Motivation:
93130b172a introduced a regression where we not "converted" an empty HttpContent to ByteBuf and just passed it on in the pipeline. This can lead to the situation that other handlers in the pipeline will see HttpContent instances which is not expected.
Modifications:
- Correctly convert HttpContent to ByteBuf when empty
- Add unit test.
Result:
Handlers in the pipeline will see the expected message type.
Motivation:
When looking for a leak, its nice to be able to request at least a
number of leaks.
Modification:
* Made all leak records up to the target amoutn recorded, and only
then enable backing off.
* Enable recording more than 32 elements. Previously the shift
amount made this impossible.
Result:
Ability to record all accesses.
Motivation:
AbstractCoalescingBufferQueue#add accounts for void promises, but AbstractCoalescingBufferQueue#addFirst does not. These methods should be consistent.
Modifications:
- AbstractCoalescingBufferQueue#addFirst should account for void promises and share code with AbstractCoalescingBufferQueue#add
Result:
More correct void promise handling in AbstractCoalescingBufferQueue.
complete
Motivation:
SslHandler removes a Buffer/Promise pair from
AbstractCoalescingBufferQueue when wrapping data. However it is possible
the SSLEngine will not consume the entire buffer. In this case
SslHandler adds the Buffer back to the queue, but doesn't add the
Promise back to the queue. This may result in the promise completing
immediately in finishFlush, and generally not correlating to the
completion of writing the corresponding Buffer
Modifications:
- AbstractCoalescingBufferQueue#addFirst should also support adding the
ChannelPromise
- In the event of a handshake timeout we should immediately fail pending
writes immediately to get a more accurate exception
Result:
Fixes https://github.com/netty/netty/issues/7378.
Motivation:
For use cases that create headers, but do not need to modify them a read only variant of HttpHeaders would be useful and may be able to provide better iteration performance for encoding.
Modifications:
- Introduce ReadOnlyHttpHeaders that is backed by a flat array
Result:
ReadOnlyHttpHeaders exists for non-modifiable HttpHeaders use cases.
Motivation:
We need to set readPending to false when we detect a EOF while issue a read as otherwise we may not unregister from the Selector / Epoll / KQueue and so keep on receving wakeups.
The important bit is that we may even get a wakeup for a read event but will still will only be able to read 0 bytes from the socket, so we need to be very careful when we clear the readPending. This can happen because we generally using edge-triggered mode for our native transports and because of the nature of edge-triggered we may schedule an read event just to find out there is nothing left to read atm (because we completely drained the socket on the previous read).
Modifications:
Set readPending to false when EOF is detected.
Result:
Fixes [#7255].
Motivation:
SslHandler only supports ByteBuf objects, but will not release objects of other types. SslHandler will also not release objects if its internal state is not correctly setup.
Modifications:
- Release non-ByteBuf objects in write
- Release all objects if the SslHandler queue is not setup
Result:
Less leaks in SslHandler.
Motivation:
HTTP/2 allows writes of 0 length data frames. However in some cases EMPTY_BUFFER is used instead of the actual buffer that was written. This may mask writes of released buffers or otherwise invalid buffer objects. It is also possible that if the buffer is invalid AbstractCoalescingBufferQueue will not release the aggregated buffer nor fail the associated promise.
Modifications:
- DefaultHttp2FrameCodec should take care to fail the promise, even if releasing the data throws
- AbstractCoalescingBufferQueue should release any aggregated data and fail the associated promise if something goes wrong during aggregation
Result:
More correct handling of invalid buffers in HTTP/2 code.
This reverts commit 413c7c2cd8 as it introduced an regression when edge-triggered mode is used which is true for our native transports by default. With 413c7c2cd8 included it was possible that we set readPending to false by mistake even if we would be interested in read more.
Motivation:
7995afee8f introduced a change that broke special handling of WebSockets 00.
Modifications:
Correctly delegate to super method which has special handling for WebSockets 00.
Result:
Fixes [#7362].
Motivation:
HttpObjectEncoder and MessageAggregator treat buffers that are not readable special. If a buffer is not readable, then an EMPTY_BUFFER is written and the actual buffer is ignored. If the buffer has already been released then this will not be correct as the promise will be completed, but in reality the original content shouldn't have resulted in any write because it was invalid.
