Motivation:
sun.misc.Unsafe allows us to handle heap ByteBuf in a more efficient matter. We should use special ByteBuf implementation when sun.misc.Unsafe can be used to increase performance.
Modifications:
- Add PooledUnsafeHeapByteBuf and UnpooledUnsafeHeapByteBuf that are used when sun.misc.Unsafe is ready to use.
- Add UnsafeHeapSwappedByteBuf
Result:
Better performance when using heap buffers and sun.misc.Unsafe is ready to use.
Motivation:
We had a bug in our implemention which double "reversed" bytes on systems which not support unaligned access.
Modifications:
- Correctly only reverse bytes if needed.
- Share code between unsafe implementations.
Result:
No more data-corruption on sytems without unaligned access.
Motivation:
We started the thread before store it in a field which could lead to an assert error when the thread is executed before we actually store it.
Modifications:
Store thread before start it.
Result:
No more assert error possible.
Motivation:
At the moment we only forward decoded messages that were added the out List once the full decode loop was completed. This has the affect that resources may not be released as fast as possible and as an application may incounter higher latency if the user triggeres a writeAndFlush(...) as a result of the decoded messages.
Modifications:
- forward decoded messages after each decode call
Result:
Forwarding decoded messages through the pipeline in a more eager fashion.
Motivation:
When dealing with case insensitive headers it can be useful to have a case insensitive contains method for CharSequence.
Modifications:
- Add containsCaseInsensative to AsciiString
Result:
More expressive utility method for case insensitive CharSequence.
Motivation:
Http2CodecUtils has some static variables which are defined as Strings instead of CharSequence. One of these defines is used as a header name and should be AsciiString.
Modifications:
- Change the String defines in Http2CodecUtils to CharSequence
Result:
Types are more consistently using CharSequence and adding the upgrade header will require less work.
Motivation:
Leak detector, when it detects a leak, will print the last 5 stack
traces that touched the ByteBuf. In some cases that might not be enough
to identify the root cause of the leak.
Also, sometimes users might not be interested in tracing all the
operations on the buffer, but just the ones that are affecting the
reference count.
Modifications:
Added command line properties to override default values:
* Allow to configure max number of stack traces to collect
* Allow to only record retain/release operation on buffers
Result:
Users can increase the number of stack traces to debug buffer leaks
with lot of retain/release operations.
Motivation:
for debugging and metrics reasons its sometimes useful to be able to get details of the the Thread that powers a SingleThreadEventExecutor.
Modifications:
- Expose ThreadProperties
- Add unit test.
Result:
It's now possible to get details of the Thread that powers a SingleThreadEventExecutor.
Motivation:
Sometimes it is useful to disable recycling completely if memory constraints are very tight.
Modifications:
Allow to use -Dio.netty.recycler.maxCapacity=0 to disable recycling completely.
Result:
It's possible to disable recycling now.
Motivation:
The javadocs are incorrect and so give false impressions of use-pattern.
Modifications:
- Fix javadocs of which operations are allowed from multiple threads concurrently.
- Let isEmpty() work concurrently.
Result:
Correctly document usage-patterns.
Motivation:
The StringUtil class creates a Formatter object, but does not close it. There are also a 2 utility methods which would be generally useful.
Modifications:
- Close the Formatter
- Add length and isNullOrEmpty
Result:
No more resource leaks. Additional utility methods.
Motivation:
A degradation in performance has been observed from the 4.0 branch as documented in https://github.com/netty/netty/issues/3962.
Modifications:
- Simplify Headers class hierarchy.
- Restore the DefaultHeaders to be based upon DefaultHttpHeaders from 4.0.
- Make various other modifications that are causing hot spots.
Result:
Performance is now on par with 4.0.
Motivation:
We noticed that the headers implementation in Netty for HTTP/2 uses quite a lot of memory
and that also at least the performance of randomly accessing a header is quite poor. The main
concern however was memory usage, as profiling has shown that a DefaultHttp2Headers
not only use a lot of memory it also wastes a lot due to the underlying hashmaps having
to be resized potentially several times as new headers are being inserted.
This is tracked as issue #3600.
