Motivation:
The wakeup logic in EpollEventLoop is overly complex
Modification:
* Simplify the race to wakeup the loop
* Dont let the event loop wake up itself (it's already awake!)
* Make event loop check if there are any more tasks after preparing to
sleep. There is small window where the non-eventloop writers can issue
eventfd writes here, but that is okay.
Result:
Cleaner wakeup logic.
Benchmarks:
```
BEFORE
Benchmark Mode Cnt Score Error Units
EpollSocketChannelBenchmark.executeMulti thrpt 20 408381.411 ± 2857.498 ops/s
EpollSocketChannelBenchmark.executeSingle thrpt 20 157022.360 ± 1240.573 ops/s
EpollSocketChannelBenchmark.pingPong thrpt 20 60571.704 ± 331.125 ops/s
Benchmark Mode Cnt Score Error Units
EpollSocketChannelBenchmark.executeMulti thrpt 20 440546.953 ± 1652.823 ops/s
EpollSocketChannelBenchmark.executeSingle thrpt 20 168114.751 ± 1176.609 ops/s
EpollSocketChannelBenchmark.pingPong thrpt 20 61231.878 ± 520.108 ops/s
```
Motivation
Pipeline handlers are free to "take control" of input buffers if they have singular refcount - in particular to mutate their raw data if non-readonly via discarding of read bytes, etc.
However there are various places (primarily unit tests) where a wrapped byte-array buffer is passed in and the wrapped array is assumed not to change (used after the wrapped buffer is passed to EmbeddedChannel.writeInbound()). This invalid assumption could result in unexpected errors, such as those exposed by #8931.
Modifications
Anywhere that the data passed to writeInbound() might be used again, ensure that either:
- A copy is used rather than wrapping a shared byte array, or
- The buffer is otherwise protected from modification by making it read-only
For the tests, copying is preferred since it still allows the "mutating" optimizations to be exercised.
Results
Avoid possible errors when pipeline assumes it has full control of input buffer.
Motivation:
Results are just wrong for small delays.
Modifications:
Switching to AvarageTime avoid to rely on OS nanoTime granularity.
Result:
Uncontended low delay results are not reliable
Motivation:
In 42742e233f we already added default methods to Channel*Handler and deprecated the Adapter classes to simplify the class hierarchy. With this change we go even further and merge everything into just ChannelHandler. This simplifies things even more in terms of class-hierarchy.
Modifications:
- Merge ChannelInboundHandler | ChannelOutboundHandler into ChannelHandler
- Adjust code to just use ChannelHandler
- Deprecate old interfaces.
Result:
Cleaner and simpler code in terms of class-hierarchy.
Motivation:
As we now us java8 as minimum java version we can deprecate ChannelInboundHandlerAdapter / ChannelOutboundHandlerAdapter and just move the default implementations into the interfaces. This makes things a bit more flexible for the end-user and also simplifies the class-hierarchy.
Modifications:
- Mark ChannelInboundHandlerAdapter and ChannelOutboundHandlerAdapter as deprecated
- Add default implementations to ChannelInboundHandler / ChannelOutboundHandler
- Refactor our code to not use ChannelInboundHandlerAdapter / ChannelOutboundHandlerAdapter anymore
Result:
Cleanup class-hierarchy and make things a bit more flexible.
Motivation:
Netty is very widely used which can lead to a lot of pain when we break API / ABI. We should make use japicmp-maven-plugin during the build to verify we do not introduce breakage by mistake.
Modifications:
- Add japicmp-maven-plugin to the build process
- Fix a method signature change in HttpProxyHandler that was flagged as a possible problem.
Result:
Ensure no API/ABI breakage accour between releases.
Motivation:
We can remove some properties for which we introduced replacements.
Modifications:
io.netty.buffer.bytebuf.checkAccessible, io.netty.leakDetectionLevel, org.jboss.netty.tryUnsafe properties removed
Result:
Code cleanup
Motivation:
We can just use Objects.requireNonNull(...) as a replacement for ObjectUtil.checkNotNull(....)
Modifications:
- Use Objects.requireNonNull(...)
Result:
Less code to maintain.
Motivation:
Just was looking through code and found 1 interesting place DateFormatter.tryParseMonth that was not very effective, so I decided to optimize it a bit.
Modification:
Changed DateFormatter.tryParseMonth method. Instead of invocation regionMatch() for every month - compare chars one by one.
Result:
DateFormatter.parseHttpDate method performance improved from ~3% to ~15%.
Benchmark (DATE_STRING) Mode Cnt Score Error Units
DateFormatter2Benchmark.parseHttpHeaderDateFormatter Sun, 27 Jan 2016 19:18:46 GMT thrpt 6 4142781.221 ± 82155.002 ops/s
DateFormatter2Benchmark.parseHttpHeaderDateFormatter Sun, 27 Dec 2016 19:18:46 GMT thrpt 6 3781810.558 ± 38679.061 ops/s
DateFormatter2Benchmark.parseHttpHeaderDateFormatterNew Sun, 27 Jan 2016 19:18:46 GMT thrpt 6 4372569.705 ± 30257.537 ops/s
DateFormatter2Benchmark.parseHttpHeaderDateFormatterNew Sun, 27 Dec 2016 19:18:46 GMT thrpt 6 4339785.100 ± 57542.660 ops/s
Motivation:
ChannelHandler.exceptionCaught(...) was marked as @deprecated as it should only exist in inbound handlers.