Modifications:
- HttpObjectEncoder should retain/write the original buffer instead of using EMPTY_BUFFER
- MessageAggregator should retain/write the original ByteBufHolder instead of using EMPTY_BUFFER
Result:
Invalid write operations which happen to not be readable correctly reflect failed status in the promise, and do not result in any writes to the channel.
infinite loop
Motivation:
If SslHandler sets jdkCompatibilityMode to false and ReferenceCountedOpenSslEngine sets jdkCompatibilityMode to true there is a chance we will get stuck in an infinite loop if the peer sends a TLS packet with length greater than 2^14 (the maximum length allowed in the TLS 1.2 RFC [1]). However there are legacy implementations which actually send larger TLS payloads than 2^14 (e.g. OpenJDK's SSLSessionImpl [2]) and in this case ReferenceCountedOpenSslEngine will return BUFFER_OVERFLOW in an attempt to notify that a larger buffer is to be used, but if the buffer is already at max size this process will repeat indefinitely.
[1] https://tools.ietf.org/html/rfc5246#section-6.2.1
[2] http://hg.openjdk.java.net/jdk8u/jdk8u/jdk/file/d5a00b1e8f78/src/share/classes/sun/security/ssl/SSLSessionImpl.java#l793
Modifications:
- Support TLS payload sizes greater than 2^14 in ReferenceCountedOpenSslEngine
- ReferenceCountedOpenSslEngine should throw an exception if a
BUFFER_OVERFLOW is impossible to rectify
Result:
No more infinite loop in ReferenceCountedOpenSslEngine due to
BUFFER_OVERFLOW and large TLS payload lengths.
Motivation:
ReferenceCountedOpenSslEngine.rejectRemoteInitiatedRenegotiation() is called in a finally block to ensure we always check for renegotiation. The problem here is that sometimes we will already shutdown the engine before we call the method which will lead to an NPE in this case as the ssl pointer was already destroyed.
Modifications:
Check that the engine is not destroyed yet before calling SSL.getHandshakeCount(...)
Result:
Fixes [#7353].
Motivation:
Some SSLEngine implementations (e.g. ReferenceCountedOpenSslContext) support unwrapping/wrapping multiple packets at a time. The SslHandler behaves differently if the SSLEngine supports this feature, but currently requires that the constructor argument between the SSLEngine creation and SslHandler are coordinated. This can be difficult, or require package private access, if extending the SslHandler.
Modifications:
- The SslHandler should inspect the SSLEngine to see if it supports jdkCompatibilityMode instead of relying on getting an extra constructor argument which maybe out of synch with the SSLEngine
Result:
Easier to override SslHandler and have consistent jdkCompatibilityMode between SSLEngine and SslHandler.
Motivation:
We should ensure we only try to load the netty-tcnative version that was compiled for the architecture we are using.
Modifications:
Include architecture into native lib name.
Result:
Only load native lib if the architecture is supported.
Motivation:
Linux kernel 4.11 introduced a new socket option,
TCP_FASTOPEN_CONNECT, that greatly simplifies making TCP Fast Open
connections on client side. Usually simply setting the flag before
connect() call is enough, no more changes are required.
Details can be found in kernel commit:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=19f6d3f3
Modifications:
TCP_FASTOPEN_CONNECT socket option was added to EpollChannelOption
class.
Result:
Netty clients can easily make TCP Fast Open connections. Simply
calling option(EpollChannelOption.TCP_FASTOPEN_CONNECT, true) in
client bootstrap is enough (given recent enough kernel).
This change allows to upgrade a plain HTTP 1.x connection to TLS
according to RFC 2817. Switching the transport layer to TLS should be
possible without removing HttpClientCodec from the pipeline,
because HTTP/1.x layer of the protocol remains untouched by the switch
and the HttpClientCodec state must be retained for proper
handling the remainder of the response message,
per RFC 2817 requirement in point 3.3:
Once the TLS handshake completes successfully, the server MUST
continue with the response to the original request.
After this commit, the upgrade can be established by simply
inserting an SslHandler at the front of the pipeline after receiving
101 SWITCHING PROTOCOLS response, exactly as described in SslHander
documentation.
