Motivation:
The encoder is currently responsible for chunking frames when writing in order to conform to max frame size. The frame writer would be a better place for this since it could perform a reuse the same promise aggregator for all the write and could also perform a single allocation for all of the frame headers.
Modifications:
Modified `DefaultHttp2FrameWriter` to perform the chunking and modified the contract in the `Http2FrameWriter` interface. Modified `DefaultHttp2ConnectionEncoder` to send give all allocated bytes to the writer.
Result:
Fixes#3966
Motivation:
The UniformStreamByteDistributor currently processes all zero-length frames, regardless of add order. This means that we would always send HEADERS for all streams, possibly taking away bandwidth for streams that actually have data.
Modifications:
Empty frames are now treated the same as any other frame except that the algorithm will pop off the any empty frames at the head of the queue.
Result:
Empty frames require no extra processing.
Motivation:
DefaultHeaders creates an array of size 16 for all headers. This may waste a good deal of memory if applications only have a small number of headers. This memory may be critical when the number of connections grows large.
Modifications:
- Make the size of the array for DefaultHeaders configurable
Result:
Applications can control the size of the DefaultHeaders array and save memory.
Motiviation:
According to jetty docs the alpn-api should use the provided scope.
Modificaitons:
- change scope to provided for alpn-api
- update for new jdk
Result:
Users of Netty don't run into alpn version conflicts.
Fixes https://github.com/netty/netty/issues/4480
Motivation:
exceptionCaught(...) will only handle inbound exceptions which means it makes not much sense to have it also on ChannelOutboundHandler. Because of this we should move it to ChannelInboundHandler.
Modifications:
Add @deprecated annotation to ChannelHandler.exceptionCaught(...).
Result:
Preapre to cleanup the API in later release.
Motivation:
6.7 is the latest stable release in RHEL/CentOS 6 line. Given that most
RHEL/CentOS users have upgraded to 6.7 via yum upgrade, we should bump
our requirement.
Modification:
s/6.6/6.7/g
Result:
'mvn release:*' must be run on RHEL/CentOS 6.7 instead of 6.6.
Motivation:
AbstractEpollStreamChannel has a queue which collects splice events. Splice is assumed not to be the most common use case of this class and thus the splice queue could be initialized in a lazy fashion to save memory. This becomes more significant when the number of connections grows.
Modifications:
- AbstractEpollStreamChannel.spliceQueue will be initialized in a lazy fashion
Result:
Less memory consumption for most use cases
Motivation:
2a2059d976 was backported from master, and included an overriden method which does not exist in 4.1.
Modifications:
- Remove the invoker method from NoPriorityByteDistributionBenchmark
Result:
No more build error
Motivation:
We should use OneTimeTask where possible to reduce object creation.
Modifications:
Replace Runnable with OneTimeTask
Result:
Less object creation
Motivation:
There is a notification ordering issue in DefaultPromise when the lateListener collection is in use. The ordering issue can be observed in situations where a late listener is added to a Future returned from a write operation. It is possible that this future will run after a read operation scheduled on the I/O thread, even if the late listener is added on the I/O thread. This can lead to unexpected ordering where a listener for a write operation which must complete in order for the read operation to happen is notified after the read operation is done.
Modifications:
- If the lateListener collection becomes empty, it should be treated as though it was null when checking if lateListeners can be notified immediatley (instead of executing a task on the executor)
Result:
Ordering is more natural and will not be perceived as being out of order relative to other tasks on the same executor.
Motivation:
We not need to store another reference to AbstractChannel as we can access it through DefaultChannelHandlerContext.
Modifications:
Remove reference.
Result:
Cleaner code.
Motivation:
If you start to have 1M+ concurrent connections memory footprint can be come a big issue. We should try to reduce it as much as possible in the core of netty.
Modifications:
- Remove HashMap that was used to store name to ctx mapping. This was only used for validation and access a handler by name. As a pipeline is not expected to be very long (like 100+ handlers) we can just walk the linked list structure to find the ctx with a given name.
Result:
Less memory footprint of the DefaultChannelPipeline.
Motivation:
If we have a lot of writes going on we currently need to lookup the IovArray for each Channel that does writes. This can have quite some perf overhead. We should not need to do this and just store a reference of the IovArray on the EpollEventLoop itself.
