1
0
mirror of https://github.com/golang/go synced 2024-10-02 08:18:32 -06:00
go/src/runtime/mstats.go

394 lines
12 KiB
Go
Raw Normal View History

// Copyright 2009 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
// Memory statistics
package runtime
import "unsafe"
// Statistics.
// Shared with Go: if you edit this structure, also edit type MemStats in mem.go.
type mstats struct {
// General statistics.
alloc uint64 // bytes allocated and not yet freed
total_alloc uint64 // bytes allocated (even if freed)
sys uint64 // bytes obtained from system (should be sum of xxx_sys below, no locking, approximate)
nlookup uint64 // number of pointer lookups
nmalloc uint64 // number of mallocs
nfree uint64 // number of frees
// Statistics about malloc heap.
// protected by mheap.lock
heap_alloc uint64 // bytes allocated and not yet freed (same as alloc above)
heap_sys uint64 // bytes obtained from system
heap_idle uint64 // bytes in idle spans
heap_inuse uint64 // bytes in non-idle spans
heap_released uint64 // bytes released to the os
heap_objects uint64 // total number of allocated objects
// Statistics about allocation of low-level fixed-size structures.
// Protected by FixAlloc locks.
stacks_inuse uint64 // this number is included in heap_inuse above
stacks_sys uint64 // always 0 in mstats
mspan_inuse uint64 // mspan structures
mspan_sys uint64
mcache_inuse uint64 // mcache structures
mcache_sys uint64
buckhash_sys uint64 // profiling bucket hash table
gc_sys uint64
other_sys uint64
// Statistics about garbage collector.
// Protected by mheap or stopping the world during GC.
next_gc uint64 // next gc (in heap_alloc time)
last_gc uint64 // last gc (in absolute time)
pause_total_ns uint64
pause_ns [256]uint64 // circular buffer of recent gc pause lengths
pause_end [256]uint64 // circular buffer of recent gc end times (nanoseconds since 1970)
numgc uint32
enablegc bool
debuggc bool
// Statistics about allocation size classes.
by_size [_NumSizeClasses]struct {
size uint32
nmalloc uint64
nfree uint64
}
runtime: introduce heap_live; replace use of heap_alloc in GC Currently there are two main consumers of memstats.heap_alloc: updatememstats (aka ReadMemStats) and shouldtriggergc. updatememstats recomputes heap_alloc from the ground up, so we don't need to keep heap_alloc up to date for it. shouldtriggergc wants to know how many bytes were marked by the previous GC plus how many bytes have been allocated since then, but this *isn't* what heap_alloc tracks. heap_alloc also includes objects that are not marked and haven't yet been swept. Introduce a new memstat called heap_live that actually tracks what shouldtriggergc wants to know and stop keeping heap_alloc up to date. Unlike heap_alloc, heap_live follows a simple sawtooth that drops during each mark termination and increases monotonically between GCs. heap_alloc, on the other hand, has much more complicated behavior: it may drop during sweep termination, slowly decreases from background sweeping between GCs, is roughly unaffected by allocation as long as there are unswept spans (because we sweep and allocate at the same rate), and may go up after background sweeping is done depending on the GC trigger. heap_live simplifies computing next_gc and using it to figure out when to trigger garbage collection. Currently, we guess next_gc at the end of a cycle and update it as we sweep and get a better idea of how much heap was marked. Now, since we're directly tracking how much heap is marked, we can directly compute next_gc. This also corrects bugs that could cause us to trigger GC early. Currently, in any case where sweep termination actually finds spans to sweep, heap_alloc is an overestimation of live heap, so we'll trigger GC too early. heap_live, on the other hand, is unaffected by sweeping. Change-Id: I1f96807b6ed60d4156e8173a8e68745ffc742388 Reviewed-on: https://go-review.googlesource.com/8389 Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-30 16:01:32 -06:00
// Statistics below here are not exported to Go directly.
