runtime: epollwait on fd 4 failed with 38 w/ go1.10.3

[neo-hu]

Compiled：Linux localhost.localdomain 3.10.0-514.26.2.el7.x86_64 #1 SMP Tue Jul 4 15:04:05 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux
RUN: Linux YF-DGE2 2.6.18-164.el5 #1 SMP Thu Sep 3 03:28:30 EDT 2009 x86_64 x86_64 x86_64 GNU/Linux

CGO_ENABLED=1 GOOS=linux GOARCH=amd64 go build -o /tmp/watchping watchping.go

[root@YF-DGE2 update]# ./watchping
watch agent
watchping: comments to xxxx

The occasional mistake
[root@YF-DGE2 update]# ./watchping
runtime: epollwait on fd 4 failed with 38
fatal error: runtime: netpoll failed

runtime stack:
runtime.throw(0xc63ded, 0x17)
/usr/local/go_1.10/src/runtime/panic.go:616 +0x81
runtime.netpoll(0x1548caff3150ff00, 0x2b7400000000)
/usr/local/go_1.10/src/runtime/netpoll_epoll.go:75 +0x216
runtime.sysmon()
/usr/local/go_1.10/src/runtime/proc.go:4265 +0x424
runtime.mstart1(0x0)
/usr/local/go_1.10/src/runtime/proc.go:1227 +0xe7
runtime.mstart()
/usr/local/go_1.10/src/runtime/proc.go:1193 +0x76

goroutine 1 [runnable, locked to thread]:
vendor/golang_org/x/net/http2/hpack.newInternalNode(...)
/usr/local/go_1.10/src/vendor/golang_org/x/net/http2/hpack/huffman.go:117
vendor/golang_org/x/net/http2/hpack.addDecoderNode(0xfffffe200000002, 0x1c)
/usr/local/go_1.10/src/vendor/golang_org/x/net/http2/hpack/huffman.go:137 +0xdf
vendor/golang_org/x/net/http2/hpack.init.0()
/usr/local/go_1.10/src/vendor/golang_org/x/net/http2/hpack/huffman.go:127 +0x75

go1.9 It's OK.

[mikioh]

What's watchping? Can you show us your snippet watchping.go? [end of handshake protocol]

Please take a look at https://github.com/golang/go/wiki/MinimumRequirements. You are using unsupported, a bit older Linux kernel: Linux YF-DGE2 2.6.18-164.el5. We can guess that the difference btw Go 1.10 and Go 1.9 on filesystem manipulation using runtime-integrated poller might be the root cause, unfortunately. Can you try your snippet on Linux kernel 2.6.23 or above?

[ianlancetaylor]

Errno value 38 is ENOSYS. The change from 1.9 to 1.10 is that in 1.9 we call the epoll_wait system call but in 1.10 we call the epoll_pwait system call. epoll_pwait was added to the Linux kernel in version 2.6.19. You are running on version 2.6.18. So that is why the failure is happening.

According to https://golang.org/wiki/MinimumRequirements we require Linux kernel version 2.6.23 or later. If 1.9 has been working for you, you've been lucky. We have no plans to support kernel versions before 2.6.23, so I'm going to close this issue. That said, you could patch runtime/sys_linux_amd64.s to go back to calling epoll_wait instead of epoll_pwait.