Modifications:
- Remove IoArrayThreadLocal
- Store the IoArray in the EventLoop itself
Result:
Less FastThreadLocal lookups
Motivation:
The current priority algorithm can yield poor per-stream goodput when either the number of streams is high or the connection window is small. When all priorities are the same (i.e. priority is disabled), we should be able to do better.
Modifications:
Added a new UniformStreamByteDistributor that ignores priority entirely and manages a queue of streams. Each stream is allocated a minimum of 1KiB on each iteration.
Result:
Improved goodput when priority is not used.
Motivation:
If ChannelOption.ALLOW_HALF_CLOSURE is true and the shutdown input operation fails we should not propagate this exception, and instead consider this socket's read as half closed.
Modifications:
- AbstractEpollChannel.shutdownInput should not propagate exceptions when attempting to shutdown the input, but instead should just close the socket
Result:
Users expecting a ChannelInputShutdownEvent will get this event even if the socket is already shutdown, and the shutdown operation fails.
Motivation:
HttpHeaders already has specific methods for such popular and simple headers like "Host", but if I need to convert POST raw body to string I need to parse complex ContentType header in my code.
Modifications:
Add getCharset and getCharsetAsString methods to parse charset from Content-Length header.
Result:
Easy to use utility method.
Motivation:
The `DefaultHttp2RemoteFlowController` does not correctly determine `hasFrame` when updating the stream state for the distributor. Adding a check to enforce `hasFrame` when `streamableBytes > 0` causes several test failures.
Modifications:
Modified `DefaultHttp2RemoteFlowController` to simplify the writing logic and to correct the bookkeeping for `hasFrame`.
Result:
The distributors are always called with valid arguments.
Motivation:
The method setBytes did not work correctly because read-only ByteBuffer
does not allow access to its underlying array.
Modifications:
New case was added for ByteBuffer's that are not direct and do not have an array.
These must be handled by copying the data into a temporary array. Unit test was
added to test this case.
Result:
It is now possible to use read-only ByteBuffer as the source
for the setBytes method.
Motivation:
The twitter hpack project does not have the support that it used to have. See discussion here: https://github.com/netty/netty/issues/4403.
Modifications:
Created a new module in Netty and copied the latest from twitter hpack master.
Result:
Netty no longer depends on twitter hpack.
Motivation:
The AsciiString.hashCode() method can be optimized. This method is frequently used while to build the DefaultHeaders data structure.
Modification:
- Add a PlatformDependent hashCode algorithm which utilizes UNSAFE if available
Result:
AsciiString hashCode is faster.
Motivation:
FullHttp[Request|Response].hashCode() uses a releasable object and in vulnerable to a IllegalRefCountException if that object has been released.
Modifications:
- Ensure the released object is not used.
Result:
No more IllegalRefCountException.
Related: #3972
Motivation:
DnsNameResolver limits the number of concurrent in-progress DNS queries
to 65536 regardless the number of DNS servers it communicates with. When
the number of available DNS servers are more than just one, we end up
using much less (65536 / numDnsServers) query IDs per DNS server, which
is non-optimal.
Modifications:
- Replace the query ID and context management with
DnsQueryContextManager
- Eash DNS server gets its own query ID space
Result:
Much bigger query ID space, and thus there's less chance of getting the
'query ID space exhaustion' error
Motivation:
The HTTP/2 client example is not validating the results of ALPN if TLS is enabled.
Modifications:
- Use ApplicationProtocolNegotiationHandler to validate ALPN results.
Result:
Client example validates ALPN results.
Motivation:
Recently a bug was found in DefaultHttp2Headers where the state of the headers could be corrupted due to the extra tracking to make pseudo headers first during iteration. Unit tests did not catch this bug.
Modifications:
- Update unit tests to cover more methods
Result:
Unit tests for DefaultHttp2Headers have better code coverage.
Motivation:
A new version of ALPN boot has been released.
Modifications:
- Update the pom to pull in this new version
Result:
New JDK get new ALPN boot.
Motivation:
Child classes of ApplicationProtocolNegotiationHandler may want to override the behavior when a handshake failure is detected.
Modifications:
- Provide a method which can be overriden when a handshake failure is detected.