Modifications:
Remove ChannelHandler.exceptionCaught(...) and adjust code / tests.
Result:
Fixes https://github.com/netty/netty/issues/8527
Motivation:
HttpHeaderDateFormat was replaced with DateFormatter many days ago and now can be easily removed.
Modification:
Remove deprecated class and related test / benchmark
Result:
Less code to maintain
Motivation:
We can use lambdas now as we use Java8.
Modification:
use lambda function for all package, #8751 only migrate transport package.
Result:
Code cleanup.
Motivation:
We need to update to a new checkstyle plugin to allow the usage of lambdas.
Modifications:
- Update to new plugin version.
- Fix checkstyle problems.
Result:
Be able to use checkstyle plugin which supports new Java syntax.
* Decouble EventLoop details from the IO handling for each transport to allow easy re-use of code and customization
Motiviation:
As today extending EventLoop implementations to add custom logic / metrics / instrumentations is only possible in a very limited way if at all. This is due the fact that most implementations are final or even package-private. That said even if these would be public there are the ability to do something useful with these is very limited as the IO processing and task processing are very tightly coupled. All of the mentioned things are a big pain point in netty 4.x and need improvement.
Modifications:
This changeset decoubled the IO processing logic from the task processing logic for the main transport (NIO, Epoll, KQueue) by introducing the concept of an IoHandler. The IoHandler itself is responsible to wait for IO readiness and process these IO events. The execution of the IoHandler itself is done by the SingleThreadEventLoop as part of its EventLoop processing. This allows to use the same EventLoopGroup (MultiThreadEventLoupGroup) for all the mentioned transports by just specify a different IoHandlerFactory during construction.
Beside this core API change this changeset also allows to easily extend SingleThreadEventExecutor / SingleThreadEventLoop to add custom logic to it which then can be reused by all the transports. The ideas are very similar to what is provided by ScheduledThreadPoolExecutor (that is part of the JDK). This allows for example things like:
* Adding instrumentation / metrics:
* how many Channels are registered on an SingleThreadEventLoop
* how many Channels were handled during the IO processing in an EventLoop run
* how many task were handled during the last EventLoop / EventExecutor run
* how many outstanding tasks we have
...
...
* Implementing custom strategies for choosing the next EventExecutor / EventLoop to use based on these metrics.
* Use different Promise / Future / ScheduledFuture implementations
* decorate Runnable / Callables when submitted to the EventExecutor / EventLoop
As a lot of functionalities are folded into the MultiThreadEventLoopGroup and SingleThreadEventLoopGroup this changeset also removes:
* AbstractEventLoop
* AbstractEventLoopGroup
* EventExecutorChooser
* EventExecutorChooserFactory
* DefaultEventLoopGroup
* DefaultEventExecutor
* DefaultEventExecutorGroup
Result:
Fixes https://github.com/netty/netty/issues/8514 .
Motivation:
Custom Netty ThreadLocalRandom and ThreadLocalRandomProvider classes are no longer needed and can be removed.
Modification:
Remove own ThreadLocalRandom
Result:
Less code to maintain
Motivation:
We can use the diamond operator these days.
Modification:
Use diamond operator whenever possible.
Result:
More modern code and less boiler-plate.
Motivation:
Invoking ChannelHandlers is not free and can result in some overhead when the ChannelPipeline becomes very long. This is especially true if most handlers will just forward the call to the next handler in the pipeline. When the user extends Channel*HandlerAdapter we can easily detect if can just skip the handler and invoke the next handler in the pipeline directly. This reduce the overhead of dispatch but also reduce the call-stack in many cases.
Modifications:
Detect if we can skip the handler when walking the pipeline.
Result:
Reduce overhead for long pipelines.
Benchmark (extraHandlers) Mode Cnt Score Error Units
DefaultChannelPipelineBenchmark.propagateEventOld 4 thrpt 10 267313.031 ± 9131.140 ops/s
DefaultChannelPipelineBenchmark.propagateEvent 4 thrpt 10 824825.673 ± 12727.594 ops/s
Motiviation:
Because of how we implemented the registration / deregistration of an EventLoop it was not possible to wrap an EventLoop implementation and use it with a Channel.
Modification:
- Introduce EventLoop.Unsafe which is responsible for the actual registration.
- Move validation of EventLoop / Channel combo to the EventLoop
- Add unit test that verifies that wrapping works
Result:
Be able to wrap an EventLoop and so add some extra functionality.
