Motivation:
When profiling it is sometimes needed to still have the native library file avaible. We should allow to disable the explicit deletion and just delete it when the JVM stops.
This is related to #6110
Modifications:
Add io.netty.native.deleteLibAfterLoading system property which allows to disable the explicit delete after laoding
Result:
Possible to profile native libraries better.
Motivation:
In later Java8 versions our Atomic*FieldUpdater are slower then the JDK implementations so we should not use ours anymore. Even worse the JDK implementations provide for example an optimized version of addAndGet(...) using intrinsics which makes it a lot faster for this use-case.
Modifications:
- Remove methods that return our own Atomic*FieldUpdaters.
- Use the JDK implementations everywhere.
Result:
Faster code.
Motivation:
c2f4daa7398a9363fde8e51bf52c0d0323f870a5 added a unit test but used a too small test timeout.
Modifications:
Increase timeout.
Result:
Test should have enough time to complete on the CI.
Motivation:
InternalLoggerFactory either sets a default logger factory
implementation based on the logging implementations on the classpath, or
applications can set a logger factory explicitly. If applications wait
too long to set the logger factory, Netty will have already set a logger
factory leading to some objects using one logging implementation and
other objets using another logging implementation. This can happen too
if the application tries to set the logger factory twice, which is
likely a bug in the application. Yet, the Javadocs for
InternalLoggerFactory warn against this saying that
InternalLoggerFactory#setLoggerFactory "should be called as early as
possible and shouldn't be called more than once". Instead, Netty should
guard against this.
Modications:
We replace the logger factory field with an atomic reference on which we
can do CAS operations to safely guard against it being set twice. We
also add an internal holder class that captures the static interface of
InternalLoggerFactory that can aid in testing.
Result:
The logging factory can not be set twice, and applications that want to
set the logging factory must do it before any Netty classes are
initialized (or the default logger factory will be set).
Motivation:
We need to ensure the tracked object can not be GC'ed before ResourceLeak.close() is called as otherwise we may get false-positives reported by the ResourceLeakDetector. This can happen as the JIT / GC may be able to figure out that we do not need the tracked object anymore and so already enqueue it for collection before we actually get a chance to close the enclosing ResourceLeak.
Modifications:
- Add ResourceLeakTracker and deprecate the old ResourceLeak
- Fix some javadocs to correctly release buffers.
- Add a unit test for ResourceLeakDetector that shows that ResourceLeakTracker has not the problems.
Result:
No more false-positives reported by ResourceLeakDetector when ResourceLeakDetector.track(...) is used.
Motivation:
Java9 will be released soon so we should ensure we can compile netty with Java9 and run all our tests. This will help to make sure Netty will be usable with Java9.
Modification:
- Add some workarounds to be able to compile with Java9, note that the full profile is not supported with Java9 atm.
- Remove some usage of internal APIs to be able to compile on java9
- Not support Alpn / Npn and so not run the tests when using Java9 for now. We will do a follow up PR to add support.
Result:
Its possible to build netty and run its testsuite with Java9.
Motivation:
42fba015ce82ab4ab30e547c888db82fe74094e9 changed the implemention of ResourceLeakDetector to improve performance. While this was done a branch was missed that can be removed. Beside this using a Boolean as value for the ConcurrentMap is sub-optimal as when calling remove(key, value) an uncessary instanceof check and cast is needed on each removal.
Modifications:
- Remove branch which is not needed anymore
- Replace usage of Boolean as value type of the ConcurrentMap and use our own special type which only compute hash-code one time and use a == operation for equals(...) to reduce overhead present when using Boolean.
Result:
Faster and cleaner ResourceLeakDetector.
Motivation:
Netty has a flag (io.netty.noUnsafe) for specifying to Netty to not be
unsafe. Yet, when initializing PlatformDependent0, Netty still tries to
be unsafe. For application that specify to Netty to not be unsafe and
run under a security manager, this can lead to an obnoxious (debug
level) stack trace. Since Netty was told not to be unsafe, Netty should
not try to be unsafe.