Result:
Child classes can override ApplicationProtocolNegotiationHandler handshake failure behavior.
Motivation:
Headers and groups of headers are frequently copied and the current mechanism is slower than it needs to be.
Modifications:
Skip name validation and hash computation when they are not necessary.
Fix emergent bug in CombinedHttpHeaders identified with better testing
Fix memory leak in DefaultHttp2Headers when clearing
Added benchmarks
Result:
Faster header copying and some collateral bug fixes
Motivation:
The previous DefaultChannelPipeline#destroy() implementation, introduced in #3156, is suboptimal as it can cause the for loop to continuously spin if the executor used by a given handler is unable to "recognize" the event loop.
It could be objected that it's the custom executor responsibility to properly implement the inEventLoop() method, but some implementetaions might not be able to do that for performance reasons, and even so, it's always better to be safe against API misuse, in particular when it is not possible to fail fast and the alternative is rather some sutle behaviour.
Modifications:
The patch simply avoids the recursive spin by explicitly passing the "in event loop" condition as a boolean parameter, preserving the same guarantees offered by #3156. A unit test has also been added.
Result:
All channel events are correctly called and no high CPU usage is seen anymore.
Motivation:
Makes the API contract of headers more consistent and simpler.
Modifications:
If self is passed to set then simply return
Result:
set and setAll will be consistent
Motivation:
For many HTTP/2 applications (such as gRPC) it is necessary to autorefill the connection window in order to prevent application-level deadlocking.
Consider an application with 2 streams, A and B. A receives a stream of messages and the application pops off one message at a time and makes a request on stream B. However, if receiving of data on A has caused the connection window to collapse, B will not be able to receive any data and the application will deadlock. The only way (currently) to get around this is 1) use multiple connections, or 2) manually refill the connection window. Both are undesirable and could needlessly complicate the application code.
Modifications:
Add a configuration option to DefaultHttp2LocalFlowController, allowing it to autorefill the connection window.
Result:
Applications can configure HTTP/2 to avoid inter-stream deadlocking.
Motivation:
If netty used as part of application, should be a way to prefix service thread name to easy distinguish such threads (for example, used in IntelliJ Platform)
Modifications:
Introduce system property io.netty.serviceThreadPrefix
Result:
ThreadDeathWatcher thread has a readable name "Netty threadDeathWatcher-2-1" if io.netty.serviceThreadPrefix set to "Netty"
Motivation:
Changing the chache of generated names to use a cache per thread. This will remove the bottleneck when many eventloops are used and names need to generate.
Modifications:
Use a FastThreadLocal to store the cached names.
Result:
Less locking between threads.
Motivation:
PriorityStreamByteDistributor saves exception state and attempts to reset state. This could be simplified by just throwing a connection error and closing the connection. PriorityStreamByteDistributor also does not handle or detect re-entry in the distribute method.
Motivation:
- PriorityStreamByteDistributor propagate an INTERNAL_ERROR if an exception occurs during writing
- PriorityStreamByteDistributor to handle re-entry on the write method
Result:
PriorityStreamByteDistributor exception code state simplified, and re-entry is detected.
Motivation:
We should allow our custom Executor to shutdown quickly.
Modifications:
Call super constructor which correct arguments.
Result:
Custom Executor can be shutdown quickly.
Motivation:
Http2ConnectionHandler verifies if the first frame after the preface is
a SETTINGS frame. However, it does not reject the SETTING ack frame
which is not expected actually.
Modifications:
Reject a SETTINGS-ack frame as well
Result:
When the first frame is a SETTINGS-ack frame, connection does not
proceed to further frame handling. (simplicity)
Motivation:
The EPOLL module was not completly respecting the half closed state. It may have missed events, or procssed events when it should not have due to checking isOpen instead of the appropriate shutdown state.
Modifications:
- use FileDescriptor's isShutdown* methods instead of isOpen to check for processing events.
Result:
Half closed code in EPOLL module is more correct.
Motivation:
transport-native-epoll is designed to be specific to Linux. However there is native code that can be extracted out and made to work on more Unix like distributions. There are a few steps to be completely decoupled but the first step is to extract out code that can run in a more general Unix environment from the Linux specific code base.