Motivation:
At the moment it’s possible to have a Channel in Netty that is not registered / assigned to an EventLoop until register(...) is called. This is suboptimal as if the Channel is not registered it is also not possible to do anything useful with a ChannelFuture that belongs to the Channel. We should think about if we should have the EventLoop as a constructor argument of a Channel and have the register / deregister method only have the effect of add a Channel to KQueue/Epoll/... It is also currently possible to deregister a Channel from one EventLoop and register it with another EventLoop. This operation defeats the threading model assumptions that are wide spread in Netty, and requires careful user level coordination to pull off without any concurrency issues. It is not a commonly used feature in practice, may be better handled by other means (e.g. client side load balancing), and therefore we propose removing this feature.
Modifications:
- Change all Channel implementations to require an EventLoop for construction ( + an EventLoopGroup for all ServerChannel implementations)
- Remove all register(...) methods from EventLoopGroup
- Add ChannelOutboundInvoker.register(...) which now basically means we want to register on the EventLoop for IO.
- Change ChannelUnsafe.register(...) to not take an EventLoop as parameter (as the EventLoop is supplied on custruction).
- Change ChannelFactory to take an EventLoop to create new Channels and introduce ServerChannelFactory which takes an EventLoop and one EventLoopGroup to create new ServerChannel instances.
- Add ServerChannel.childEventLoopGroup()
- Ensure all operations on the accepted Channel is done in the EventLoop of the Channel in ServerBootstrap
- Change unit tests for new behaviour
Result:
A Channel always has an EventLoop assigned which will never change during its life-time. This ensures we are always be able to call any operation on the Channel once constructed (unit the EventLoop is shutdown). This also simplifies the logic in DefaultChannelPipeline a lot as we can always call handlerAdded / handlerRemoved directly without the need to wait for register() to happen.
Also note that its still possible to deregister a Channel and register it again. It's just not possible anymore to move from one EventLoop to another (which was not really safe anyway).
Fixes https://github.com/netty/netty/issues/8513.
Motivation:
ByteBuf supports “marker indexes”. The intended use case for these is if a speculative operation (e.g. decode) is in process the user can “mark” and interface and refer to it later if the operation isn’t successful (e.g. not enough data). However this is rarely used in practice,
requires extra memory to maintain, and introduces complexity in the state management for derived/pooled buffer initialization, resizing, and other operations which may modify reader/writer indexes.
Modifications:
Remove support for marking and adjust testcases / code.
Result:
Fixes https://github.com/netty/netty/issues/8535.
Motivation:
Netty executors doesn't have yet any means to compare with each others
nor to compare with the j.u.c. executors
Modifications:
A new benchmark measuring execute burst cost is being added
Result:
It's now possible to compare some of Netty executors with each others
and with the j.u.c. executors
Motivation:
CompositeByteBuf is a powerful and versatile abstraction, allowing for
manipulation of large data without copying bytes. There is still a
non-negligible cost to reading/writing however relative to "singular"
ByteBufs, and this can be mostly eliminated with some rework of the
internals.
My use case is message modification/transformation while zero-copy
proxying. For example replacing a string within a large message with one
of a different length
Modifications:
- No longer slice added buffers and unwrap added slices
- Components store target buf offset relative to position in
composite buf
- Less allocations, object footprint, pointer indirection, offset
arithmetic
- Use Component[] rather than ArrayList<Component>
- Avoid pointer indirection and duplicate bounds check, more
efficient backing array growth
- Facilitates optimization when doing bulk-inserts - inserting n
ByteBufs behind m is now O(m + n) instead of O(mn)
- Avoid unnecessary casting and method call indirection via superclass
- Eliminate some duplicate range/ref checks via non-checking versions of
toComponentIndex and findComponent
- Add simple fast-path for toComponentIndex(0); add racy cache of
last-accessed Component to findComponent(int)
- Override forEachByte0(...) and forEachByteDesc0(...) methods
- Make use of RecyclableArrayList in nioBuffers(int, int) (in line with
FasterCompositeByteBuf impl)
- Modify addComponents0(boolean,int,Iterable) to use the Iterable
directly rather than copy to an array first (and possibly to an
ArrayList before that)
- Optimize addComponents0(boolean,int,ByteBuf[],int) to not perform
repeated array insertions and avoid second loop for offset updates
- Simplify other logic in various places, in particular the general
pattern used where a sub-range is iterated over
- Add benchmarks to demonstrate some improvements
While refactoring I also came across a couple of clear bugs. They are
fixed in these changes but I will open another PR with unit tests and
fixes to the current version.
Result:
Much faster creation, manipulation, and access; many fewer allocations
and smaller footprint. Benchmark results to follow.
* Optimize AbstractByteBuf.getCharSequence() in US_ASCII case
Motivation:
Inspired by https://github.com/netty/netty/pull/8388, I noticed this
simple optimization to avoid char[] allocation (also suggested in a TODO
here).
Modifications:
Return an AsciiString from AbstractByteBuf.getCharSequence() if
requested charset is US_ASCII or ISO_8859_1 (latter thanks to
@Scottmitch's suggestion). Also tweak unit tests not to require Strings
and include a new benchmark to demonstrate the speedup.