Modifications:
The initialization logic in PlatformDependent0 should take into account
that Netty was told not to be unsafe. This means that we need to
initialize PlatformDependent#IS_EXPLICIT_NO_UNSAFE as soon as possible,
before the static initializer for PlatformDependent0 has a chance to
run. Thus the following modifications are made:
- initialize PlatformDependent#IS_EXPLICIT_NO_UNSAFE before any other
code in PlatformDependent causes PlatformDependent0 to initialize
- expose the value of PlatformDependent#IS_EXPLICIT_NO_UNSAFE for
reading in PlatformDependent0
- take the value of PlatformDependent#IS_EXPLICIT_NO_UNSAFE into
account in PlatformDependent0
Result:
Netty does not try to be unsafe when told not to be unsafe.
Motivation:
For applications that set their own logger factory, they want that
logger factory to be the one logger factory. Yet, Netty eagerly
initializes this and then triggers initialization of other classes
before the application has had a chance to set its preferred logger
factory.
Modifications:
With this commit there are two key changes:
- Netty does not attempt to eagerly initialize the default logger
factory, only doing so if the application layer above Netty has not
already set a logger factory
- do not eagerly initialize unrelated classes from the logger factory;
while the motivation behind this was to initialize ThreadLocalRandom
as soon as possible in case it has to block reading from /dev/random,
this can be worked around for applications where it is problematic by
setting securerandom.source=file:/dev/urandom in their Java system
security policy (no, it is not less secure; do not even get me
started on myths about /dev/random)
Result:
Netty uses the logger factory that the application prefers, and does not
initialize unrelated classes.
Motivation:
PlatformDependent#getSystemClassLoader may throw a wide variety of exceptions based upon the environment. We should handle all exceptions and continue initializing the slow path if an exception occurs.
Modifications:
- Catch Throwable in cases where PlatformDependent#getSystemClassLoader is used
Result:
Fixes https://github.com/netty/netty/issues/6038
Motiviation:
We used ReferenceCountUtil.releaseLater(...) in our tests which simplifies a bit the releasing of ReferenceCounted objects. The problem with this is that while it simplifies stuff it increase memory usage a lot as memory may not be freed up in a timely manner.
Modifications:
- Deprecate releaseLater(...)
- Remove usage of releaseLater(...) in tests.
Result:
Less memory needed to build netty while running the tests.
Motivation:
00fc239995ce7b683993c88bb82b30dbd5fe7982 introduced a change to HashedWheelTimerTest which attempted to wait for an explicit event notification until more timer events can be added. However HashedWheelTimer will execute the timer Runnable before removing it from the queue and decrementing the total count. This make it difficult for users to know when it is safe to add another timer task as the limit is approached.
Modifications:
- HashedWheelTimer should remove the timer Runnable before executing the task.
Result:
Users can more reliably add new timers when the limit is reached and HashedWheelTimerTest will no longer fail spuriously due to this race condition.
Motivation:
HashWheelTimerTest has busy/wait and sleep statements which are not necessary. We also depend upon a com.google.common.base.Supplier which isn't necessary.
Modifications:
- Remove buys wait loops and timeouts where possible
Result:
HashWheelTimerTest more explicit in verifying conditions and less reliant on wait times.
Motivation:
If the rate at which new timeouts are created is very high and the created timeouts are not cancelled, then the JVM can crash because of out of heap space. There should be a guard in the implementation to prevent this.
Modifications:
The constructor of HashedWheelTimer now takes an optional max pending timeouts parameter beyond which it will reject new timeouts by throwing RejectedExecutionException.
Result:
After this change, if the max pending timeouts parameter is passed as constructor argument to HashedWheelTimer, then it keeps a track of pending timeouts that aren't yet expired or cancelled. When a new timeout is being created, it checks for current pending timeouts and if it's equal to or greater than provided max pending timeouts, then it throws RejectedExecutionException.
Motivation:
ResourceLeakDetector shows two main problems, racy access and heavy lock contention.