Modifications:
- Move all non-Linux specific stuff from Native.java into the io.netty.channel.unix package.
- io.netty.channel.unix.FileDescriptor will inherit all the native methods that are specific to file descriptors.
- io_netty_channel_epoll_Native.[c|h] will only have code that is specific to Linux.
Result:
Code is decoupled and design is streamlined in FileDescriptor.
Motivation:
The HTTP/2 RFC (https://tools.ietf.org/html/rfc7540#section-8.1.2) indicates that header names consist of ASCII characters. We currently use ByteString to represent HTTP/2 header names. The HTTP/2 RFC (https://tools.ietf.org/html/rfc7540#section-10.3) also eludes to header values inheriting the same validity characteristics as HTTP/1.x. Using AsciiString for the value type of HTTP/2 headers would allow for re-use of predefined HTTP/1.x values, and make comparisons more intuitive. The Headers<T> interface could also be expanded to allow for easier use of header types which do not have the same Key and Value type.
Motivation:
- Change Headers<T> to Headers<K, V>
- Change Http2Headers<ByteString> to Http2Headers<CharSequence, CharSequence>
- Remove ByteString. Having AsciiString extend ByteString complicates equality comparisons when the hash code algorithm is no longer shared.
Result:
Http2Header types are more representative of the HTTP/2 RFC, and relationship between HTTP/2 header name/values more directly relates to HTTP/1.x header names/values.
Motivation:
When DnsNameResolverContext succeeds to get the address(es), it cancels
the promise of other queries in progress.
Unlike expectation, DnsNameResolverContext.query() attempts to retry
even when the query has failed due to cancellation.
As a result, the resolver sends unnecessary extra queries to a DNS
server and then tries to mark the promised that's been fulfilled
already, leading to unnecessarily verbose 'failed to notify success to a
promise' messages.
Modifications:
Do not perform an extra query when the previous query has failed due to
cancellation
Result:
DnsNameResolver does not send unnecessary extra queries and thus does
not log the 'failed to notify success to a promise' message.
Motivation:
DefaultPromise.toString() returns 'DefaultPromise(incomplete)' when it's
actually complete with non-null result.
Modifications:
Handle the case where the promise is done and its result is non-null in
toString()
Result:
The String returned by DefaultPromise.toString() is not confusing
anymore.
Motivation:
As reported in #4402, the FastThreadLocalBenchmark shows that the JDK ThreadLocal
is actually faster than Netty's custom thread local implementation.
I was looking forward to doing some deep digging, but got disappointed :(.
Modifications:
The microbenchmark was not using FastThreadLocalThreads and would thus always hit the slow path.
I updated the JMH command line flags, so that FastThreadLocalThreads would be used.
Result:
FastThreadLocalBenchmark shows FastThreadLocal to be faster than JDK's ThreadLocal implementation,
by about 56% in this particular benchmark. Run on OSX El Capitan with OpenJDK 1.8u60.
Benchmark Mode Cnt Score Error Units
FastThreadLocalBenchmark.fastThreadLocal thrpt 20 55452.027 ± 725.713 ops/s
FastThreadLocalBenchmark.jdkThreadLocalGet thrpt 20 35481.888 ± 1471.647 ops/s
Motivation:
To prove one implementation is faster as the other we should have a benchmark.
Modifications:
Add benchmark which benchmarks the unsafe and non-unsafe implementation of HeapByteBuf.
Result:
Able to compare speed of implementations easily.
Motivation:
We should only call ReferenceCountUtil.touch(...) if needed as otherwise we pay the overhead of instanceof and cast
everytime.
Modifications:
Add boolean flag which indicates if touch(...) should be called.
Result:
Less overhead when leak detection is not enabled.
Motivation:
Modulo operations are slow, we can use bitwise operation to detect if resource leak detection must be done while sampling.
Modifications:
- Ensure the interval is a power of two
- Use bitwise operation for sampling
- Add benchmark.
Result:
Faster sampling.
Motivation:
When the ImmediateEventExecutor is in use it is possible to get a StackOverFlowException if when a promise completes a new listener is added to that promise.
Modifications:
- Protect against the case where LateListeners.run() smashes the stack.
Result:
Fixes https://github.com/netty/netty/issues/4395