Modifications:
We redesigned the DefaultHeaders to simply take a Map object in its constructor and
reimplemented the class using only the Map primitives. That way the implementation
is very concise and hopefully easy to understand and it allows each concrete headers
implementation to provide its own map or to even use a different headers implementation
for processing requests and writing responses i.e. incoming headers need to provide
fast random access while outgoing headers need fast insertion and fast iteration. The
new implementation can support this with hardly any code changes. It also comes
with the advantage that if the Netty project decides to add a third party collections library
as a dependency, one can simply plug in one of those very fast and memory efficient map
implementations and get faster and smaller headers for free.
For now, we are using the JDK's TreeMap for HTTP and HTTP/2 default headers.
Result:
- Significantly fewer lines of code in the implementation. While the total commit is still
roughly 400 lines less, the actual implementation is a lot less. I just added some more
tests and microbenchmarks.
- Overall performance is up. The current implementation should be significantly faster
for insertion and retrieval. However, it is slower when it comes to iteration. There is simply
no way a TreeMap can have the same iteration performance as a linked list (as used in the
current headers implementation). That's totally fine though, because when looking at the
benchmark results @ejona86 pointed out that the performance of the headers is completely
dominated by insertion, that is insertion is so significantly faster in the new implementation
that it does make up for several times the iteration speed. You can't iterate what you haven't
inserted. I am demonstrating that in this spreadsheet [1]. (Actually, iteration performance is
only down for HTTP, it's significantly improved for HTTP/2).
- Memory is down. The implementation with TreeMap uses on avg ~30% less memory. It also does not
produce any garbage while being resized. In load tests for GRPC we have seen a memory reduction
of up to 1.2KB per RPC. I summarized the memory improvements in this spreadsheet [1]. The data
was generated by [2] using JOL.
- While it was my original intend to only improve the memory usage for HTTP/2, it should be similarly
improved for HTTP, SPDY and STOMP as they all share a common implementation.
[1] https://docs.google.com/spreadsheets/d/1ck3RQklyzEcCLlyJoqDXPCWRGVUuS-ArZf0etSXLVDQ/edit#gid=0
[2] https://gist.github.com/buchgr/4458a8bdb51dd58c82b4
Motivation:
The HttpObjectDecoder is on the hot code path for the http codec. There are a few hot methods which can be modified to improve performance.
Modifications:
- Modify AppendableCharSequence to provide unsafe methods which don't need to re-check bounds for every call.
- Update HttpObjectDecoder methods to take advantage of new AppendableCharSequence methods.
Result:
Peformance boost for decoding http objects.
Motivation:
We should support XXXCollections methods for all primitive map types.
Modifications:
Removed PrimitiveCollections and added a template for XXXCollections.
Result:
Fixes#4001
Motivation:
It would be useful to support the Java `Map` interface in our primitive maps.
Modifications:
Renamed current methods to "pXXX", where p is short for "primitive". Made the template for all primitive maps extend the appropriate Map interface.
Result:
Fixes#3970
Motivation:
Prior we used a purge task that would remove previous canceled scheduled tasks from the internal queue. This could introduce some delay and so use a lot of memory even if the task itself is already canceled.
Modifications:
Schedule removal of task from queue via EventLoop if cancel operation is not done in the EventLoop Thread or just remove directly if the Thread that cancels the scheduled task is in the EventLoop.
Result:
Faster possibility to GC a canceled ScheduledFutureTask.
Motivation:
PoolThreadCache did only cache allocations if the allocation and deallocation Thread were the same. This is not optimal as often people write from differen thread then the actual EventLoop thread.
Modification:
- Add MpscArrayQueue which was forked from jctools and lightly modified.
- Use MpscArrayQueue for caches and always add buffer back to the cache that belongs to the allocation thread.
Result:
ThreadPoolCache is now also usable and so gives performance improvements when allocation and deallocation thread are different.
Performance when using same thread for allocation and deallocation is noticable worse then before.
Motivation:
The PooledByteBufAllocator is more or less a black-box atm. We need to expose some metrics to allow the user to get a better idea how to tune it.
Modifications:
- Expose different metrics via PooledByteBufAllocator
- Add *Metrics interfaces
Result:
It is now easy to gather metrics and detail about the PooledByteBufAllocator and so get a better understanding about resource-usage etc.
Motivation:
There is an error in the ByteString test logic which is resulting in test failures.
Modifications:
- Fix the loop iteration to use the loop iteration variable instead of a fixed index.
Result:
Tests are less buggy.
Motivation:
In the SslHandler we schedule a timeout at which we close the Channel if a timeout was detected during close_notify. Because this can race with notify the flushFuture we can see an IllegalStateException when the Channel is closed.