Modifications:
This PR fixes this by doing two things:
1. Replace the sampling counter with a ThreadLocalRandom. This has two benefits.
First, it makes the sampling ration no longer have to be a power of two. Second,
it de-noises the continuous races that fight over this single value. Instead,
this change uses slightly more CPU to decide if it should sample by using TLR.
2. DefaultResourceLeaks need to be kept alive in order to catch leaks. The means
by which this happens is by a singular, doubly-linked list. This creates a
large amount of contention when allocating quickly. This is noticeable when
running on a multi core machine.
Instead, this uses a concurrent hash map to keep track of active resources
which has much better contention characteristics.
Results:
Better concurrent hygiene. Running the gRPC QPS benchmark showed RLD taking about
3 CPU seconds for every 1 wall second when runnign with 12 threads.
There are some minor perks to this as well. DefaultResourceLeak accounting is
moved to a central place which probably has better caching behavior.
Motivation:
ResourceLeakDetector reports leak for first call to open(obj) as its leakCheckCnt starts with value 0 and increment subsequently. with value of leakCheckCnt =0, it always returns ResourceLeak. Our application calls ResourceLeakDetector.open(obj) to validate Leak and it fails at very first call even though there is no leak in application.
Modifications:
ResourceLeakDetector.leakCheckCnt value will not be 0 while deriving leak and it will not return incorrect value of ResourceLeak.
Result:
Fix false leak report on first call on ResourceLeakDetector.
Motivation:
At the moment we log very confusing messages when trying to load a native library which kind of suggest that the whole loading process failed even if just one mechanism failed and the library could be loaded at the end.
Modifications:
Make the mesage less confusing and also log a successful load of the native library.
Result:
Less confusing logs.
Motivation:
Since netty shaded JCTools the OSGi manifest no longer is correct. It claims to
have an optional import "org.jctools.queues;resolution:=optional,org.jctools.qu
eues.atomic;resolution:=optional,org.jctools.util;resolution:=optional"
However since it is shaded, this is no longer true.
This was noticed when making JCTools a real bundle and netty resolved it as
optional import.
Modifications:
Modify the generated manifest by no longer analyzing org.jctools for imports.
A manual setting of sun.misc as optional was required.
Result:
Netty OSGi bundle will no longer interfere with a JCTools bundle.
Motivation:
To make it easier to debug why notification of a promise failed we should log extra info and make it consistent.
Modifications:
- Create a new PromiseNotificationUtil that has static methods that can be used to try notify a promise and log.
- Reuse this in AbstractChannelHandlerContext, ChannelOutboundBuffer and PromiseNotifier
Result:
Easier to debug why a promise could not be notified.
Motivation:
We can share the code in retain() and retain(...) and also in release() and release(...).
Modifications:
Share code.
Result:
Less duplicated code.
Motivation:
Windows refuses to load a .DLL file when it's opened by other process.
Recent modification in NativeLibraryLoader causes NativeLibraryLoader to
attempt to load a .DLL before closing its OutputStream. As a result,
loading a .DLL file in Windows always fails.
Modifications:
Close the OutputStream explicitly before loading a shared library.
Result:
Native library loading in Windows works again.
Motivation:
SystemPropertyUtil requires InternalLoggerFactory requires ThreadLocalRandom requires SystemPropertyUtil. This can lead to a dead-lock.
Modifications:
Ensure ThreadLocalRandom does not require SystemPropertyUtil during initialization.
Result:
No more deadlock possible.
Motivation:
Sometimes it is useful to be able to wrap an existing memory address (a.k.a pointer) and create a ByteBuf from it. This way its easier to interopt with other libraries.
Modifications:
Add a new Unpooled.wrappedBuffer(....) method that takes a memory address.
Result:
Be able to wrap an existing memory address into a ByteBuf.
Motivation:
As the issue #5539 say, the OpenSsl.class will throw `java.lang.UnsatisfiedLinkError: org.apache.tomcat.jni.Library.version(I)I` when it is invoked. This path try to resolve the problem by modifying the native library loading logic of OpenSsl.class.
