Since TestPingPongHog tests the scheduler, it's ultimately
probabilistic. Currently, it requires the result be at most of factor
of 2 off of the ideal. It turns out this isn't quite enough in
practice, with factors on 1000 iterations on linux/amd64 ranging from
0.48 to 2.5. If the test were failing, we would expect a factor closer
to 1000X, so it's pretty safe to expand the accepted factor from 2 to
5.
Fixes#20494.
Change-Id: If8f2e96194fe66f1fb981a965d1167fe74ff38d7
Reviewed-on: https://go-review.googlesource.com/44859
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Mostly unnecessary *testing.T arguments.
Found with github.com/mvdan/unparam.
Change-Id: Ifb955cb88f2ce8784ee4172f4f94d860fa36ae9a
Reviewed-on: https://go-review.googlesource.com/41691
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Currently it's possible for user code to exploit the high scheduler
priority of the GC worker in conjunction with the runnext optimization
to elevate a user goroutine to high priority so it will always run
even if there are other runnable goroutines.
For example, if a goroutine is in a tight allocation loop, the
following can happen:
1. Goroutine 1 allocates, triggering a GC.
2. G 1 attempts an assist, but fails and blocks.
3. The scheduler runs the GC worker, since it is high priority.
Note that this also starts a new scheduler quantum.
4. The GC worker does enough work to satisfy the assist.
5. The GC worker readies G 1, putting it in runnext.
6. GC finishes and the scheduler runs G 1 from runnext, giving it
the rest of the GC worker's quantum.
7. Go to 1.
Even if there are other goroutines on the run queue, they never get a
chance to run in the above sequence. This requires a confluence of
circumstances that make it unlikely, though not impossible, that it
would happen in "real" code. In the test added by this commit, we
force this confluence by setting GOMAXPROCS to 1 and GOGC to 1 so it's
easy for the test to repeated trigger GC and wake from a blocked
assist.
We fix this by making GC always put user goroutines at the end of the
run queue, instead of in runnext. This makes it so user code can't
piggy-back on the GC's high priority to make a user goroutine act like
it has high priority. The only other situation where GC wakes user
goroutines is waking all blocked assists at the end, but this uses the
global run queue and hence doesn't have this problem.
Fixes#15706.
Change-Id: I1589dee4b7b7d0c9c8575ed3472226084dfce8bc
Reviewed-on: https://go-review.googlesource.com/23172
Reviewed-by: Rick Hudson <rlh@golang.org>
Runqempty is a critical predicate for scheduler. If runqempty spuriously
returns true, then scheduler can fail to schedule arbitrary number of
runnable goroutines on idle Ps for arbitrary long time. With the addition
of runnext runqempty predicate become broken (can spuriously return true).
Consider that runnext is not nil and the main array is empty. Runqempty
observes that the array is empty, then it is descheduled for some time.
Then queue owner pushes another element to the queue evicting runnext
into the array. Then queue owner pops runnext. Then runqempty resumes
and observes runnext is nil and returns true. But there were no point
in time when the queue was empty.
Fix runqempty predicate to not return true spuriously.
Change-Id: Ifb7d75a699101f3ff753c4ce7c983cf08befd31e
Reviewed-on: https://go-review.googlesource.com/20858
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Dmitry Vyukov <dvyukov@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
During random stealing we steal 4*GOMAXPROCS times from random procs.