tinyallocs uint64 // number of tiny allocations that didn't cause actual allocation; not exported to go directly
runtime: introduce heap_live; replace use of heap_alloc in GC Currently there are two main consumers of memstats.heap_alloc: updatememstats (aka ReadMemStats) and shouldtriggergc. updatememstats recomputes heap_alloc from the ground up, so we don't need to keep heap_alloc up to date for it. shouldtriggergc wants to know how many bytes were marked by the previous GC plus how many bytes have been allocated since then, but this *isn't* what heap_alloc tracks. heap_alloc also includes objects that are not marked and haven't yet been swept. Introduce a new memstat called heap_live that actually tracks what shouldtriggergc wants to know and stop keeping heap_alloc up to date. Unlike heap_alloc, heap_live follows a simple sawtooth that drops during each mark termination and increases monotonically between GCs. heap_alloc, on the other hand, has much more complicated behavior: it may drop during sweep termination, slowly decreases from background sweeping between GCs, is roughly unaffected by allocation as long as there are unswept spans (because we sweep and allocate at the same rate), and may go up after background sweeping is done depending on the GC trigger. heap_live simplifies computing next_gc and using it to figure out when to trigger garbage collection. Currently, we guess next_gc at the end of a cycle and update it as we sweep and get a better idea of how much heap was marked. Now, since we're directly tracking how much heap is marked, we can directly compute next_gc. This also corrects bugs that could cause us to trigger GC early. Currently, in any case where sweep termination actually finds spans to sweep, heap_alloc is an overestimation of live heap, so we'll trigger GC too early. heap_live, on the other hand, is unaffected by sweeping. Change-Id: I1f96807b6ed60d4156e8173a8e68745ffc742388 Reviewed-on: https://go-review.googlesource.com/8389 Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-30 16:01:32 -06:00
// heap_live is the number of bytes considered live by the GC.
// That is: retained by the most recent GC plus allocated
// since then. heap_live <= heap_alloc, since heap_live
// excludes unmarked objects that have not yet been swept.
heap_live uint64
// heap_marked is the number of bytes marked by the previous
// GC. After mark termination, heap_live == heap_marked, but
// unlike heap_live, heap_marked does not change until the
// next mark termination.
heap_marked uint64
runtime: use reachable heap estimate to set trigger/goal Currently, we set the heap goal for the next GC cycle using the size of the marked heap at the end of the current cycle. This can lead to a bad feedback loop if the mutator is rapidly allocating and releasing pointers that can significantly bloat heap size. If the GC were STW, the marked heap size would be exactly the reachable heap size (call it stwLive). However, in concurrent GC, marked=stwLive+floatLive, where floatLive is the amount of "floating garbage": objects that were reachable at some point during the cycle and were marked, but which are no longer reachable by the end of the cycle. If the GC cycle is short, then the mutator doesn't have much time to create floating garbage, so marked≈stwLive. However, if the GC cycle is long and the mutator is allocating and creating floating garbage very rapidly, then it's possible that marked≫stwLive. Since the runtime currently sets the heap goal based on marked, this will cause it to set a high heap goal. This means that 1) the next GC cycle will take longer because of the larger heap and 2) the assist ratio will be low because of the large distance between the trigger and the goal. The combination of these lets the mutator produce even more floating garbage in the next cycle, which further exacerbates the problem. For example, on the garbage benchmark with GOMAXPROCS=1, this causes the heap to grow to ~500MB and the garbage collector to retain upwards of ~300MB of heap, while the true reachable heap size is ~32MB. This, in turn, causes the GC cycle to take upwards of ~3 seconds. Fix this bad feedback loop by estimating the true reachable heap size (stwLive) and using this rather than the marked heap size (stwLive+floatLive) as the basis for the GC trigger and heap goal. This breaks the bad feedback loop and causes the mutator to assist more, which decreases the rate at which it can create floating garbage. On the same garbage benchmark, this reduces the maximum heap size to ~73MB, the retained heap to ~40MB, and the duration of the GC cycle to ~200ms. Change-Id: I7712244c94240743b266f9eb720c03802799cdd1 Reviewed-on: https://go-review.googlesource.com/9177 Reviewed-by: Rick Hudson <rlh@golang.org>
2015-04-21 12:24:25 -06:00
// heap_reachable is an estimate of the reachable heap bytes
// at the end of the previous GC.
heap_reachable uint64
}
var memstats mstats
// A MemStats records statistics about the memory allocator.
type MemStats struct {
// General statistics.