Result:
Speed-up of AbstractByteBuf.getCharSequence() in ascii and iso 8859/1
cases
Motiviation:
At the moment whenever ensureAccessible() is called in our ByteBuf implementations (which is basically on each operation) we will do a volatile read. That per-se is not such a bad thing but the problem here is that it will also reduce the the optimizations that the compiler / jit can do. For example as these are volatile it can not eliminate multiple loads of it when inline the methods of ByteBuf which happens quite frequently because most of them a quite small and very hot. That is especially true for all the methods that act on primitives.
It gets even worse as people often call a lot of these after each other in the same method or even use method chaining here.
The idea of the change is basically just ue a non-volatile read for the ensureAccessible() check as its a best-effort implementation to detect acting on already released buffers anyway as even with a volatile read it could happen that the user will release it in another thread before we actual access the buffer after the reference check.
Modifications:
- Try to do a non-volatile read using sun.misc.Unsafe if we can use it.
- Add a benchmark
Result:
Big performance win when multiple ByteBuf methods are called from a method.
With the change:
UnsafeByteBufBenchmark.setGetLongUnsafeByteBuf thrpt 20 281395842,128 ± 5050792,296 ops/s
Before the change:
UnsafeByteBufBenchmark.setGetLongUnsafeByteBuf thrpt 20 217419832,801 ± 5080579,030 ops/s
Motivation:
We should use the latest jmh version which also supports -prof dtraceasm on MacOS.
Modifications:
Update to latest jmh version.
Result:
Better benchmark / profiling support on MacOS.
Motivation:
Some of the flags we used are not supported anymore on more recent JDK versions. We should just remove all of them and only keep what we really need. This may also reflect better what people use in production.
Modifications:
Remove some flags when running the benchmarks.
Result:
Benchmarks also run with JDK11.
Motivation:
It is sometimes useful to be able to run benchmarks easily from the commandline and passs different arguments / options here. We should support this.
Modifications:
Add the benchmark-jar profile which allows to generate such an "uber-jar" that can be used directly to run benchmarks as documented at http://openjdk.java.net/projects/code-tools/jmh/.
Result:
More flexible way to run benchmarks.
Motivation:
Optimizing the Epoll channel needs an objective measure of how fast
it is.
Modification:
Add a simple, closed loop, ping-pong benchmark.
Result:
Benchmark can be used to measure #7816
Initial numbers:
```
Result "io.netty.microbench.channel.epoll.EpollSocketChannelBenchmark.pingPong":
22614.403 ±(99.9%) 797.263 ops/s [Average]
(min, avg, max) = (21093.160, 22614.403, 24977.387), stdev = 918.130
CI (99.9%): [21817.140, 23411.666] (assumes normal distribution)
Benchmark Mode Cnt Score Error Units
EpollSocketChannelBenchmark.pingPong thrpt 20 22614.403 ± 797.263 ops/s
```
Motivation:
The JVM isn't always able to hoist out/reduce bounds checking (due to ref counting operations etc etc) hence making it configurable could improve performances for most CPU intensive use cases.
Modifications:
Each AbstractByteBuf bounds check has been tested against a new static final configuration property similar to checkAccessible ie io.netty.buffer.bytebuf.checkBounds.
Result:
Any user could disable ByteBuf bounds checking in order to get extra performances.
Motivation:
The usage of Invocation level for JMH fixture methods (setup/teardown) inccurs in a significant overhead
in the benchmark time (see org.openjdk.jmh.annotations.Level documentation).
In the case of CodecInputListBenchmark, benchmarks are far too small (less than 50ns) and the Invocation
level setup offsets the measurement considerably.
On such cases, the recommended fix patch is to include the setup/teardown code in the benchmark method.
Modifications:
Include the setup/teardown code in the relevant benchmark methods.
Remove the setup/teardown methods from the benchmark class.
Result:
We run the entire benchmark 10 times with default parameters we observed:
- ArrayList benchmark affected directly by JMH overhead is now from 15-80% faster.
- CodecList benchmark is now 50% faster than original (even with the setup code being measured).
- Recyclable ArrayList is ~30% slower.
- All benchmarks have significant different means (ANOVA) and medians (Moore)
Mode: Throughput (Higher the better)
Method Full params Factor Modified (Median) Original (Median)
recyclableArrayList (elements = 1) 0.615520967 21719082.75 35285691.2
recyclableArrayList (elements = 4) 0.699553431 17149442.76 24514843.31
arrayList (elements = 4) 1.152666631 27120407.18 23528404.88
codecOutList (elements = 1) 1.527275908 67251089.04 44033359.47
codecOutList (elements = 4) 1.596917095 59174088.78 37055204.03
arrayList (elements = 1) 1.878616889 62188238.24 33103204.06
Environment:
Tests run on a Computational server with CPU: E5-1660-3.3GHZ (6 cores + HT), 64 GB RAM.
