Motivation
#8563 highlighted race conditions introduced by the prior optimistic
update optimization in 83a19d5650. These
were known at the time but considered acceptable given the perf
benefit in high contention scenarios.
This PR proposes a modified approach which provides roughly half the
gains but stronger concurrency semantics. Race conditions still exist
but their scope is narrowed to much less likely cases (releases
coinciding with retain overflow), and even in those
cases certain guarantees are still assured. Once release() returns true,
all subsequent release/retains are guaranteed to throw, and in
particular deallocate will be called at most once.
Modifications
- Use even numbers internally (including -ve) for live refcounts
- "Final" release changes to odd number (equivalent to refcount 0)
- Retain still uses faster getAndAdd, release uses CAS loop
- First CAS attempt uses non-volatile read
- Thread.yield() after a failed CAS provides a net gain
Result
More (though not completely) robust concurrency semantics for ref
counting; increased latency under high contention, but still roughly
twice as fast as the original logic. Bench results to follow
Motivation:
When a buffer is over-released, the current error message of `IllegalReferenceCountException` is `refCnt: XXX, increment: XXX`, which is confusing. The correct message should be `refCnt: XXX, decrement: XXX`.
Modifications:
Pass `-decrement` to create `IllegalReferenceCountException`.
Result:
The error message will be `refCnt: XXX, decrement: XXX` when a buffer is over-released.
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
```
Motivation:
The documentation for field updates says:
> Note that the guarantees of the {@code compareAndSet}
> method in this class are weaker than in other atomic classes.
> Because this class cannot ensure that all uses of the field
> are appropriate for purposes of atomic access, it can
> guarantee atomicity only with respect to other invocations of
> {@code compareAndSet} and {@code set} on the same updater.
This implies that volatiles shouldn't use normal assignment; the
updater should set them.
Modifications:
Use setter for field updaters that make use of compareAndSet.
Result:
Concurrency compliant code
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:
We can share the code in retain() and retain(...) and also in release() and release(...).
Modifications:
Share code.
Result:
Less duplicated code.
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:
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:
As part of GSOC 2013 we had @mbakkar working on a DNS codec but did not integrate it yet as it needs some cleanup. This commit is based on @mbakkar's work and provide the codec for DNS.
Modifications:
Add DNS codec
Result:
Reusable DNS codec will be included in netty.
This PR also includes a AsynchronousDnsResolver which allows to resolve DNS entries in a non blocking way by make use
of the dns codec and netty transport itself.
- Related: #2163
- Add ResourceLeakHint to allow a user to provide a meaningful information about the leak when touching it
- DefaultChannelHandlerContext now implements ResourceLeakHint to tell where the message is going.
- Cleaner resource leak report by excluding noisy stack trace elements