[maker-devel] (no subject)

[Panos Ioannidis]

Thanks for the help guys. All really helpful!

Unfortunately, full Perl reinstall is not an option at this moment on this
machine, since it's used by others in our group and they need it running...
I'll try to find another solution with our admin.


On Tue, Jul 15, 2014 at 4:10 PM, Carson Holt <carsonhh at gmail.com> wrote:

> If you don't have MPI support, it's not an issue, and  your Seg fault is
> likely something else.  Your reference to perl 5.18 and forks.pm should
> not be a segfault error either, and would not represent your error. The
> Perl 5.18/forks.pm is a different issue where perl actually tells itself
> to die because hash reshuffling isn't safe whereas segfaults are causes by
> binary corruption or incorrect memory access issues (very different
> issues). I'd actually recommend a full perl reinstall if you are getting
> segfaults, because it suggests a deeper issue.
>
> --Carson
>
>
> From: Panos Ioannidis <panos.ioannidis at gmail.com>
> Date: Tuesday, July 15, 2014 at 8:03 AM
>
> To: Carson Holt <carsonhh at gmail.com>
> Cc: Daniel Ence <dence at genetics.utah.edu>, maker-devel <
> maker-devel at yandell-lab.org>
> Subject: Re: [maker-devel] (no subject)
>
> Carson, many thanks for the info!
>
> I haven't installed Maker with MPI support. Is this segfault only
> occurring when you install it with MPI support?
>
>
> On Tue, Jul 15, 2014 at 3:58 PM, Carson Holt <carsonhh at gmail.com> wrote:
>
>> If you are getting a segfault.  It is more likely an MPI error especially
>> if you are using OpenMPI or MVAPICH2.  They both use OpenFabrics libraries
>> that have bugs on forks and system calls.
>>
>> If it is OpenMPI, run the following command before launching MAKER -->
>> export LD_PRELOAD=Š/openmpi_location/lib/libmpi.so
>>
>> Make sure to set replace openmpi_location with the location of your
>> OpenMPI.
>>
>> Also add the following to your MPI command before running MAKER.
>> --> -mca btl ^openib
>> Example --> mpiexec -mca btl ^openib -n 40 maker
>>
>>
>> If you are using MVAPICH2, then you need to switch to OpenMPI.
>>
>> --Carson
>>
>>
>>
>>
>> From: Panos Ioannidis <panos.ioannidis at gmail.com>
>> Date: Tuesday, July 15, 2014 at 12:59 AM
>> To: Carson Holt <carsonhh at gmail.com>
>> Cc: Daniel Ence <dence at genetics.utah.edu>, maker-devel <
>> maker-devel at yandell-lab.org>
>>
>> Subject: Re: [maker-devel] (no subject)
>>
>> I didn't know there are more than one forks.pm files! We'll give it
>> another try later today.
>>
>> As for the error, it's just "Segmentation fault"! And we know this
>> segfault is because of forks.pm, because if you remove the "use forks;"
>> line script execution continues without segfault (till it crashes later for
>> another reason, of course). In fact, even if you create a script with just
>> the line "use forks;" and try to run it, you'll get a segfault. So it looks
>> like it's something pretty general and serious, and I'm really surprised I
>> can't find anything by googling (except your fix!)...
>>
>>
>> On Mon, Jul 14, 2014 at 4:49 PM, Carson Holt <carsonhh at gmail.com> wrote:
>>
>>> Also one more question.  What is the exact error text you get for the
>>> forks error?  Is it a forks.pm error or is it an MPI warn on fork error
>>> (which are actually very different).
>>>
>>> --Carson
>>>
>>>
>>> From: Carson Holt <carsonhh at gmail.com>
>>> Date: Monday, July 14, 2014 at 8:46 AM
>>> To: Panos Ioannidis <panos.ioannidis at gmail.com>, Daniel Ence <
>>> dence at genetics.utah.edu>
>>>
>>> Cc: maker-devel <maker-devel at yandell-lab.org>
>>> Subject: Re: [maker-devel] (no subject)
>>>
>>> If you do the BLAST's yourself the results could be dramatically worse.
>>>  The filtering and polishing done by MAKER is rather significant (direct
>>> BLAST is actually worse with homology searches than many people realize).
>>>
>>> With respect to forks.pm, your admin most likely edited the wrong
>>> forks.pm.  There may be more than one on your system. If you let maker
>>> install some prerequisites for you (because it requires a specific version
>>> of forks.pm), it may be in .../maker/perl/lib/forks.pm.  Otherwise you
>>> have to identify the exact location of the forks.pm being used. Or if
>>> he is editing it as part of the install tarball, his edits may actually be