Motivation:
The usage of Invocation level for JMH fixture methods (setup/teardown) inccurs in a significant impact in
in the benchmark time (see org.openjdk.jmh.annotations.Level documentation).
When the benchmark and the setup/teardown is too small (less than a milisecond) the Invocation level might saturate the system with
timestamp requests and iteration synchronizations which introduce artificial latency, throughput, and scalability bottlenecks.
In the HeadersBenchmark, all benchmarks take less than 100ns and the Invocation level setup offsets the measurement considerably.
As fixture methods is defined for the entire class, this overhead also impacts every single benchmark in this class, not only
the ones that use the emptyHttpHeaders object (cleaned in the setup).
The recommended fix patch here is to include the setup/teardown code in the benchmark where the object is used.
Modifications:
Include the setup/teardown code in the relevant benchmark methods.
Remove the setup/teardown method of Invocation level from the benchmark class.
Result:
We run all benchmarks from HeadersBenchmark 10 times with default parameter, we observe:
- Benchmarks that were not directly affected by the fix patch, improved execution time.
For instance, http2Remove with (exampleHeader = THREE) had its median reported as 2x faster than the original version.
- Benchmarks that had the setup code inserted (eg. http2AddAllFastest) did not suffer a significant punch in the execution time,
as the benchmarks are not dominated by the clear().
Environment:
Tests run on a Computational server with CPU: E5-1660-3.3GHZ (6 cores + HT), 64 GB RAM.
Motivation:
NetUtilBenchmark is using out of date data, throws an exception in the benchmark, and allocates a Set on each run.
Modifications:
- Update the benchmark and reduce each run's overhead
Result:
NetUtilBenchmark is updated.
Motivation:
Read-only heap ByteBuffer doesn't expose array: the existent method to perform copies to direct ByteBuf involves the creation of a (maybe pooled) additional heap ByteBuf instance and copy
Modifications:
To avoid stressing the allocator with additional (and stealth) heap ByteBuf allocations is provided a method to perform copies using the (pooled) internal NIO buffer
Result:
Copies from read-only heap ByteBuffer to direct ByteBuf won't create any intermediate ByteBuf
Motivation:
According to the spec:
All pseudo-header fields MUST appear in the header block before regular
header fields. Any request or response that contains a pseudo-header
field that appears in a header block after
a regular header field MUST be treated as malformed (Section 8.1.2.6).
Pseudo-header fields are only valid in the context in which they are defined.
Pseudo-header fields defined for requests MUST NOT appear in responses;
pseudo-header fields defined for responses MUST NOT appear in requests.
Pseudo-header fields MUST NOT appear in trailers.
Endpoints MUST treat a request or response that contains undefined or
invalid pseudo-header fields as malformed (Section 8.1.2.6).
Clients MUST NOT accept a malformed response. Note that these requirements
are intended to protect against several types of common attacks against HTTP;
they are deliberately strict because being permissive can expose
implementations to these vulnerabilities.
Modifications:
- Introduce validation in HPackDecoder
Result:
- Requests with unknown pseudo-field headers are rejected
- Requests with containing response specific pseudo-headers are rejected
- Requests where pseudo-header appear after regular header are rejected
- h2spec 8.1.2.1 pass
Motivation:
We used Recycler for the CodecOutputList which is not optimized for the use-case of access only from the same Thread all the time.
Modifications:
- Use FastThreadLocal for CodecOutputList
- Add benchmark
Result:
Less overhead in our codecs.
Motivation:
The JMH doc suggests to use BlackHoles to avoid dead code elimination hence would be better to follow this best practice.
Modifications:
Each benchmark method is returning the ByteBuf/ByteBuffer to avoid the JVM to perform any dead code elimination.
Result:
The results are more reliable and comparable to the others provided by other ByteBuf benchmarks (eg HeapByteBufBenchmark)
Motivation:
HttpMethod#valueOf shows up on profiler results in the top set of
results. Since it is a relatively simple operation it can be improved in
isolation.
Modifications:
- Introduce a special case map which assigns each HttpMethod to a unique
index in an array and provides constant time lookup from a hash code
algorithm. When the bucket is matched we can then directly do equality
comparison instead of potentially following a linked structure when
HashMap has hash collisions.
Result:
~10% improvement in benchmark results for HttpMethod#valueOf
Benchmark Mode Cnt Score Error Units
HttpMethodMapBenchmark.newMapKnownMethods thrpt 16 31.831 ± 0.928 ops/us
HttpMethodMapBenchmark.newMapMixMethods thrpt 16 25.568 ± 0.400 ops/us
HttpMethodMapBenchmark.newMapUnknownMethods thrpt 16 51.413 ± 1.824 ops/us
HttpMethodMapBenchmark.oldMapKnownMethods thrpt 16 29.226 ± 0.330 ops/us
HttpMethodMapBenchmark.oldMapMixMethods thrpt 16 21.073 ± 0.247 ops/us
HttpMethodMapBenchmark.oldMapUnknownMethods thrpt 16 49.081 ± 0.577 ops/us
Motivation:
DefaultHttp2FrameWriter#writeData allocates a DataFrameHeader for each write operation. DataFrameHeader maintains internal state and allocates multiple slices of a buffer which is a maximum of 30 bytes. This 30 byte buffer may not always be necessary and the additional slice operations can utilize retainedSlice to take advantage of pooled objects. We can also save computation and object allocations if there is no padding which is a common case in practice.
