It looks like there are some data races in your code http://paste.ubuntu.com/23945891/
It's not possible to debug runtime problems until there are no data races. On Tuesday, 7 February 2017 16:07:55 UTC+11, Dave Cheney wrote: > > > Since there is nothing that changes in the processes over time, the > fact that it kicks in after a few minutes makes me think it may be a > garbage collection issue. > > Running with GODEBUG=gctrace=1 will confirm / deny this hypothesis. > > On Tuesday, 7 February 2017 16:06:09 UTC+11, Jason E. Aten wrote: >> >> >> On linux/amd64 using go1.8rc3, I'm seeing one Go built process peg one >> cpu to 100% after running for a few minutes. >> >> >> Tasks: 499 total, 1 running, 497 sleeping, 0 stopped, 1 >> zombie >> >> >> %Cpu(s): 2.8 us, 3.5 sy, 0.0 ni, 93.7 id, 0.0 wa, 0.0 hi, 0.0 si, >> 0.0 >> st >> >> >> KiB Mem : 12356040+total, 65169064 free, 5018704 used, 53372628 >> buff/cache >> >> >> KiB Swap: 0 total, 0 free, 0 used. 11790566+avail >> Mem >> >> >> >> >> >> >> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ >> COMMAND >> >> >> 84359 jaten 20 0 1156916 52152 4632 S 100.7 0.0 9:05.82 >> vhaline -info -web -addr >> loc+ >> >> >> 84486 jaten 20 0 941964 19884 4268 S 0.7 0.0 0:05.58 >> vhaline -info -addr >> localhos+ >> >> >> 84501 jaten 20 0 1020484 14712 4216 S 0.7 0.0 0:04.95 >> vhaline -info -addr localhos+ >> >> >> when I profile, I get alot fo runtime._ExternalCode. What is that and >> what does it mean? >> >> $ go tool pprof --nodefraction=0.1 vhaline >> ./profile.cpu.runaway.rootnode >> Entering interactive mode (type "help" for >> commands) >> >> >> (pprof) >> top10 >> >> >> 21.16s of 29.97s total >> (70.60%) >> >> >> Dropped 55 nodes (cum <= >> 3s) >> >> >> Showing top 10 nodes out of 16 (cum >= >> 10.61s) >> >> >> flat flat% sum% cum >> cum% >> >> >> 13.55s 45.21% 45.21% 13.55s 45.21% >> runtime._ExternalCode >> >> >> 6.23s 20.79% 66.00% 7.85s 26.19% >> syscall.Syscall >> >> >> 0.27s 0.9% 66.90% 16.42s 54.79% >> github.com/glycerine/vhaline/vhaline.(*srvReader).start.func1 >> >> >> 0.25s 0.83% 67.73% 10.20s 34.03% >> net.(*netFD).Read >> >> >> 0.23s 0.77% 68.50% 3.96s 13.21% >> net.setDeadlineImpl >> >> >> 0.20s 0.67% 69.17% 10.96s 36.57% >> bufio.(*Reader).Peek >> >> >> 0.12s 0.4% 69.57% 10.76s 35.90% >> bufio.(*Reader).fill >> >> >> 0.11s 0.37% 69.94% 11.09s 37.00% >> github.com/glycerine/tmframe2.(*FrameReader).NextFrame >> >> >> 0.11s 0.37% 70.30% 7.96s 26.56% >> syscall.read >> >> >> 0.09s 0.3% 70.60% 10.61s 35.40% >> net.(*conn).Read >> >> >> (pprof) >> >> source code is here; the processes can be built from >> https://github.com/glycerine/vhaline at commit commit >> ca32a98799b87627798e05bb8e6cc29d9cac7ec3 or using tag "runaway". >> >> pprof profiling was turned on; could this account for this peg? >> >> I was running three vhaline processes in a chain; the cpu peg happened to >> the first process, but I've also seen it happen for the 2nd. >> >> they were started like this, in three separate shells >> $ vhaline -info -web -addr localhost:9000 # >> process 84359 >> $ vhaline -info -addr localhost:9001 -parent localhost:9000 -cpu # >> process 84486 >> $ vhaline -info -addr localhost:9002 -parent localhost:9001 -cpu # >> process 84501 >> >> Since there is nothing that changes in the processes over time, the fact >> that it kicks in after a few minutes makes me think it may be a garbage >> collection issue. >> >> >> >> >> > -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