Alloc uint64 // bytes allocated and not yet freed
TotalAlloc uint64 // bytes allocated (even if freed)
Sys uint64 // bytes obtained from system (sum of XxxSys below)
Lookups uint64 // number of pointer lookups
Mallocs uint64 // number of mallocs
Frees uint64 // number of frees
// Main allocation heap statistics.
HeapAlloc uint64 // bytes allocated and not yet freed (same as Alloc above)
HeapSys uint64 // bytes obtained from system
HeapIdle uint64 // bytes in idle spans
HeapInuse uint64 // bytes in non-idle span
HeapReleased uint64 // bytes released to the OS
HeapObjects uint64 // total number of allocated objects
// Low-level fixed-size structure allocator statistics.
// Inuse is bytes used now.
// Sys is bytes obtained from system.
StackInuse uint64 // bytes used by stack allocator
StackSys uint64
MSpanInuse uint64 // mspan structures
MSpanSys uint64
MCacheInuse uint64 // mcache structures
MCacheSys uint64
BuckHashSys uint64 // profiling bucket hash table
GCSys uint64 // GC metadata
OtherSys uint64 // other system allocations
// Garbage collector statistics.
NextGC uint64 // next collection will happen when HeapAlloc ≥ this amount
LastGC uint64 // end time of last collection (nanoseconds since 1970)
PauseTotalNs uint64
PauseNs [256]uint64 // circular buffer of recent GC pause durations, most recent at [(NumGC+255)%256]
PauseEnd [256]uint64 // circular buffer of recent GC pause end times
NumGC uint32
EnableGC bool
DebugGC bool
// Per-size allocation statistics.
// 61 is NumSizeClasses in the C code.
BySize [61]struct {
Size uint32
Mallocs uint64
Frees uint64
}
}
// Size of the trailing by_size array differs between Go and C,
// and all data after by_size is local to runtime, not exported.
// NumSizeClasses was changed, but we can not change Go struct because of backward compatibility.
// sizeof_C_MStats is what C thinks about size of Go struct.
var sizeof_C_MStats = unsafe.Offsetof(memstats.by_size) + 61*unsafe.Sizeof(memstats.by_size[0])
func init() {
var memStats MemStats
if sizeof_C_MStats != unsafe.Sizeof(memStats) {
println(sizeof_C_MStats, unsafe.Sizeof(memStats))
throw("MStats vs MemStatsType size mismatch")
}
}
// ReadMemStats populates m with memory allocator statistics.
func ReadMemStats(m *MemStats) {
// Have to acquire worldsema to stop the world,
// because stoptheworld can only be used by
// one goroutine at a time, and there might be
// a pending garbage collection already calling it.
semacquire(&worldsema, false)
gp := getg()
gp.m.preemptoff = "read mem stats"
systemstack(stoptheworld)
systemstack(func() {
readmemstats_m(m)
})
gp.m.preemptoff = ""
gp.m.locks++
semrelease(&worldsema)
systemstack(starttheworld)
gp.m.locks--
}
func readmemstats_m(stats *MemStats) {
updatememstats(nil)
// Size of the trailing by_size array differs between Go and C,
// NumSizeClasses was changed, but we can not change Go struct because of backward compatibility.
memmove(unsafe.Pointer(stats), unsafe.Pointer(&memstats), sizeof_C_MStats)
// Stack numbers are part of the heap numbers, separate those out for user consumption
stats.StackSys += stats.StackInuse
stats.HeapInuse -= stats.StackInuse
stats.HeapSys -= stats.StackInuse
}
//go:linkname readGCStats runtime/debug.readGCStats
func readGCStats(pauses *[]uint64) {
systemstack(func() {
readGCStats_m(pauses)
})
}
func readGCStats_m(pauses *[]uint64) {
p := *pauses
// Calling code in runtime/debug should make the slice large enough.
if cap(p) < len(memstats.pause_ns)+3 {
throw("short slice passed to readGCStats")
}
// Pass back: pauses, pause ends, last gc (absolute time), number of gc, total pause ns.