Modifications:
- Remove DataFrameHeader
- Add a fast path for padding == 0
Result:
Less object allocation in DefaultHttp2FrameWriter
Motivation:
`AbstractScheduledEventExecutor` uses a standard `java.util.PriorityQueue` to keep track of task deadlines. `ScheduledFuture.cancel` removes tasks from this `PriorityQueue`. Unfortunately, `PriorityQueue.remove` has `O(n)` performance since it must search for the item in the entire queue before removing it. This is fast when the future is at the front of the queue (e.g., already triggered) but not when it's randomly located in the queue.
Many servers will use `ScheduledFuture.cancel` on all requests, e.g., to manage a request timeout. As these cancellations will be happen in arbitrary order, when there are many scheduled futures, `PriorityQueue.remove` is a bottleneck and greatly hurts performance with many concurrent requests (>10K).
Modification:
Use netty's `DefaultPriorityQueue` for scheduling futures instead of the JDK. `DefaultPriorityQueue` is almost identical to the JDK version except it is able to remove futures without searching for them in the queue. This means `DefaultPriorityQueue.remove` has `O(log n)` performance.
Result:
Before - cancelling futures has varying performance, capped at `O(n)`
After - cancelling futures has stable performance, capped at `O(log n)`
Benchmark results
After - cancelling in order and in reverse order have similar performance within `O(log n)` bounds
```
Benchmark (num) Mode Cnt Score Error Units
ScheduledFutureTaskBenchmark.cancelInOrder 100 thrpt 20 137779.616 ± 7709.751 ops/s
ScheduledFutureTaskBenchmark.cancelInOrder 1000 thrpt 20 11049.448 ± 385.832 ops/s
ScheduledFutureTaskBenchmark.cancelInOrder 10000 thrpt 20 943.294 ± 12.391 ops/s
ScheduledFutureTaskBenchmark.cancelInOrder 100000 thrpt 20 64.210 ± 1.824 ops/s
ScheduledFutureTaskBenchmark.cancelInReverseOrder 100 thrpt 20 167531.096 ± 9187.865 ops/s
ScheduledFutureTaskBenchmark.cancelInReverseOrder 1000 thrpt 20 33019.786 ± 4737.770 ops/s
ScheduledFutureTaskBenchmark.cancelInReverseOrder 10000 thrpt 20 2976.955 ± 248.555 ops/s
ScheduledFutureTaskBenchmark.cancelInReverseOrder 100000 thrpt 20 362.654 ± 45.716 ops/s
```
Before - cancelling in order and in reverse order have significantly different performance at higher queue size, orders of magnitude worse than the new implementation.
```
Benchmark (num) Mode Cnt Score Error Units
ScheduledFutureTaskBenchmark.cancelInOrder 100 thrpt 20 139968.586 ± 12951.333 ops/s
ScheduledFutureTaskBenchmark.cancelInOrder 1000 thrpt 20 12274.420 ± 337.800 ops/s
ScheduledFutureTaskBenchmark.cancelInOrder 10000 thrpt 20 958.168 ± 15.350 ops/s
ScheduledFutureTaskBenchmark.cancelInOrder 100000 thrpt 20 53.381 ± 13.981 ops/s
ScheduledFutureTaskBenchmark.cancelInReverseOrder 100 thrpt 20 123918.829 ± 3642.517 ops/s
ScheduledFutureTaskBenchmark.cancelInReverseOrder 1000 thrpt 20 5099.810 ± 206.992 ops/s
ScheduledFutureTaskBenchmark.cancelInReverseOrder 10000 thrpt 20 72.335 ± 0.443 ops/s
ScheduledFutureTaskBenchmark.cancelInReverseOrder 100000 thrpt 20 0.743 ± 0.003 ops/s
```
Motivation:
Highly retained and released objects have contention on their ref
count. Currently, the ref count is updated using compareAndSet
with care to make sure the count doesn't overflow, double free, or
revive the object.
Profiling has shown that a non trivial (~1%) of CPU time on gRPC
latency benchmarks is from the ref count updating.
Modification:
Rather than pessimistically assuming the ref count will be invalid,
optimistically update it assuming it will be. If the update was
wrong, then use the slow path to revert the change and throw an
execption. Most of the time, the ref counts are correct.
This changes from using compareAndSet to getAndAdd, which emits a
different CPU instruction on x86 (CMPXCHG to XADD). Because the
CPU knows it will modifiy the memory, it can avoid contention.
On a highly contended machine, this can be about 2x faster.
There is a downside to the new approach. The ref counters can
temporarily enter invalid states if over retained or over released.