One would expect that most of the time we check all procs this way,
but due to low quality PRNG we actually miss procs with frightening
probability. Below are modelling experiment results for 1e6 tries:
GOMAXPROCS = 2 : missed 1 procs 7944 times
GOMAXPROCS = 3 : missed 1 procs 101620 times
GOMAXPROCS = 3 : missed 2 procs 3571 times
GOMAXPROCS = 4 : missed 1 procs 63916 times
GOMAXPROCS = 4 : missed 2 procs 61 times
GOMAXPROCS = 4 : missed 3 procs 16 times
GOMAXPROCS = 5 : missed 1 procs 133136 times
GOMAXPROCS = 5 : missed 2 procs 1025 times
GOMAXPROCS = 5 : missed 3 procs 101 times
GOMAXPROCS = 5 : missed 4 procs 15 times
GOMAXPROCS = 8 : missed 1 procs 151765 times
GOMAXPROCS = 8 : missed 2 procs 5057 times
GOMAXPROCS = 8 : missed 3 procs 1726 times
GOMAXPROCS = 8 : missed 4 procs 68 times
GOMAXPROCS = 12 : missed 1 procs 199081 times
GOMAXPROCS = 12 : missed 2 procs 27489 times
GOMAXPROCS = 12 : missed 3 procs 3113 times
GOMAXPROCS = 12 : missed 4 procs 233 times
GOMAXPROCS = 12 : missed 5 procs 9 times
GOMAXPROCS = 16 : missed 1 procs 237477 times
GOMAXPROCS = 16 : missed 2 procs 30037 times
GOMAXPROCS = 16 : missed 3 procs 9466 times
GOMAXPROCS = 16 : missed 4 procs 1334 times
GOMAXPROCS = 16 : missed 5 procs 192 times
GOMAXPROCS = 16 : missed 6 procs 5 times
GOMAXPROCS = 16 : missed 7 procs 1 times
GOMAXPROCS = 16 : missed 8 procs 1 times
A missed proc won't lead to underutilization because we check all procs
again after dropping P. But it can lead to an unpleasant situation
when we miss a proc, drop P, check all procs, discover work, acquire P,
miss the proc again, repeat.
Improve stealing logic to cover all procs.
Also don't enter spinning mode and try to steal when there is nobody around.
Change-Id: Ibb6b122cc7fb836991bad7d0639b77c807aab4c2
Reviewed-on: https://go-review.googlesource.com/20836
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Dmitry Vyukov <dvyukov@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
[Repeat of CL 18343 with build fixes.]
Before, NumGoroutine counted system goroutines and Stack (usually) didn't show them,
which was inconsistent and confusing.
To resolve which way they should be consistent, it seems like
package main
import "runtime"
func main() { println(runtime.NumGoroutine()) }
should print 1 regardless of internal runtime details. Make it so.
Fixes#11706.
Change-Id: If26749fec06aa0ff84311f7941b88d140552e81d
Reviewed-on: https://go-review.googlesource.com/18432
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Russ Cox <rsc@golang.org>
Before, NumGoroutine counted system goroutines and Stack (usually) didn't show them,
which was inconsistent and confusing.
To resolve which way they should be consistent, it seems like
package main
import "runtime"
func main() { println(runtime.NumGoroutine()) }
should print 1 regardless of internal runtime details. Make it so.
Fixes#11706.
Change-Id: I6bfe26a901de517728192cfb26a5568c4ef4fe47
Reviewed-on: https://go-review.googlesource.com/18343
Reviewed-by: Austin Clements <austin@google.com>
Now there are just three programs to compile instead of many,
and repeated tests can reuse the compilation result instead of
rebuilding it.
Combined, these changes reduce the time spent testing runtime
during all.bash on my laptop from about 60 to about 30 seconds.
(All.bash itself runs in 5½ minutes.)
For #10571.
Change-Id: Ie2c1798b847f1a635a860d11dcdab14375319ae9
Reviewed-on: https://go-review.googlesource.com/18085
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
Currently we wake up new worker threads whenever we pass
through the scheduler with nmspinning==0. This leads to
lots of unnecessary thread wake ups.
Instead let only spinning threads wake up new spinning threads.
For the following program:
package main
import "runtime"
func main() {
for i := 0; i < 1e7; i++ {
runtime.Gosched()
}
}
Before:
$ time ./test
real 0m4.278s
user 0m7.634s
sys 0m1.423s
$ strace -c ./test
% time seconds usecs/call calls errors syscall
99.93 9.314936 3 2685009 17536 futex
After:
$ time ./test
real 0m1.200s
user 0m1.181s
sys 0m0.024s
$ strace -c ./test
% time seconds usecs/call calls errors syscall
3.11 0.000049 25 2 futex
Fixes#13527
Change-Id: Ia1f5bf8a896dcc25d8b04beb1f4317aa9ff16f74
Reviewed-on: https://go-review.googlesource.com/17540
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Dmitry Vyukov <dvyukov@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
It's a bad test and it's worst on uniprocessors.
Fixes#11143.
Change-Id: I0164231ada294788d7eec251a2fc33e02a26c13b
Reviewed-on: https://go-review.googlesource.com/12522
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TestGoroutineParallelism can deadlock if the GC runs during the
test. Currently it tries to prevent this by forcing a GC before the
test, but this is best effort and fails completely if GOGC is very low
for testing.
