Re: [MooseFS-Users] Strange problems compiling on a moosefs share.

[Matt W. <mat...@gm...>]

Thank you for looking into this. It might be good to first build in /tmp
for a reference baseline. If I run Megatest from a location where it was
compiled before the upgrade to Moosefs 3.0.81 it runs fine. The problem
seems to happen during the compile.

How to reproduce:

1. Install chicken + IUP gui (alternative: use Makefile.installall from
utils dir in Megatest)

wget http://www.kiatoa.com/matt/chicken_4.11rc2_x86.tar.gz
/tmp/chicken.tar.gz
cd /opt;tar xf /tmp/chicken.tar.gz
source /opt/chicken/4.11rc2_x86_64/setup-chicken4x.sh

2. Get Megatest source for v1.61 (project is at
https://www.kiatoa.com/fossils/megatest)

 wget http://www.kiatoa.com/matt/megatest_v1.61_src.tar.gz
/tmp/megatest.tar.gz
tar xf /tmp/megatest.tar.gz
cd megatest
make clean;make -j && make install
cd tests
source fixpath.sh
make dashboard
========== results in ==========
cd fullrun && /mfs/tmp/megatest/bin/dashboard -skip-version-check -rows 20 &
matt@xena:/mfs/tmp/megatest/tests$
Error: segmentation violation

    Call history:

    dcommon.scm:68: hash-table-set!
    tree.scm:12: ##sys#require
    tree.scm:13: ##sys#require
    tree.scm:15: ##sys#require
    tree.scm:17: ##sys#require
    tree.scm:17: ##sys#require
    tree.scm:17: ##sys#require
    altdb.scm:2: make-hash-table
    altdb.scm:3: ##sys#require
    altdb.scm:3: hash-table-set!
    vg.scm:13: ##sys#require
    vg.scm:16: ##sys#require
    vg.scm:16: ##sys#require
    vg_records.scm:4: ##sys#require
    vg_records.scm:5: make-exception
    vg_records.scm:17: ##sys#require          <--

If I compile Megatest v1.62 it also fails but I get a different signature:

matt@xena:/mfs/tmp/megatest/tests$ [panic] Detected corrupted data in stack
- execution terminated

...more...
altdb.scm:3: hash-table-set!
db.scm:19: ##sys#require
db.scm:19: ##sys#require
db.scm:19: ##sys#require
db.scm:19: ##sys#require
db.scm:19: ##sys#require
db.scm:19: ##sys#require
db.scm:19: ##sys#require
db.scm:19: ##sys#require
db.scm:19: ##sys#require
db.scm:19: ##sys#require
db.scm:19: ##sys#require
altdb.scm:2: make-hash-table
altdb.scm:3: ##sys#require
altdb.scm:3: hash-table-set!
db.scm:35: make-mutex     <--


On Mon, Sep 19, 2016 at 2:31 AM, Jakub Kruszona-Zawadzki <
jak...@ge...> wrote:

>
> On 18 Sep, 2016, at 2:34, Matt Welland <mat...@gm...> wrote:
>
> Previous (rock solid) setup:
>
> Moosefs 2.0 installed via apt all on 64bit x86.
>
> Current (problematic) setup:
>
> Moosefs 3.0.81 installed from source using "dpkg-buildpackage -b"
>
> Master is on a cubietruck running Debian Jessie. chunkservers are on x86.
>
> The problem:
>
> Compiling a Chicken Scheme program results in the program crashing.
> Compiling the program on local disk works fine.
>
>
> We need to reproduce this behaviour on our instance. Could you write more?
> I tried simple 'hello world' and it works:
>
> $ cat hello.scm
> (print "Hello, world!")
> $ csi -s hello.scm
> Hello, world!
> $ csc hello.scm
> $ ./hello
> Hello, world!
>
> I've also downloaded some benchmarks from:
>
> https://github.com/ashinn/chibi-scheme/tree/master/benchmarks/gabriel
>
> I've tested 'kanren.sch' using MooseFS 3.0.81, Ubuntu 16.04 and chicken
> scheme installed from packages (version 4.9 as I remember)
>
> core@worker6:/mnt/mfstest/scheme/chibi-scheme/benchmarks/gabriel$ csc
> kanren.sch
> core@worker6:/mnt/mfstest/scheme/chibi-scheme/benchmarks/gabriel$ ./kanren
> Testing eigen
> eigens: (!x_!$gen$!x3 !y_!$gen$!x4)
> Testing test-unify/pairs-oleg1
> Testing test-unify/pairs-oleg2
> (...) <--- long output
> Testing logo-3-3
> Testing powers-of-3
> Testing powers-of-exp-3
> 2.7s CPU time, 0.432s GC time (major), 4601 mutations, 202/5439 GCs
> (major/minor)
>
> It looks correct.
>
> I've also checked it on Debian Jessie (64bit) and it also works fine.
>
>
> Maybe we need something more sophisticated. Could you send us something
> that triggers this problem?
>
>
>
>
> I also see these messages in syslog:
>
> [11622.232964] mfsmount (13759): /proc/13759/oom_adj is deprecated, please
> use /proc/13759/oom_score_adj instead.
>
>
> It doesn't matter. I use both 'oom_adj' and 'oom_score_adj', so it works
> fine. This is only disabling 'out of memory killer', so this is not related
> to any file operations.
>
>
> Should I revert to my old setup?
> ------------------------------------------------------------
> ------------------
> _________________________________________
> moosefs-users mailing list
> moo...@li...
> https://lists.sourceforge.net/lists/listinfo/moosefs-users
>
>
> --
> Regards,
> Jakub Kruszona-Zawadzki
> - - - - - - - - - - - - - - - -
> Segmentation fault (core dumped)
> Phone: +48 602 212 039
>
>
> ------------------------------------------------------------
> ------------------
>
> _________________________________________
> moosefs-users mailing list
> moo...@li...
> https://lists.sourceforge.net/lists/listinfo/moosefs-users
>
>