lock(&mheap_.lock)
n := memstats.numgc
if n > uint32(len(memstats.pause_ns)) {
n = uint32(len(memstats.pause_ns))
}
// The pause buffer is circular. The most recent pause is at
// pause_ns[(numgc-1)%len(pause_ns)], and then backward
// from there to go back farther in time. We deliver the times
// most recent first (in p[0]).
p = p[:cap(p)]
for i := uint32(0); i < n; i++ {
j := (memstats.numgc - 1 - i) % uint32(len(memstats.pause_ns))
p[i] = memstats.pause_ns[j]
p[n+i] = memstats.pause_end[j]
}
p[n+n] = memstats.last_gc
p[n+n+1] = uint64(memstats.numgc)
p[n+n+2] = memstats.pause_total_ns
unlock(&mheap_.lock)
*pauses = p[:n+n+3]
}
//go:nowritebarrier
func updatememstats(stats *gcstats) {
if stats != nil {
*stats = gcstats{}
}
for mp := allm; mp != nil; mp = mp.alllink {
if stats != nil {
src := (*[unsafe.Sizeof(gcstats{}) / 8]uint64)(unsafe.Pointer(&mp.gcstats))
dst := (*[unsafe.Sizeof(gcstats{}) / 8]uint64)(unsafe.Pointer(stats))
for i, v := range src {
dst[i] += v
}
mp.gcstats = gcstats{}
}
}
memstats.mcache_inuse = uint64(mheap_.cachealloc.inuse)
memstats.mspan_inuse = uint64(mheap_.spanalloc.inuse)
memstats.sys = memstats.heap_sys + memstats.stacks_sys + memstats.mspan_sys +
memstats.mcache_sys + memstats.buckhash_sys + memstats.gc_sys + memstats.other_sys
// Calculate memory allocator stats.
// During program execution we only count number of frees and amount of freed memory.
// Current number of alive object in the heap and amount of alive heap memory
// are calculated by scanning all spans.
// Total number of mallocs is calculated as number of frees plus number of alive objects.
// Similarly, total amount of allocated memory is calculated as amount of freed memory
// plus amount of alive heap memory.
memstats.alloc = 0
memstats.total_alloc = 0
memstats.nmalloc = 0
memstats.nfree = 0
for i := 0; i < len(memstats.by_size); i++ {
memstats.by_size[i].nmalloc = 0
memstats.by_size[i].nfree = 0
}
// Flush MCache's to MCentral.
systemstack(flushallmcaches)
// Aggregate local stats.
cachestats()
// Scan all spans and count number of alive objects.
lock(&mheap_.lock)
for i := uint32(0); i < mheap_.nspan; i++ {
s := h_allspans[i]
if s.state != mSpanInUse {
continue
}
if s.sizeclass == 0 {
memstats.nmalloc++
memstats.alloc += uint64(s.elemsize)
} else {
memstats.nmalloc += uint64(s.ref)
memstats.by_size[s.sizeclass].nmalloc += uint64(s.ref)
memstats.alloc += uint64(s.ref) * uint64(s.elemsize)
}
}
unlock(&mheap_.lock)
// Aggregate by size class.
smallfree := uint64(0)
memstats.nfree = mheap_.nlargefree
for i := 0; i < len(memstats.by_size); i++ {
memstats.nfree += mheap_.nsmallfree[i]
memstats.by_size[i].nfree = mheap_.nsmallfree[i]
memstats.by_size[i].nmalloc += mheap_.nsmallfree[i]
smallfree += uint64(mheap_.nsmallfree[i]) * uint64(class_to_size[i])
}
memstats.nfree += memstats.tinyallocs
memstats.nmalloc += memstats.nfree
// Calculate derived stats.
memstats.total_alloc = uint64(memstats.alloc) + uint64(mheap_.largefree) + smallfree
memstats.heap_alloc = memstats.alloc
memstats.heap_objects = memstats.nmalloc - memstats.nfree
}
//go:nowritebarrier
func cachestats() {
for i := 0; ; i++ {
p := allp[i]
if p == nil {
break
}
c := p.mcache
if c == nil {
continue
}
purgecachedstats(c)
}
}
//go:nowritebarrier
func flushallmcaches() {
for i := 0; ; i++ {
p := allp[i]
if p == nil {
break
}
c := p.mcache
if c == nil {
continue
}
mCache_ReleaseAll(c)
stackcache_clear(c)
}
}
//go:nosplit
func purgecachedstats(c *mcache) {
// Protected by either heap or GC lock.