Modifications:
The OpenSsl.class loads the tcnative lib by `NativeLibraryLoader.loadFirstAvailable()`. The native library will be loaded in the bundle `netty-common`'s ClassLoader, which is diff with the native class's ClassLoader. That is the root cause of throws `UnsatisfiedLinkError` when the native method is invoked.
So, it should load the native library by the its bundle classloader firstly, then the embedded resources if failed.
Result:
First of all, the error threw by native method problem will be resolved.
Secondly, the native library should work as normal in non-OSGi env. But, this is hard. The loading logic of `Library.class` in `netty-tcnative` bundle is simple: try to load the library in PATH env. If not found, it falls back to the originally logic `NativeLibraryLoader.loadFirstAvailable()`.
Signed-off-by: XU JINCHUAN <xsir@msn.com>
Motivation:
When try to call Cleaner.run() via reflection on Java9 you may see an IllegalAccessException.
Modifications:
Just cast the Cleaner to Runnable to prevent IllegalAccessException to be raised.
Result:
Free direct buffers also work on Java9+ as expected.
Motivation:
Our current strategy in NativeLibraryLoader is to mark the temporary .so file to be deleted on JVM exit. This has the drawback to not delete the file in the case the JVM dies or is killed.
Modification:
Just directly try to delete the file one we loaded the native library and if this fails mark the file to be removed once the JVM exits.
Result:
Less likely to have temporary files still on the system in case of JVM kills.
Motiviation:
Preparing platform dependent code for using unsafe requires executing
privileged code. The privileged code for initializing unsafe is executed
in a manner that would require all code leading up to the initialization
to have the requisite permissions. Yet, in a restrictive environment
(e.g., under a security policy that only grants the requisite
permissions the Netty common jar but not to application code triggering
the Netty initialization), then initializing unsafe will not succeed
even if the security policy would otherwise permit it.
Modifications:
This commit marks the necessary blocks as privileged. This enables
access to the necessary resources for initialization unsafe. The idea is
that we are saying the Netty code is trusted, and as long as the Netty
code has been granted the necessary permissions, then we will allow the
caller access to these resources even though the caller itself might not
have the requisite permissions.
Result:
Unsafe can be initialized in a restrictive security environment.
Motivation:
We not need to do an extra conditional check in retain(...) as we can just check for overflow after we did the increment.
Modifications:
- Remove extra conditional check
- Add test code.
Result:
One conditional check less.
Motivation:
If the user uses 0 as quiet period we should shutdown without any delay if possible.
Modifications:
Ensure we not introduce extra delay when a shutdown quit period of 0 is used.
Result:
EventLoop shutdown as fast as expected.
Motivation:
To better restrict resource usage we should limit the number of WeakOrderQueue instances per Thread. Once this limit is reached object that are recycled from a different Thread then the allocation Thread are dropped on the floor.
Modifications:
Add new system property io.netty.recycler.maxDelayedQueuesPerThread and constructor that allows to limit the max number of WeakOrderQueue instances per Thread for Recycler instance. The default is 2 * cores (the same as the default number of EventLoop instances per EventLoopGroup).
Result:
Better way to restrict resource / memory usage per Recycler instance.
Motivation:
Commons logger is dead and not updated for more than 2 years. #5615.
Modifications:
Added @Deprecated annotation to CommonsLoggerFactory and CommonsLogger.
Result:
Commons logger now deprecated.
Motivation:
When Netty components are initialized, Netty attempts to determine if it
has access to unsafe. If Netty is not able to access unsafe (because of
security permissions, or because the JVM was started with an explicit
flag to tell Netty to not look for unsafe), Netty logs an info-level
message that looks like a warning:
Your platform does not provide complete low-level API for accessing
direct buffers reliably. Unless explicitly requested, heap buffer will
always be preferred to avoid potential system unstability.
This log message can appear in applications that depend on Netty for
networking, and this log message can be scary to end-users of such
platforms. This log message should not be emitted if the application was
started with an explicit flag telling Netty to not look for unsafe.
