cmd/go: fork/exec /home/rsc/g/go/bin/tool/cgo: text file busy

[rsc]

On Linux I sometimes get the error 

fork/exec /home/rsc/g/go/bin/tool/cgo: text file busy

when cmd/go tries to build runtime/cgo during all.bash.
I don't know what would cause this.  I didn't know you could
get that error from exec.

[ianlancetaylor]

Comment 1:

That error from exec means that you are trying to exec a file that some process has open
for write access.

[rsc]

Comment 2:

Appears to be fixed by recent build changes, which are more careful about using the
right cgo during cross-compiles.

Status changed to Fixed.

[alberts]

Comment 3:

This just happened again on our linux/amd64 machine with a build of e285ee3ad289.

[robpike]

Comment 4:

Status changed to Accepted.

[robpike]

Comment 5:

Issue #3114 has been merged into this issue.

[alberts]

Comment 6:

Sadly, stracing the build slows it down to the extent that you don't see this problem
anymore.

[minux]

Comment 7:

I think the real reason is: we didn't take cmd/cgo into dependence graph for packages
using cgo.

[rsc]

Comment 8:

Yes we do.

[rsc]

Comment 9:

I think we may end up with the building of cmd/cgo in the build graph multiple times.
I started on this last night.

Owner changed to @rsc.

Status changed to Started.

[rsc]

Comment 10:

This issue was closed by revision b03a5f6.

Status changed to Fixed.

[gopherbot]

Comment 11 by giacomo.tartari:

Hi,
this still happens to me on a rocks cluster (CentOS):
fork/exec /home/giacomo/go/pkg/tool/linux_amd64/cgo: text file busy
Also this during the tests:
ok      old/template    0.020s
ok      os  0.040s
--- FAIL: TestExtraFiles (0.10 seconds)
    exec_test.go:158: Something already leaked - closed fd 3
    exec_test.go:199: CombinedOutput: exit status 1; output "leaked parent file. fd = 12; want 9\n"
FAIL
FAIL    os/exec 0.166s
ok      os/signal   0.005s
ok      os/user 0.006s
Is it the same problem?
$ uname -rsv
Linux 2.6.18-194.17.4.el5 #1 SMP Mon Oct 25 15:50:53 EDT 2010
$ go version
go version weekly.2012-03-04 +f4470a54e6db

[rsc]

Comment 12:

Status changed to Accepted.

[gopherbot]

Comment 13 by giacomo.tartari:

some more info on the machine, just in case:
$ cat /proc/version
Linux version 2.6.18-194.17.4.el5 ([email protected]) (gcc version 4.1.2
20080704 (Red Hat 4.1.2-48)) #1 SMP Mon Oct 25 15:50:53 EDT 2010
$ cat /etc/issue
CentOS release 5.5 (Final)
Kernel \r on an \m

[rsc]

Comment 14:

If you can reproduce this reliably, could you change
the go commands in make.bash and run.bash to say
go install -x and go test -x (add the -x part) and then
run
all.bash 2>&1 | tee all.log
If you can capture a failure, please attach all.log to this
bug report.  I would like to understand why this is happening,
and the -x will include all the commands being run.
Thanks.

[gopherbot]

Comment 15 by giacomo.tartari:

If I missed any -x flags let me know.

Attachments:

1. all.log (207925 bytes)

[rsc]

Comment 16:

Thanks.  The logging looks great but that looks like a log of a successful run.
Can you try to get a log of a failure (with the fork/exec message)?

[gopherbot]

Comment 17 by giacomo.tartari:

That seems to happen randomly. I did a few more builds hg updating back and forth
between tip and weekly and could not reproduce it.
But the os/exec test always failed on this machine.
ok      os  0.019s
--- FAIL: TestExtraFiles (0.10 seconds)
    exec_test.go:158: Something already leaked - closed fd 3
    exec_test.go:158: Something already leaked - closed fd 8
    exec_test.go:158: Something already leaked - closed fd 14
    exec_test.go:158: Something already leaked - closed fd 16
    exec_test.go:199: CombinedOutput: exit status 1; output "leaked parent file. fd = 13; want 10\n"
FAIL
FAIL    os/exec 0.168s
ok      os/signal   0.005s

[rsc]

Comment 18:

TestExtraFiles is a different issue.  (Maybe you have a virus checker or some other
thing forcing itself into your program's address space and leaking files?)

[minux]

Comment 19:

I can confirm TestExtraFiles always fail on CentOS 5/RHEL 5, see, for example, a log
here:
http://pastebin.com/xF7cN5Xw

[rsc]

Comment 20:

This issue is about cgo.
Please move discussion of TestExtraFiles elsewhere (it's just CentOS
ignoring O_CLOEXEC).

[rsc]

Comment 21:

I think I found the problem.  5845044.

Status changed to Started.

[minux]

Comment 22:

This issue was closed by revision a4b2c5e.

Status changed to Fixed.