Modifications:
- Use a trySuccess() and tryFailure(...) to guard against race.
Result:
No more race.
Motivation:
All read operations should be safe to execute from multiple threads which was not the case and so could produce a livelock.
Modifications:
Modify methods so these are safe to be called from multiple threads.
Result:
No more livelock.
Motivation:
There are various known issues in netty-codec-dns:
- Message types are not interfaces, which can make it difficult for a
user to implement his/her own message implementation.
- Some class names and field names do not match with the terms in the
RFC.
- The support for decoding a DNS record was limited. A user had to
encode and decode by him/herself.
- The separation of DnsHeader from DnsMessage was unnecessary, although
it is fine conceptually.
- Buffer leak caused by DnsMessage was difficult to analyze, because the
leak detector tracks down the underlying ByteBuf rather than the
DnsMessage itself.
- DnsMessage assumes DNS-over-UDP.
- To send an EDNS message, a user have to create a new DNS record class
instance unnecessarily.
Modifications:
- Make all message types interfaces and add default implementations
- Rename some classes, properties, and constants to match the RFCs
- DnsResource -> DnsRecord
- DnsType -> DnsRecordType
- and many more
- Remove DnsClass and use an integer to support EDNS better
- Add DnsRecordEncoder/DnsRecordDecoder and their default
implementations
- DnsRecord does not require RDATA to be ByteBuf anymore.
- Add DnsRawRecord as the catch-all record type
- Merge DnsHeader into DnsMessage
- Make ResourceLeakDetector track AbstractDnsMessage
- Remove DnsMessage.sender/recipient properties
- Wrap DnsMessage with AddressedEnvelope
- Add DatagramDnsQuest and DatagramDnsResponse for ease of use
- Rename DnsQueryEncoder to DatagramDnsQueryEncoder
- Rename DnsResponseDecoder to DatagramDnsResponseDecoder
- Miscellaneous changes
- Add StringUtil.TAB
Result:
- Cleaner APi
- Can support DNS-over-TCP more easily in the future
- Reduced memory footprint in the default DnsQuery/Response
implementations
- Better leak tracking for DnsMessages
- Possibility to introduce new DnsRecord types in the future and provide
full record encoder/decoder implementation.
- No unnecessary instantiation for an EDNS pseudo resource record
Motivation:
Many projects need some kind a Channel/Connection pool implementation. While the protocols are different many things can be shared, so we should provide a generic API and implementation.
Modifications:
Add ChannelPool / ChannelPoolMap API and implementations.
Result:
Reusable / Generic pool implementation that users can use.
Motivation:
'length2 == 0' is not reachable because length1 and length2 are same at this point.
Motification:
Removed 'length2 == 0'.
Result:
Cleaner code.
Motivation:
Currently, valueOf() and newInstance() use almost same code to create new constant.
For maintainability, it's better to share duplicate code among them.
Motification:
Added new private functions.
- checkNotNullAndNotEmpty() is for checking whether the name of a constant is null and empty.
- newConstant0() is for creating a new constant.
Result:
- Compact source code
- Improvement of maintainability
Motivation:
When a SecurityManager is in place that preven reading the somaxconn file trying to bootstrap a channel later will result in a ClassNotFoundError.
Modifications:
- Reading the file in a privileged block.
Result:
No more ClassNotFoundError when a SecurityManager is in place.
Motivation:
The ByteString class currently assumes the underlying array will be a complete representation of data. This is limiting as it does not allow a subsection of another array to be used. The forces copy operations to take place to compensate for the lack of API support.
Modifications:
- add arrayOffset method to ByteString
- modify all ByteString and AsciiString methods that loop over or index into the underlying array to use this offset
- update all code that uses ByteString.array to ensure it accounts for the offset
- add unit tests to test the implementation respects the offset
Result:
ByteString and AsciiString can represent a sub region of a byte[].
Motivation:
static Package getPackage(Class<?> c) uses synchronized block internally.
Thanks to @jingene for the hint and initial report of the issue.
Modifications:
-Use simple lastIndexOf(...) and substring for a faster implementation
Result:
No more lock condition.
Motivation:
Each different *ChannelOption did extend ChannelOption in 4.0, which we changed in 4.1. This is a breaking change in terms of the API so we need to ensure we keep the old hierarchy.