The code does handle these overflow and underflow scenarios, but it
is possible that another concurrent access may push the failure to
a different location. For example:
Time 1 Thread 1: obj.retain(INT_MAX - 1)
Time 2 Thread 1: obj.retain(2)
Time 2 Thread 2: obj.retain(1)
Previously Thread 2 would always succeed and Thread 1 would always
fail on the second access. Now, thread 2 could fail while thread 1
is rolling back its change.
====
There are a few reasons why I think this is okay:
1. Buggy code is going to have bugs. An exception _is_ going to be
thrown. This just causes the other threads to notice the state
is messed up and stop early.
2. If high retention counts are a use case, then ref count should
be a long rather than an int.
3. The critical section is greatly reduced compared to the previous
version, so the likelihood of this happening is lower
4. On error, the code always rollsback the change atomically, so
there is no possibility of corruption.
Result:
Faster refcounting
```
BEFORE:
Benchmark (delay) Mode Cnt Score Error Units
AbstractReferenceCountedByteBufBenchmark.retainRelease_contended 1 sample 2901361 804.579 ± 1.835 ns/op
AbstractReferenceCountedByteBufBenchmark.retainRelease_contended 10 sample 3038729 785.376 ± 16.471 ns/op
AbstractReferenceCountedByteBufBenchmark.retainRelease_contended 100 sample 2899401 817.392 ± 6.668 ns/op
AbstractReferenceCountedByteBufBenchmark.retainRelease_contended 1000 sample 3650566 2077.700 ± 0.600 ns/op
AbstractReferenceCountedByteBufBenchmark.retainRelease_contended 10000 sample 3005467 19949.334 ± 4.243 ns/op
AbstractReferenceCountedByteBufBenchmark.retainRelease_uncontended 1 sample 456091 48.610 ± 1.162 ns/op
AbstractReferenceCountedByteBufBenchmark.retainRelease_uncontended 10 sample 732051 62.599 ± 0.815 ns/op
AbstractReferenceCountedByteBufBenchmark.retainRelease_uncontended 100 sample 778925 228.629 ± 1.205 ns/op
AbstractReferenceCountedByteBufBenchmark.retainRelease_uncontended 1000 sample 633682 2002.987 ± 2.856 ns/op
AbstractReferenceCountedByteBufBenchmark.retainRelease_uncontended 10000 sample 506442 19735.345 ± 12.312 ns/op
AFTER:
Benchmark (delay) Mode Cnt Score Error Units
AbstractReferenceCountedByteBufBenchmark.retainRelease_contended 1 sample 3761980 383.436 ± 1.315 ns/op
AbstractReferenceCountedByteBufBenchmark.retainRelease_contended 10 sample 3667304 474.429 ± 1.101 ns/op
AbstractReferenceCountedByteBufBenchmark.retainRelease_contended 100 sample 3039374 479.267 ± 0.435 ns/op
AbstractReferenceCountedByteBufBenchmark.retainRelease_contended 1000 sample 3709210 2044.603 ± 0.989 ns/op
AbstractReferenceCountedByteBufBenchmark.retainRelease_contended 10000 sample 3011591 19904.227 ± 18.025 ns/op
AbstractReferenceCountedByteBufBenchmark.retainRelease_uncontended 1 sample 494975 52.269 ± 8.345 ns/op
AbstractReferenceCountedByteBufBenchmark.retainRelease_uncontended 10 sample 771094 62.290 ± 0.795 ns/op
AbstractReferenceCountedByteBufBenchmark.retainRelease_uncontended 100 sample 763230 235.044 ± 1.552 ns/op
AbstractReferenceCountedByteBufBenchmark.retainRelease_uncontended 1000 sample 634037 2006.578 ± 3.574 ns/op
AbstractReferenceCountedByteBufBenchmark.retainRelease_uncontended 10000 sample 506284 19742.605 ± 13.729 ns/op
```
Motiviation:
The ResourceLeakDetector helps to detect and troubleshoot resource leaks and is often used even in production enviroments with a low level. Because of this its import that we try to keep the overhead as low as overhead. Most of the times no leak is detected (as all is correctly handled) so we should keep the overhead for this case as low as possible.
Modifications:
- Only call getStackTrace() if a leak is reported as it is a very expensive native call. Also handle the filtering and creating of the String in a lazy fashion
- Remove the need to mantain a Queue to store the last access records
- Add benchmark
Result:
Huge decrease of performance overhead.