Modifications:
- Don't set HttpObjectDecoder into UPGRADED state if
101 SWITCHING_PROTOCOLS response contains HTTP/1.0 or HTTP/1.1 in
the protocol stack described by the Upgrade header.
- Skip pairing comparison for 101 SWITCHING_PROTOCOLS, similar
to 100 CONTINUE, since 101 is not the final response to the original
request and the final response is expected after TLS handshake.
Fixes#7293.
Motivation:
Fix NullPointerExceptions that occur when running netty-tcnative inside the bootstrap class loader.
Modifications:
- Replace loader.getResource(...) with ClassLoader.getSystemResource(...) when loader is null.
- Replace loader.loadClass(...) with Class.forName(..., false, loader) which works when loader is both null and non-null.
Result:
Support running native libs in bootstrap class loader
Motivation:
If a child channel's read is triggered outside the parent channel's read
loop then it is possible a WINDOW_UPDATE will be written, but not
flushed.
If a child channel's beginRead processes data from the inboundBuffer and
then readPending is set to false, which will result in data not being
delivered if in the parent's read loop and more data is attempted to be
delievered to that child channel.
Modifications:
- The child channel must force a flush if a frame is written as a result
of reading a frame, and this is not in the parent channel's read loop
- The child channel must allow a transition from dequeueing from
beginRead into the parent channel's read loop to deliver more data
Result:
The child channel flushes data when reading outside the parent's read
loop, and has frames delivered more reliably.
Motivation:
readPending is currently only set to false if data is delivered to the application, however this may result in duplicate events being received from the selector in the event that the socket was closed.
Modifications:
- We should set readPending to false before each read attempt for all
transports besides NIO.
- Based upon the Javadocs it is possible that NIO may have spurious
wakeups [1]. In this case we should be more cautious and only set
readPending to false if data was actually read.
[1] https://docs.oracle.com/javase/7/docs/api/java/nio/channels/SelectionKey.html
That a selection key's ready set indicates that its channel is ready for some operation category is a hint, but not a guarantee, that an operation in such a category may be performed by a thread without causing the thread to block.
Result:
Notification from the selector (or simulated events from kqueue/epoll ET) in the event of socket closure.
Fixes https://github.com/netty/netty/issues/7255
Motivation:
Http2StreamFrameToHttpObjectCodec was not properly encoding and
decoding 100-Continue HttpResponse/Http2SettingsFrame properly. It was
encoding 100-Continue FullHttpResponse as an Http2SettingFrame with
endStream=true, causing the child channel to terminate. It was not
decoding 100-Continue Http2SettingsFrame (endStream=false) as
FullHttpResponse. This should be fixed as it would cause http2 child
stream to prematurely close, and could cause HttpObjectAggregator to
fail if it's in the pipeline.
Modification:
- Fixed encode() to properly encode 100-Continue FullHttpResponse as
Http2SettingsFrame with endStream=false
- Reject 100-Continue HttpResponse that are NOT FullHttpResponse
- Fixed decode() to properly decode 100-Continue Http2SettingsFrame
(endStream=false) as a FullHttpResponse
- made Http2StreamFrameToHttpObjectCodec sharable so that it can b used
among child streams within the same Http2MultiplexCodec
Result:
Now Http2StreamFrameToHttpObjectCodec should be properly handling
100-Continue responses.
Motivation:
I am receiving a multipart/form_data upload from a Mailgun webhook. This webhook used to send parts like this:
--74e78d11b0214bdcbc2f86491eeb4902
Content-Disposition: form-data; name="attachment-2"; filename="attached_�айл.txt"
Content-Type: text/plain
Content-Length: 32
This is the content of the file
--74e78d11b0214bdcbc2f86491eeb4902--
but now it posts parts like this:
--74e78d11b0214bdcbc2f86491eeb4902
Content-Disposition: form-data; name="attachment-2"; filename*=utf-8''attached_%D1%84%D0%B0%D0%B9%D0%BB.txt
This is the content of the file
--74e78d11b0214bdcbc2f86491eeb4902--
This new format uses field parameter encoding described in RFC 5987. More about this encoding can be found here.