h := &mheap_
runtime: introduce heap_live; replace use of heap_alloc in GC Currently there are two main consumers of memstats.heap_alloc: updatememstats (aka ReadMemStats) and shouldtriggergc. updatememstats recomputes heap_alloc from the ground up, so we don't need to keep heap_alloc up to date for it. shouldtriggergc wants to know how many bytes were marked by the previous GC plus how many bytes have been allocated since then, but this *isn't* what heap_alloc tracks. heap_alloc also includes objects that are not marked and haven't yet been swept. Introduce a new memstat called heap_live that actually tracks what shouldtriggergc wants to know and stop keeping heap_alloc up to date. Unlike heap_alloc, heap_live follows a simple sawtooth that drops during each mark termination and increases monotonically between GCs. heap_alloc, on the other hand, has much more complicated behavior: it may drop during sweep termination, slowly decreases from background sweeping between GCs, is roughly unaffected by allocation as long as there are unswept spans (because we sweep and allocate at the same rate), and may go up after background sweeping is done depending on the GC trigger. heap_live simplifies computing next_gc and using it to figure out when to trigger garbage collection. Currently, we guess next_gc at the end of a cycle and update it as we sweep and get a better idea of how much heap was marked. Now, since we're directly tracking how much heap is marked, we can directly compute next_gc. This also corrects bugs that could cause us to trigger GC early. Currently, in any case where sweep termination actually finds spans to sweep, heap_alloc is an overestimation of live heap, so we'll trigger GC too early. heap_live, on the other hand, is unaffected by sweeping. Change-Id: I1f96807b6ed60d4156e8173a8e68745ffc742388 Reviewed-on: https://go-review.googlesource.com/8389 Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-30 16:01:32 -06:00
memstats.heap_live += uint64(c.local_cachealloc)
c.local_cachealloc = 0
if trace.enabled {
traceHeapAlloc()
}
memstats.tinyallocs += uint64(c.local_tinyallocs)
c.local_tinyallocs = 0
memstats.nlookup += uint64(c.local_nlookup)
c.local_nlookup = 0
h.largefree += uint64(c.local_largefree)
c.local_largefree = 0
h.nlargefree += uint64(c.local_nlargefree)
c.local_nlargefree = 0
for i := 0; i < len(c.local_nsmallfree); i++ {
h.nsmallfree[i] += uint64(c.local_nsmallfree[i])
c.local_nsmallfree[i] = 0
}
}
// Atomically increases a given *system* memory stat. We are counting on this
// stat never overflowing a uintptr, so this function must only be used for
// system memory stats.
//
// The current implementation for little endian architectures is based on
// xadduintptr(), which is less than ideal: xadd64() should really be used.
// Using xadduintptr() is a stop-gap solution until arm supports xadd64() that
// doesn't use locks. (Locks are a problem as they require a valid G, which
// restricts their useability.)
//
// A side-effect of using xadduintptr() is that we need to check for
// overflow errors.
//go:nosplit
func mSysStatInc(sysStat *uint64, n uintptr) {
if _BigEndian != 0 {
xadd64(sysStat, int64(n))
return
}
if val := xadduintptr((*uintptr)(unsafe.Pointer(sysStat)), n); val < n {
print("runtime: stat overflow: val ", val, ", n ", n, "\n")
exit(2)
}
}
// Atomically decreases a given *system* memory stat. Same comments as
// mSysStatInc apply.
//go:nosplit
func mSysStatDec(sysStat *uint64, n uintptr) {
if _BigEndian != 0 {
xadd64(sysStat, -int64(n))
return
}
if val := xadduintptr((*uintptr)(unsafe.Pointer(sysStat)), uintptr(-int64(n))); val+n < n {
print("runtime: stat underflow: val ", val, ", n ", n, "\n")
exit(2)
}
}