我有 Go 代码，我已经为它编写了单元测试。测试运行 6.5 秒，它会产生巨大的 CPU 负载。我在 docker 容器中运行测试，它的 CPU 使用率达到 400-500%。我得到了 cpuprofile，它显示了以下结果：
(pprof) top
Showing nodes accounting for 4260ms, 69.49% of 6130ms total
Dropped 80 nodes (cum <= 30.65ms)
Showing top 10 nodes out of 79
flat flat% sum% cum cum%
1250ms 20.39% 20.39% 1250ms 20.39% runtime._ExternalCode /usr/local/go/src/runtime/proc.go
640ms 10.44% 30.83% 960ms 15.66% runtime.scanobject /usr/local/go/src/runtime/mgcmark.go
580ms 9.46% 40.29% 580ms 9.46% runtime.memmove /usr/local/go/src/runtime/memmove_amd64.s
420ms 6.85% 47.15% 730ms 11.91% runtime.mapassign_faststr /usr/local/go/src/runtime/hashmap_fast.go
310ms 5.06% 52.20% 310ms 5.06% runtime.duffcopy /usr/local/go/src/runtime/duff_amd64.s
260ms 4.24% 56.44% 260ms 4.24% runtime.heapBitsForObject /usr/local/go/src/runtime/mbitmap.go
220ms 3.59% 60.03% 220ms 3.59% runtime.scanobject /usr/local/go/src/runtime/mbitmap.go
200ms 3.26% 63.30% 1630ms 26.59% runtime.mallocgc /usr/local/go/src/runtime/malloc.go
200ms 3.26% 66.56% 200ms 3.26% runtime.memclrNoHeapPointers /usr/local/go/src/runtime/memclr_amd64.s
180ms 2.94% 69.49% 2250ms 36.70% runtime.systemstack /usr/local/go/src/runtime/asm_amd64.s
我不明白这里的 ExternalCode 是什么。在这个结果中也没有我的实际代码的功能。我知道我的一些函数可能会调用这些内置的 go 东西，最终可能会在那里花费大量时间。但是它并没有给我方向来改进我的代码的性能。这个结果有什么明显的地方需要注意吗？
我尝试list在 pprof 交互模式下使用命令来显示 CPU 使用率高的功能。
(pprof) list _ExternalCode
Total: 6.13s
ROUTINE ======================== runtime._ExternalCode in /usr/local/go/src/runtime/proc.go
1.25s 1.25s (flat, cum) 20.39% of Total
. . 3231: signalLock uint32
. . 3232: hz int32
. . 3233:}
. . 3234:
. . 3235:func _System() { _System() }
1.25s 1.25s 3236:func _ExternalCode() { _ExternalCode() }
. . 3237:func _LostExternalCode() { _LostExternalCode() }
. . 3238:func _GC() { _GC() }
. . 3239:func _LostSIGPROFDuringAtomic64() { _LostSIGPROFDuringAtomic64() }
. . 3240:
. . 3241:// Counts SIGPROFs received while in atomic64 critical section, on mips{,le}
(pprof)
(pprof)
(pprof)
(pprof) list scanobject
Total: 6.13s
ROUTINE ======================== runtime.scanobject in /usr/local/go/src/runtime/mbitmap.go
220ms 220ms (flat, cum) 3.59% of Total
. . 479:
. . 480:// The caller can test morePointers and isPointer by &-ing with bitScan and bitPointer.
. . 481:// The result includes in its higher bits the bits for subsequent words
. . 482:// described by the same bitmap byte.
. . 483:func (h heapBits) bits() uint32 {
. . 484: // The (shift & 31) eliminates a test and conditional branch
. . 485: // from the generated code.
200ms 200ms 486: return uint32(*h.bitp) >> (h.shift & 31)
. . 487:}
. . 488:
. . 489:// morePointers returns true if this word and all remaining words in this object
. . 490:// are scalars.
. . 491:// h must not describe the second word of the object.
ROUTINE ======================== runtime.scanobject in /usr/local/go/src/runtime/mgcmark.go
640ms 960ms (flat, cum) 15.66% of Total
. . 1155: n = maxObletBytes
. . 1156: }
. . 1157: }
. . 1158:
. . 1159: var i uintptr
100ms 100ms 1160: for i = 0; i < n; i += sys.PtrSize {
. . 1161: // Find bits for this word.
10ms 10ms 1162: if i != 0 {
. . 1163: // Avoid needless hbits.next() on last iteration.
10ms 10ms 1164: hbits = hbits.next()
. . 1165: }
. . 1166: // Load bits once. See CL 22712 and issue 16973 for discussion.
. . 1167: bits := hbits.bits()
. . 1168: // During checkmarking, 1-word objects store the checkmark
. . 1169: // in the type bit for the one word. The only one-word objects
. . 1170: // are pointers, or else they'd be merged with other non-pointer
. . 1171: // data into larger allocations.
80ms 80ms 1172: if i != 1*sys.PtrSize && bits&bitScan == 0 {
. . 1173: break // no more pointers in this object
. . 1174: }
10ms 10ms 1175: if bits&bitPointer == 0 {
. . 1176: continue // not a pointer
. . 1177: }
. . 1178:
. . 1179: // Work here is duplicated in scanblock and above.
. . 1180: // If you make changes here, make changes there too.
20ms 20ms 1181: obj := *(*uintptr)(unsafe.Pointer(b + i))
. . 1182:
. . 1183: // At this point we have extracted the next potential pointer.
. . 1184: // Check if it points into heap and not back at the current object.
340ms 340ms 1185: if obj != 0 && arena_start <= obj && obj < arena_used && obj-b >= n {
. . 1186: // Mark the object.
60ms 300ms 1187: if obj, hbits, span, objIndex := heapBitsForObject(obj, b, i); obj != 0 {
10ms 90ms 1188: greyobject(obj, b, i, hbits, span, gcw, objIndex)
. . 1189: }
. . 1190: }
. . 1191: }
. . 1192: gcw.bytesMarked += uint64(n)
. . 1193: gcw.scanWork += int64(i)