Netty does not parse this format. The result is the filename is not decoded and the part is not parsed into a FileUpload.
Modification:
Added failing test in HttpPostRequestDecoderTest.java and updated HttpPostMultipartRequestDecoder.java
Refactored to please Netkins
Result:
Fixes:
HttpPostMultipartRequestDecoder identifies the RFC 5987 format and parses it.
Previous functionality is retained.
Motivation:
Phantom references are for cleaning up resources that were
forgotten, which means they keep their referent alive. This
means garbage is kept around until the refqueue is drained, rather
than when the reference is unreachable.
Modification:
Use Weak References instead of Phantoms
Result:
More punctual leak detection.
Motivation:
We can end up with a buffer leak if SSLEngine.wrap(...) throws.
Modifications:
Correctly release the ByteBuf if SSLEngine.wrap(...) throws.
Result:
Fixes [#7337].
Motivation:
Before this commit, it is impossible to access the path component of the
URI before it has been decoded. This makes it impossible to distinguish
between the following URIs:
/user/title?key=value
/user%2Ftitle?key=value
The user could already access the raw uri value, but they had to calculate
pathEndIdx themselves, even though it might already be cached inside
QueryStringDecoder.
Result:
The user can easily and efficiently access the undecoded path and query.
Motivation:
Previously client Http2ConnectionHandler trigger a user event
immediately when the HTTP/2 connection preface is sent. Any attempt to
immediately send a new request could cause the server to terminate the
connection, as it might not have received the SETTINGS frame from the
client. Per RFC7540 Section 3.5, the preface "MUST be followed by a
SETTINGS frame (Section 6.5), which MAY be empty."
(https://tools.ietf.org/html/rfc7540#section-3.5)
This event could be made more meaningful if it also indicates that the
initial client SETTINGS frame has been sent to signal that the channel
is ready to send new requests.
Modification:
- Renamed event to Http2ConnectionPrefaceAndSettingsFrameWrittenEvent.
- Modified Http2ConnectionHandler to trigger the user event only if it
is a client and it has sent both the preface and SETTINGS frame.
Result:
It is now safe to use the event as an indicator that the HTTP/2
connection is ready to send new requests.
Motivation:
A regression was introduced in 86e653e which had the effect that the writability was not updated for a Channel while queueing data in the SslHandler.
Modifications:
- Factor out code that will increment / decrement pending bytes and use it in AbstractCoalescingBufferQueue and PendingWriteQueue
- Add test-case
Result:
Channel writability changes are triggered again.
Motivation:
We tried to set IPV6 opts on an ipv4 only system and so failed to set / get the traffic opts. This resulted in a test-error when trying to compile netty on ipv4 only systems.
Modifications:
Use the correct opts depending on if the system is ipv4 only or not.
Result:
Be able to build and use on ipv4 only systems.
Motivation:
Bug in capacity calculation: occurs auto convert to string instead of sum up.
Modifications:
Use `eventName.length()` in sum.
Result:
Less trash in logs.
Motivation:
We should also enforce the handshake timeout on the server-side to allow closing connections which will not finish the handshake in an expected amount of time.
Modifications:
- Enforce the timeout on the server and client side
- Add unit test.
Result:
Fixes [#7230].
Motviation:
DnsNameResolverContext#followCname attempts to build a query to follow a CNAME, but puts the original hostname in the DnsQuery instead of the CNAME hostname. This will result in not following CNAME redirects correctly.
Result:
- DnsNameResolverContext#followCname should use the CNAME instead of the original hostname when building the DnsQuery
Result:
More correct handling of redirect queries.
Motivation:
An `origin`/`sec-websocket-origin` header value in websocket client is filling incorrect in some cases:
- Hostname is not converting to lower-case as prescribed by RFC 6354 (see [1]).
- Selecting a `http` scheme when source URI has `wss`/`https` scheme and non-standard port.
Modifications:
- Convert uri-host to lower-case.
- Use a `https` scheme if source URI scheme is `wss`/`https`, or if source scheme is null and port == 443.
Result:
Correct filling an `origin` header for WS client.
