFlush(...) call will still
be flushed individually. When these calls are close enough, it can be
beneficial to consolidate them even outside of a read loop.
Modifications:
When we allow a flush to "go through", don't perform it immediately, but
submit it on the channel's executor. Under high pressure, this gives
other writes a chance to enqueue before the task gets executed, and so
we flush multiple writes at once.
Result:
Lower CPU usage and less context switching.
Motivation:
We need to ensure we release all direct memory once the DirectPoolArena is collected. Otherwise we may never reclaim the memory and so leak memory.
Modifications:
Ensure we destroy all PoolChunk memory when DirectPoolArena is collected.
Result:
Free up unreleased memory when DirectPoolArena is collected.