Modifications:
This commit refactors the unsafe detection logic to expose whether or
not the JVM was started with a flag telling Netty to not look for
unsafe. With this exposed information, the log message on unsafe being
unavailable can be modified to not be emitted when Netty is explicitly
told to not look for unsafe.
Result:
No log message is produced when unsafe is unavailable because Netty was
told to not look for it.
Motivation:
Old code doesn't needed anymore due to logger factory initialization.
Modifications :
Removed static section and useless static variables;
Logging concatenations replaced with placeholders.
Result:
Cleaner, simpler code doing the same
Motivation:
Some usages of findNextPositivePowerOfTwo assume that bounds checking is taken care of by this method. However bounds checking is not taken care of by findNextPositivePowerOfTwo and instead assert statements are used to imply the caller has checked the bounds. This can lead to unexpected non power of 2 return values if the caller is not careful and thus invalidate any logic which depends upon a power of 2.
Modifications:
- Add a safeFindNextPositivePowerOfTwo method which will do runtime bounds checks and always return a power of 2
Result:
Fixes https://github.com/netty/netty/issues/5601
Motivation
Starting from Java 7 method String.split in java is optimized for splitting by single char string. Thus custom StringUtil.split method doesn't have any sense anymore. Even more - it slower than optimized java version.
Modifications:
Replaced all occurrences of StringUtil.split with String.split.
Result:
Less code and faster on jdk7
Motivation:
NewLine initializing is complex, with unnecessary allocations and non-standard.
Static section is overloaded with StringBuilders for simple "s" + "s" concatenation pattern that compiler optimizes perfectly.
Modifications:
NewLine initializing replaced with standard System.getProperty("line.separator").
Removed StringBuilders in static section.
Result:
Less complex code.
Motivation:
When we try to close the Channel due a timeout we need to ensure we not log if the notification of the promise fails as it may be completed in the meantime.
Modifications:
Add another constructor to ChannelPromiseNotifier and PromiseNotifier which allows to log on notification failure.
Result:
No more miss-leading logs.
Motivation:
At the moment the Recyler is very sensitive to allocation bursts which means that if there is a need for X objects for only one time these will most likely end up in the Recycler and sit there forever as the normal workload only need a subset of this number.
Modifications:
Add a ratio which sets how many objects should be pooled for each new allocation. This allows to slowly increase the number of objects in the Recycler while not be to sensitive for bursts.
Result:
Less unused objects in the Recycler if allocation rate sometimes bursts.
Motivation:
Using Attribute.remove() and Attribute.getAndRemove() in a multi-threaded enviroment has its drawbacks. Make sure we document these.
Modifications:
Add javadocs and mark Attribute.remove() and Attribute.getAndRemove() as @Deprecated.
Result:
Hopefully less suprising behaviour.
Motivation:
We used a very high number for DEFAULT_INITIAL_MAX_CAPACITY (over 200k) which is not very relastic and my lead to very surprising memory usage if allocations happen in bursts.
Modifications:
Use a more sane default value of 32k
Result:
Less possible memory usage by default
Motivation:
AbstractReferenceCounted as independent conditional statements to check the bounds of the retain IllegalReferenceCountException condition. One of the exceptions also uses the incorrect increment.
Modifications:
- Combined independent conditional checks into 1 where possible
- Correct IllegalReferenceCountException with incorrect increment
- Remove the subtract to check for overflow and re-use the addition and check for overflow to remove 1 arithmetic operation (see http://docs.oracle.com/javase/specs/jls/se7/html/jls-15.html#jls-15.18.2)
Result:
AbstractReferenceCounted has less independent branch statements and more correct IllegalReferenceCountException. Compilation size of AbstractReferenceCounted.retain() is reduced from 58 bytes to 47 bytes.
Motivation:
We saw some sporadic test failures for GlobalEventExecutorTest.testAutomaticStartStop test. This is caused parallel execution of tests in combination with assert checks that will be affected.
Modifications:
Remove fragile assert checks.
Result:
No more sporadic test failures