This change replaces this best-effort fix with simply setting GOGC to
off for the duration of the test.
Change-Id: I8229310833f241b149ebcd32845870c1cb14e9f8
Reviewed-on: https://go-review.googlesource.com/10454
Reviewed-by: Russ Cox <rsc@golang.org>
The master goroutine was returning before
the child goroutine had done its final i < b.N
(the one that fails and causes it to exit the loop)
and then the benchmark harness was updating
b.N, causing a read+write race on b.N.
Change-Id: I2504270a0de30544736f6c32161337a25b505c3e
Reviewed-on: https://go-review.googlesource.com/9368
Reviewed-by: Austin Clements <austin@google.com>
Currently, when the runtime ready()s a G, it adds it to the end of the
current P's run queue and continues running. If there are many other
things in the run queue, this can result in a significant delay before
the ready()d G actually runs and can hurt fairness when other Gs in
the run queue are CPU hogs. For example, if there are three Gs sharing
a P, one of which is a CPU hog that never voluntarily gives up the P
and the other two of which are doing small amounts of work and
communicating back and forth on an unbuffered channel, the two
communicating Gs will get very little CPU time.
Change this so that when G1 ready()s G2 and then blocks, the scheduler
immediately hands off the remainder of G1's time slice to G2. In the
above example, the two communicating Gs will now act as a unit and
together get half of the CPU time, while the CPU hog gets the other
half of the CPU time.
This fixes the problem demonstrated by the ping-pong benchmark added
in the previous commit:
benchmark old ns/op new ns/op delta
BenchmarkPingPongHog 684287 825 -99.88%
On the x/benchmarks suite, this change improves the performance of
garbage by ~6% (for GOMAXPROCS=1 and 4), and json by 28% and 36% for
GOMAXPROCS=1 and 4. It has negligible effect on heap size.
This has no effect on the go1 benchmark suite since those benchmarks
are mostly single-threaded.
Change-Id: I858a08eaa78f702ea98a5fac99d28a4ac91d339f
Reviewed-on: https://go-review.googlesource.com/9289
Reviewed-by: Rick Hudson <rlh@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
This benchmark demonstrates a current problem with the scheduler where
a set of frequently communicating goroutines get very little CPU time
in the presence of another goroutine that hogs that CPU, even if one
of those communicating goroutines is always runnable.
Currently it takes about 0.5 milliseconds to switch between
ping-ponging goroutines in the presence of a CPU hog:
BenchmarkPingPongHog 2000 684287 ns/op
Change-Id: I278848c84f778de32344921ae8a4a8056e4898b0
Reviewed-on: https://go-review.googlesource.com/9288
Reviewed-by: Rick Hudson <rlh@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
Currently we always create context objects for closures that capture variables.
However, it is completely unnecessary for direct calls of closures
(whether it is func()(), defer func()() or go func()()).
This change transforms any OCALLFUNC(OCLOSURE) to normal function call.
Closed variables become function arguments.
This transformation is especially beneficial for go func(),
because we do not need to allocate context object on heap.
But it makes direct closure calls a bit faster as well (see BenchmarkClosureCall).
On implementation level it required to introduce yet another compiler pass.
However, the pass iterates only over xtop, so it should not be an issue.
Transformation consists of two parts: closure transformation and call site
transformation. We can't run these parts on different sides of escape analysis,
because tree state is inconsistent. We can do both parts during typecheck,
we don't know how to capture variables and don't have call site.
We can't do both parts during walk of OCALLFUNC, because we can walk
OCLOSURE body earlier.
So now capturevars pass only decides how to capture variables
(this info is required for escape analysis). New transformclosure
pass, that runs just before order/walk, does all transformations
of a closure. And later walk of OCALLFUNC(OCLOSURE) transforms call site.
benchmark old ns/op new ns/op delta
BenchmarkClosureCall 4.89 3.09 -36.81%
BenchmarkCreateGoroutinesCapture 1634 1294 -20.81%
benchmark old allocs new allocs delta
BenchmarkCreateGoroutinesCapture 6 2 -66.67%
benchmark old bytes new bytes delta
BenchmarkCreateGoroutinesCapture 176 48 -72.73%
Change-Id: Ic85e1706e18c3235cc45b3c0c031a9c1cdb7a40e
Reviewed-on: https://go-review.googlesource.com/4050
Reviewed-by: Russ Cox <rsc@golang.org>
Language specification says that variables are captured by reference.