Before the patch:
Benchmark (recordTimes) Mode Cnt Score Error Units
ResourceLeakDetectorRecordBenchmark.record 8 thrpt 20 4358.367 ± 116.419 ops/s
ResourceLeakDetectorRecordBenchmark.record 16 thrpt 20 2306.027 ± 55.044 ops/s
ResourceLeakDetectorRecordBenchmark.recordWithHint 8 thrpt 20 4220.979 ± 114.046 ops/s
ResourceLeakDetectorRecordBenchmark.recordWithHint 16 thrpt 20 2250.734 ± 55.352 ops/s
With this patch:
Benchmark (recordTimes) Mode Cnt Score Error Units
ResourceLeakDetectorRecordBenchmark.record 8 thrpt 20 71398.957 ± 2695.925 ops/s
ResourceLeakDetectorRecordBenchmark.record 16 thrpt 20 38643.963 ± 1446.694 ops/s
ResourceLeakDetectorRecordBenchmark.recordWithHint 8 thrpt 20 71677.882 ± 2923.622 ops/s
ResourceLeakDetectorRecordBenchmark.recordWithHint 16 thrpt 20 38660.176 ± 1467.732 ops/s
Motivation:
1. Some encoders used a `ByteBuf#writeBytes` to write short constant byte array (2-3 bytes). This can be replaced with more faster `ByteBuf#writeShort` or `ByteBuf#writeMedium` which do not access the memory.
2. Two chained calls of the `ByteBuf#setByte` with constants can be replaced with one `ByteBuf#setShort` to reduce index checks.
3. The signature of method `HttpHeadersEncoder#encoderHeader` has an unnecessary `throws`.
Modifications:
1. Use `ByteBuf#writeShort` or `ByteBuf#writeMedium` instead of `ByteBuf#writeBytes` for the constants.
2. Use `ByteBuf#setShort` instead of chained call of the `ByteBuf#setByte` with constants.
3. Remove an unnecessary `throws` from `HttpHeadersEncoder#encoderHeader`.
Result:
A bit faster writes constants into buffers.
Motivation:
Right now HttpRequestEncoder does insertion of slash for url like http://localhost?pararm=1 before the question mark. It is done not effectively.
Modification:
Code:
new StringBuilder(len + 1)
.append(uri, 0, index)
.append(SLASH)
.append(uri, index, len)
.toString();
Replaced with:
new StringBuilder(uri)
.insert(index, SLASH)
.toString();
Result:
Faster HttpRequestEncoder. Additional small test. Attached benchmark in PR.
Benchmark Mode Cnt Score Error Units
HttpRequestEncoderInsertBenchmark.newEncoder thrpt 40 3704843.303 ± 98950.919 ops/s
HttpRequestEncoderInsertBenchmark.oldEncoder thrpt 40 3284236.960 ± 134433.217 ops/s
Motivation:
A `StringUtil#escapeCsv` creates new `StringBuilder` on each value even if the same string is returned in the end.
Modifications:
Create new `StringBuilder` only if it really needed. Otherwise, return the original string (or just trimmed substring).
Result:
Less GC load. Up to 4x faster work for not changed strings.
Motivation:
When I run Netty micro benchmarks I get many warnings like:
WARNING: -Dio.netty.noResourceLeakDetection is deprecated. Use '-Dio.netty.leakDetection.level=simple' instead.
Modification:
-Dio.netty.noResourceLeakDetection replaced with -Dio.netty.leakDetection.level=disabled.
Result:
No warnings.
Motivation:
IPv4/6 validation methods use allocations, which can be avoided.
IPv4 parse method use StringTokenizer.
Modifications:
Rewriting IPv4/6 validation methods to avoid allocations.
Rewriting IPv4 parse method without use StringTokenizer.
Result:
IPv4/6 validation and IPv4 parsing faster up to 2-10x.
Motivation:
We currently don't have a native transport which supports kqueue https://www.freebsd.org/cgi/man.cgi?query=kqueue&sektion=2. This can be useful for BSD systems such as MacOS to take advantage of native features, and provide feature parity with the Linux native transport.
Modifications:
- Make a new transport-native-unix-common module with all the java classes and JNI code for generic unix items. This module will build a static library for each unix platform, and included in the dynamic libraries used for JNI (e.g. transport-native-epoll, and eventually kqueue).
- Make a new transport-native-unix-common-tests module where the tests for the transport-native-unix-common module will live. This is so each unix platform can inherit from these test and ensure they pass.
- Add a new transport-native-kqueue module which uses JNI to directly interact with kqueue
Result:
JNI support for kqueue.
Fixes https://github.com/netty/netty/issues/2448
Fixes https://github.com/netty/netty/issues/4231
Motivation:
Java9 added a new method to Unsafe which allows to allocate a byte[] without memset it. This can have a massive impact in allocation times when the byte[] is big. This change allows to enable this when using Java9 with the io.netty.tryAllocateUninitializedArray property when running Java9+. Please note that you will need to open up the jdk.internal.misc package via '--add-opens java.base/jdk.internal.misc=ALL-UNNAMED' as well.
Modifications:
Allow to allocate byte[] without memset on Java9+
Result:
Better performance when allocate big heap buffers and using java9.
Motivation:
There are two files that still use `system.out.println` to log their status
Modification:
Replace `system.out.println` with a `debug` function inside an instance of `InternalLoggerFactory`
Result:
Introduce an instance of `InternalLoggerFactory` in class `AbstractMicrobenchmark.java` and `AbstractSharedExecutorMicrobenchmark.java`