Modifications:
- Let all *ChannelOption extend ChannelOption
- Add back constructor and mark it as @deprecated
Result:
No API breakage between 4.0 and 4.1
Motivation:
The current implementation does byte by byte comparison, which we have seen
can be a performance bottleneck when the AsciiString is used as the key in
a Map.
Modifications:
Use sun.misc.Unsafe (on supporting platforms) to compare up to eight bytes at a time
and get closer to the performance of String.equals(Object).
Result:
Significant improvement (2x - 6x) in performance over the current implementation.
Benchmark (size) Mode Samples Score Score error Units
i.n.m.i.PlatformDependentBenchmark.arraysBytesEqual 10 thrpt 10 118843477.518 2347259.347 ops/s
i.n.m.i.PlatformDependentBenchmark.arraysBytesEqual 50 thrpt 10 43910319.773 198376.996 ops/s
i.n.m.i.PlatformDependentBenchmark.arraysBytesEqual 100 thrpt 10 26339969.001 159599.252 ops/s
i.n.m.i.PlatformDependentBenchmark.arraysBytesEqual 1000 thrpt 10 2873119.030 20779.056 ops/s
i.n.m.i.PlatformDependentBenchmark.arraysBytesEqual 10000 thrpt 10 306370.450 1933.303 ops/s
i.n.m.i.PlatformDependentBenchmark.arraysBytesEqual 100000 thrpt 10 25750.415 108.391 ops/s
i.n.m.i.PlatformDependentBenchmark.unsafeBytesEqual 10 thrpt 10 248077563.510 635320.093 ops/s
i.n.m.i.PlatformDependentBenchmark.unsafeBytesEqual 50 thrpt 10 128198943.138 614827.548 ops/s
i.n.m.i.PlatformDependentBenchmark.unsafeBytesEqual 100 thrpt 10 86195621.349 1063959.307 ops/s
i.n.m.i.PlatformDependentBenchmark.unsafeBytesEqual 1000 thrpt 10 16920264.598 61615.365 ops/s
i.n.m.i.PlatformDependentBenchmark.unsafeBytesEqual 10000 thrpt 10 1687454.747 6367.602 ops/s
i.n.m.i.PlatformDependentBenchmark.unsafeBytesEqual 100000 thrpt 10 153717.851 586.916 ops/s
Motivation:
ByteString#hashCode() trashes its own hash code if it's being accessed concurrently
Modifications:
Pull the ByteString#hash into a local variable and calculate it locally.
Result:
ByteString#hashCode() is no longer returning a junk value.
Motivation:
The IntObjectHashMap benchmarks show the Agrona collections to be faster on put, lookup, and remove. One major difference is that we're using 2 modulus operations each time we increment the position index while iterating. Agrona uses a mask instead.
Modifications:
Modified the KObjectHashMap to use masking rather than modulus when wrapping the position index. This requires that the capacity be a power of 2.
Result:
Improved performance of IntObjectHashMap.
Motivation:
The usage and code within AsciiString has exceeded the original design scope for this class. Its usage as a binary string is confusing and on the verge of violating interface assumptions in some spots.
Modifications:
- ByteString will be created as a base class to AsciiString. All of the generic byte handling processing will live in ByteString and all the special character encoding will live in AsciiString.
Results:
The AsciiString interface will be clarified. Users of AsciiString can now be clear of the limitations the class imposes while users of the ByteString class don't have to live with those limitations.
Motivation:
Attribute.getAndRemove() will return the value but also remove the AttributeKey itself from the AttributeMap. This may not
what you want as you may want to keep an instance of it and just set it later again. Document the contract so the user know what to expect.
Modifications:
- Make it clear when to use AttributeKey.getAndRemove() / AttributeKey.remove() and when AttributeKey.getAndSet(null) / AttributeKey.set(null).
Result:
Less suprising behaviour.
Motivation:
To support HTTP2 we need APLN support. This was not provided before when using OpenSslEngine, so SSLEngine (JDK one) was the only bet.
Beside this CipherSuiteFilter was not supported
Modifications:
- Upgrade netty-tcnative and make use of new features to support ALPN and NPN in server and client mode.
- Guard against segfaults after the ssl pointer is freed
- support correctly different failure behaviours
- add support for CipherSuiteFilter
Result:
Be able to use OpenSslEngine for ALPN / NPN for server and client.