And that is what gc compiler does. However, in lots of cases it is
possible to capture variables by value under the hood without
affecting visible behavior of programs. For example, consider
the following typical pattern:
func (o *Obj) requestMany(urls []string) []Result {
wg := new(sync.WaitGroup)
wg.Add(len(urls))
res := make([]Result, len(urls))
for i := range urls {
i := i
go func() {
res[i] = o.requestOne(urls[i])
wg.Done()
}()
}
wg.Wait()
return res
}
Currently o, wg, res, and i are captured by reference causing 3+len(urls)
allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap).
But all of them can be captured by value without changing behavior.
This change implements simple strategy for capturing by value:
if a captured variable is not addrtaken and never assigned to,
then it is captured by value (it is effectively const).
This simple strategy turned out to be very effective:
~80% of all captures in std lib are turned into value captures.
The remaining 20% are mostly in defers and non-escaping closures,
that is, they do not cause allocations anyway.
benchmark old allocs new allocs delta
BenchmarkCompressedZipGarbage 153 126 -17.65%
BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18%
BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53%
BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40%
BenchmarkEncodeDigitsDefault1e4 100 75 -25.00%
BenchmarkEncodeDigitsDefault1e5 193 139 -27.98%
BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63%
BenchmarkEncodeDigitsCompress1e4 100 75 -25.00%
BenchmarkEncodeDigitsCompress1e5 193 139 -27.98%
BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63%
BenchmarkEncodeTwainSpeed1e4 109 81 -25.69%
BenchmarkEncodeTwainSpeed1e5 211 151 -28.44%
BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92%
BenchmarkEncodeTwainDefault1e4 103 77 -25.24%
BenchmarkEncodeTwainDefault1e5 199 143 -28.14%
BenchmarkEncodeTwainDefault1e6 1324 917 -30.74%
BenchmarkEncodeTwainCompress1e4 103 77 -25.24%
BenchmarkEncodeTwainCompress1e5 190 137 -27.89%
BenchmarkEncodeTwainCompress1e6 1327 919 -30.75%
BenchmarkConcurrentDBExec 16223 16220 -0.02%
BenchmarkConcurrentStmtQuery 17687 16182 -8.51%
BenchmarkConcurrentStmtExec 5191 5186 -0.10%
BenchmarkConcurrentTxQuery 17665 17661 -0.02%
BenchmarkConcurrentTxExec 15154 15150 -0.03%
BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52%
BenchmarkConcurrentTxStmtExec 3677 3673 -0.11%
BenchmarkConcurrentRandom 14000 13614 -2.76%
BenchmarkManyConcurrentQueries 25 22 -12.00%
BenchmarkDecodeComplex128Slice 318 252 -20.75%
BenchmarkDecodeFloat64Slice 318 252 -20.75%
BenchmarkDecodeInt32Slice 318 252 -20.75%
BenchmarkDecodeStringSlice 2318 2252 -2.85%
BenchmarkDecode 11 8 -27.27%
BenchmarkEncodeGray 64 56 -12.50%
BenchmarkEncodeNRGBOpaque 64 56 -12.50%
BenchmarkEncodeNRGBA 67 58 -13.43%
BenchmarkEncodePaletted 68 60 -11.76%
BenchmarkEncodeRGBOpaque 64 56 -12.50%
BenchmarkGoLookupIP 153 139 -9.15%
BenchmarkGoLookupIPNoSuchHost 508 466 -8.27%
BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76%
BenchmarkClientServer 62 59 -4.84%
BenchmarkClientServerParallel4 62 59 -4.84%
BenchmarkClientServerParallel64 62 59 -4.84%
BenchmarkClientServerParallelTLS4 79 76 -3.80%
BenchmarkClientServerParallelTLS64 112 109 -2.68%
BenchmarkCreateGoroutinesCapture 10 6 -40.00%
BenchmarkAfterFunc 1006 1005 -0.10%
Fixes#6632.
Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca
Reviewed-on: https://go-review.googlesource.com/3166
Reviewed-by: Russ Cox <rsc@golang.org>