>>> undone during the installation procedure.
>>>
>>> Use this command line to identify the location of the forks.pm module
>>> that would have to be edited -->
>>> maker --debug 2>&1 | grep "forks.pm"
>>>
>>> You can even send me a copy of the file once it has been edited, and I
>>> can tell you if it was done correctly.
>>>
>>> --Carson
>>>
>>>
>>>
>>>
>>> From: Panos Ioannidis <panos.ioannidis at gmail.com>
>>> Date: Monday, July 14, 2014 at 1:20 AM
>>> To: Daniel Ence <dence at genetics.utah.edu>
>>> Cc: maker-devel <maker-devel at yandell-lab.org>
>>> Subject: Re: [maker-devel] (no subject)
>>>
>>> Daniel, thanks for the info.
>>>
>>> Regarding (3), the only reason I think of running BLASTs separately is
>>> because I'm currently not able to run Maker on our cluster due to a problem
>>> in the Perl "forks" library. And it looks like there isn't much I can do
>>> about it; I tried Perlbrew but it doesn't work when I try to install
>>> versions <5.18 (the problem in forks occurs on 5.18 and later versions).
>>> Our admin also tried to change the code in the forks.pm file as per
>>> Carson's suggestion in another thread, but that didn't work either... As a
>>> result I'm running Maker on my workstation (really slooow) till a solution
>>> is found and since BLAST is a time-consuming step I was thinking of running
>>> it separately.
>>>
>>>
>>> On Fri, Jul 11, 2014 at 4:08 PM, Daniel Ence <dence at genetics.utah.edu>
>>> wrote:
>>>
>>>> Hi Panos,
>>>>
>>>> 1) You'll only use est2genome and protein2genome for creating models
>>>> that will be used for training the ab-initio predictors (like SNAP).
>>>> Sometimes that means one run of MAKER for training; sometimes that means
>>>> two runs of MAKER. You usually don't gain any accuracy after the second
>>>> round of training. It's ok to use both EST and protein data for this
>>>> training step.
>>>>
>>>> 2) If you're using both ESTs and protein sequence to train your
>>>> ab-initio predictors, then both est2genome and protein2genome should be set
>>>> to 1.
>>>>
>>>> 3) If you want to pass Blast results to MAKER, you'll need to pass
>>>> those results as GFF3, but MAKER will install and run blast for you, and
>>>> does a good job of keeping track of all those results and making them
>>>> accessible to you in the end, so it's going to be a lot of work to do those
>>>> blasts on your own outside of MAKER. I seriously suggest that you use blast
>>>> internal to maker.
>>>>
>>>> Daniel Ence
>>>> Graduate Student
>>>> Eccles Institute of Human Genetics
>>>> University of Utah
>>>> 15 North 2030 East, Room 2100
>>>> Salt Lake City, UT 84112-5330
>>>> ------------------------------
>>>> *From:* maker-devel [maker-devel-bounces at yandell-lab.org] on behalf of
>>>> Panos Ioannidis [panos.ioannidis at gmail.com]
>>>> *Sent:* Friday, July 11, 2014 5:56 AM
>>>> *To:* maker-devel
>>>> *Subject:* [maker-devel] (no subject)
>>>>
>>>> I got back to my annotations this past week and have a couple of
>>>> questions!
>>>>
>>>> 1) Since my organism isn't closely related with any other that's
>>>> already sequenced, I will have to run maker twice (according to the
>>>> tutorial). So for the first run I see that some people use only the ESTs
>>>> and some others use ESTs and a protein database (CEGMA, Uniref50,
>>>> Swiss-Prot, etc). I guess that the ESTs will give better models, but for
>>>> the cases where genes aren't covered by an EST, it's okay to have a protein
>>>> database to detect them as well. Am I right? What do you think?
>>>>
>>>> 2) In case I use both ESTs and a protein database how should I set the
>>>> est2genome and protein2genome parameters in the maker_opts.ctl file?
>>>> Should they both equal to "1"?
>>>>
>>>> 3) I've been thinking of running the BLAST searches separately and
>>>> giving Maker directly the results. I guess that in this case, I'll have to
>>>> first convert the BLAST output to a gff3 file and give it to the
>>>> protein_gff parameter, right?
>>>>
>>>> Thanks,
>>>> Panos
>>>>
>>>
>>> _______________________________________________ maker-devel mailing list
>>> maker-devel at box290.bluehost.com
>>> http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org
>>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://yandell-lab.org/pipermail/maker-devel_yandell-lab.org/attachments/20140716/75cae652/attachment-0003.html>