[1] https://tools.ietf.org/html/rfc6454#section-4
Motivation:
Use actual links to new locations of Protobuf repo and documentation to
avoid problems when redirect will not work.
Modification:
Links in comments and all/pom.xml
Result:
Correct links to Protobuf resources
Motivation:
Objects of java.util.TreeMap or java.util.TreeSet will become
non-Serializable if instantiated with Comparators, which are not also
Serializable. This can result in unexpected and difficult-to-diagnose
bugs.
Modifications:
Implements Serializable for all classes, which implements Comparator.
Result:
Proper Comparators which will not force collections to
non-Serializable mode.
Motivation:
Even if it's a super micro-optimization (most JVM could optimize such
cases in runtime), in theory (and according to some perf tests) it
may help a bit. It also makes a code more clear and allows you to
access such methods in the test scope directly, without instance of
the class.
Modifications:
Add 'static' modifier for all methods, where it possible. Mostly in
test scope.
Result:
Cleaner code with proper 'static' modifiers.
Motivation:
Javadoc of the `ByteBufUtil#copy(AsciiString, int, ByteBuf, int, int)` is incorrect.
Modifications:
Fix it.
Result:
The description of the `#copy` method is not misleading.
Motivation:
In the `ByteBufOutputStream` we can use an appropriate methods of `ByteBuf`
to reduce calls of virtual methods and do not copying converting logic.
Modifications:
- Use an appropriate methods of `ByteBuf`
- Remove redundant conversions (int -> byte, int -> char).
- Use `ByteBuf#writeCharSequence` in the `writeBytes(String)'.
Result:
Less code duplication. A `writeBytes(String)` method is faster.
No unnecessary conversions. More consistent and cleaner code.
Motivation:
Without a 'serialVersionUID' field, any change to a class will make
previously serialized versions unreadable.
Modifications:
Add missed 'serialVersionUID' field for all Serializable
classes.
Result:
Proper deserialization of previously serialized objects.
Motivation:
During code read of the Netty codebase I noticed that the Netty
HttpServerUpgradeHandler unconditionally sets a Content-Length: 0
header on 101 Switching Protocols responses. This explicitly
contravenes RFC 7230 Section 3.3.2 (Content-Length), which notes
that:
A server MUST NOT send a Content-Length header field in any
response with a status code of 1xx (Informational) or 204
(No Content).
While it is unlikely that any client will ever be confused by
this behaviour, there is no reason to contravene this part of the
specification.
Modifications:
Removed the line of code setting the header field and changed the
only test that expected it to be there.
Result:
When performing the server portion of HTTP upgrade, the 101
Switching Protocols response will no longer contain a
Content-Length: 0 header field.
Configuring this is tough because there is split between highly shared (and accessed) objects and lightly accessed objects.
Modification:
There are a number of changes here. In relative order of importance:
API / Functionality changes:
* Max records and max sample records are gone. Only "target" records, the number of records tries to retain is exposed.
* Records are sampled based on the number of already stored records. The likelihood of recording a new sample is `2^(-n)`, where `n` is the number of currently stored elements.
* Records are stored in a concurrent stack structure rather than a list. This avoids a head and tail. Since the stack is only read once, there is no need to maintain head and tail pointers
* The properties of this imply that the very first and very last access are always recorded. When deciding to sample, the top element is replaced rather than pushed.
* Samples that happen between the first and last accesses now have a chance of being recorded. Previously only the final few were kept.
* Sampling is no longer deterministic. Previously, a deterministic access pattern meant that you could conceivably always miss some access points.
* Sampling has a linear ramp for low values and and exponentially backs off roughly equal to 2^n. This means that for 1,000,000 accesses, about 20 will actually be kept. I have an elegant proof for this which is too large to fit in this commit message.
Code changes:
* All locks are gone. Because sampling rarely needs to do a write, there is almost 0 contention. The dropped records counter is slightly contentious, but this could be removed or changed to a LongAdder. This was not done because of memory concerns.
* Stack trace exclusion is done outside of RLD. Classes can opt to remove some of their methods.
* Stack trace exclusion is faster, since it uses String.equals, often getting a pointer compare due to interning. Previously it used contains()
* Leak printing is outputted fairly differently. I tried to preserve as much of the original formatting as possible, but some things didn't make sense to keep.
