Hello All! I'm afraid I'm stuck with one of my packages. It fails to build on i686 in Rawhide but builds fine if koji was started with --arch-override=i686 (no other builds in parallel).
I suppose I run into some "limits exceeded" issue. Unfortunately all I've got is the segfault during the documentation build.
Technically I could disable docs on i686, but I really don't want to. Could someone take a look inside - what's going on there?
Here is my latest build attempt:
http://koji.fedoraproject.org/koji/taskinfo?taskID=12563046
On Fri, 15 Jan 2016 16:14:03 +0100 Peter Lemenkov lemenkov@gmail.com wrote:
Hello All! I'm afraid I'm stuck with one of my packages. It fails to build on i686 in Rawhide but builds fine if koji was started with --arch-override=i686 (no other builds in parallel).
I suppose I run into some "limits exceeded" issue. Unfortunately all I've got is the segfault during the documentation build.
hmm, OOM killer in kernel in action? because both i686 and x86_64 tasks have used buildhw-11.phx2.fedoraproject.org
Technically I could disable docs on i686, but I really don't want to. Could someone take a look inside - what's going on there?
Here is my latest build attempt:
Dan
On Fri, 15 Jan 2016 16:23:54 +0100 Dan Horák dan@danny.cz wrote:
On Fri, 15 Jan 2016 16:14:03 +0100 Peter Lemenkov lemenkov@gmail.com wrote:
Hello All! I'm afraid I'm stuck with one of my packages. It fails to build on i686 in Rawhide but builds fine if koji was started with --arch-override=i686 (no other builds in parallel).
I suppose I run into some "limits exceeded" issue. Unfortunately all I've got is the segfault during the documentation build.
hmm, OOM killer in kernel in action? because both i686 and x86_64 tasks have used buildhw-11.phx2.fedoraproject.org
Technically I could disable docs on i686, but I really don't want to. Could someone take a look inside - what's going on there?
Here is my latest build attempt:
Doesn't seem to be OOM to me... in the i686 build.log:
/bin/sh: line 6: 18505 Segmentation fault (core dumped) erl -boot start_clean -noshell -pa ebin -s erl_html_tools top_index src /builddir/build/BUILD/otp-OTP-18.2.2 ../html 18 -s erlang halt
Is the difference between a working and failed i686 build if it gets a buildvm or buildhw?
kevin
2016-01-16 18:49 GMT+01:00 Kevin Fenzi kevin@scrye.com:
On Fri, 15 Jan 2016 16:23:54 +0100 Dan Horák dan@danny.cz wrote:
On Fri, 15 Jan 2016 16:14:03 +0100 Peter Lemenkov lemenkov@gmail.com wrote:
Hello All! I'm afraid I'm stuck with one of my packages. It fails to build on i686 in Rawhide but builds fine if koji was started with --arch-override=i686 (no other builds in parallel).
I suppose I run into some "limits exceeded" issue. Unfortunately all I've got is the segfault during the documentation build.
hmm, OOM killer in kernel in action? because both i686 and x86_64 tasks have used buildhw-11.phx2.fedoraproject.org
Technically I could disable docs on i686, but I really don't want to. Could someone take a look inside - what's going on there?
Here is my latest build attempt:
Doesn't seem to be OOM to me... in the i686 build.log:
/bin/sh: line 6: 18505 Segmentation fault (core dumped) erl -boot start_clean -noshell -pa ebin -s erl_html_tools top_index src /builddir/build/BUILD/otp-OTP-18.2.2 ../html 18 -s erlang halt
Is the difference between a working and failed i686 build if it gets a buildvm or buildhw?
The only difference was --arch-override=i686 for a successful build.
This was an out-of-memory issue. I've changed Java environmental variables from -Xmx1024m to -Xmx512m and it builds fine now!
http://koji.fedoraproject.org/koji/taskinfo?taskID=12578827
Although I think it's safe to say that the issue is fixed now (workaround applied). I'd like to have more details about Koji jobs.
Also it would be great if different build tasks won't interfere with each other. I suppose that the limits should be applied to each arch-build separately rather than for the entire task.
On Sat, 16 Jan 2016 21:21:40 +0100 Peter Lemenkov lemenkov@gmail.com wrote:
The only difference was --arch-override=i686 for a successful build.
So that was also a scratch build vs a real one?
This was an out-of-memory issue. I've changed Java environmental variables from -Xmx1024m to -Xmx512m and it builds fine now!
http://koji.fedoraproject.org/koji/taskinfo?taskID=12578827
Although I think it's safe to say that the issue is fixed now (workaround applied). I'd like to have more details about Koji jobs.
Also it would be great if different build tasks won't interfere with each other. I suppose that the limits should be applied to each arch-build separately rather than for the entire task.
They should only apply to each task. Each task is done on whatever builders are available. Its pretty rare that multiple tasks from the same build would end up on the same builder (but obviously not impossible). Each task is done in it's own mock chroot, so settings in one shouldn't matter for settings in another.
The only thing I can think of here is somehow the failed builds always landed on buildhw vs buildvm or something.
kevin
2016-01-16 21:51 GMT+01:00 Kevin Fenzi kevin@scrye.com:
On Sat, 16 Jan 2016 21:21:40 +0100 Peter Lemenkov lemenkov@gmail.com wrote:
The only difference was --arch-override=i686 for a successful build.
So that was also a scratch build vs a real one?
No. If I start scratchbuild w/o arch-override it also fails. if I add --arch-override it works fine.