Result:
More useful leak reporting.
Faster:
```
Before:
Benchmark (recordTimes) Mode Cnt Score Error Units
ResourceLeakDetectorRecordBenchmark.record 8 thrpt 20 136293.404 ± 7669.454 ops/s
ResourceLeakDetectorRecordBenchmark.record 16 thrpt 20 72805.720 ± 3710.864 ops/s
ResourceLeakDetectorRecordBenchmark.recordWithHint 8 thrpt 20 139131.215 ± 4882.751 ops/s
ResourceLeakDetectorRecordBenchmark.recordWithHint 16 thrpt 20 74146.313 ± 4999.246 ops/s
After:
Benchmark (recordTimes) Mode Cnt Score Error Units
ResourceLeakDetectorRecordBenchmark.record 8 thrpt 20 155281.969 ± 5301.399 ops/s
ResourceLeakDetectorRecordBenchmark.record 16 thrpt 20 77866.239 ± 3821.054 ops/s
ResourceLeakDetectorRecordBenchmark.recordWithHint 8 thrpt 20 153360.036 ± 8611.353 ops/s
ResourceLeakDetectorRecordBenchmark.recordWithHint 16 thrpt 20 78670.804 ± 2399.149 ops/s
```
Motivation:
Due a bug we happen to sometimes fail the connectPromise with a ClosedChannelException when using the kqueue transport and the remote peer refuses the connection. We need to ensure we fail it with the correct exception.
Modifications:
Call finishConnect() before calling close() to ensure we preserve the correct exception.
Result:
KQueueSocketConnectionAttemptTest.testConnectionRefused will pass always on macOS.
Motivation: Today when Netty encounters a general error while decoding
it treats this as a decoder exception. However, for fatal causes this
should not be treated as such, instead the fatal error should be carried
up the stack without the callee having to unwind causes. This was
probably done for byte to byte message decoder but is now done for all
decoders.
Modifications: Instead of translating any error to a decoder exception,
we let those unwind out the stack (note that finally blocks still
execute) except in places where an event needs to fire where we fire
with the error instead of wrapping in a decoder exception.
Result: Fatal errors will not be treated as innocent decoder exceptions.
Motivation:
We need to upgrade our dependencies to versions which use ASM 6.0.0+ to support compiling on java9.
Modifications:
Update animal-sniffer-maven-plugin and maven-shade-plugin.
Result:
Fixes https://github.com/netty/netty/issues/6100
Motivation: Today when Netty encounters a general error while decoding
it treats this as a decoder exception. However, for fatal causes this
should not be treated as such, instead the fatal error should be carried
up the stack without the callee having to unwind causes.
Modifications: Instead of translating any error to a decoder exception,
we let those unwind out the stack (note that finally blocks still
execute).
Result: Fatal errors will not be treated as innocent decoder exceptions.
Motivation:
There are 2 motivations, the first depends on the second:
Loading Netty Epoll statically stopped working in 4.1.16, due to
`Native` always loading the arch specific shared object. In a
static binary, there is no arch specific SO.
Second, there are a ton of exceptions that can happen when loading
a native library. When loading native code, Netty tries a bunch of
different paths but a failure in any given may not be fatal.
Additionally: turning on debug logging is not always feasible so
exceptions get silently swallowed.
Modifications:
* Change Epoll and Kqueue to try the static load second
* Modify NativeLibraryLoader to record all the locations where
exceptions occur.
* Attempt to use `addSuppressed` from Java 7 if available.
Alternatives Considered:
An alternative would be to record log messages at each failure. If
all load attempts fail, the log messages are printed as warning,
else as debug. The problem with this is there is no `LogRecord` to
create like in java.util.logging. Buffering the args to
logger.log() at the end of the method loses the call site, and
changes the order of events to be confusing.
Another alternative is to teach NativeLibraryLoader about loading
the SO first, and then the static version. This would consolidate
the code fore Epoll, Kqueue, and TCNative. I think this is the
long term better option, but this PR is changing a lot already.
Someone else can take a crack at it later
Results:
Epoll Still Loads and easier debugging.
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
```
