Segmentation fault when running Minia v3 with some kmer values

Question

[matthew.nicotra]

Hi,

I am attempting to assemble a set of 151 bp paired end reads using Minia v3 on my university's cluster. I am using the precompiled linux binary.

I ran kmergenie on the data (~60 million reads) and it suggested setting kmer-size to 47. If I submit the job with 1 thread and 8G memory, it completes without problems. The maxRSS is 4.9G.

However, when I submit the job with 8 threads, I get a segmentation fault

Minia 3, git commit 4b32fec
setting storage type to hdf5
[Approximating frequencies of minimizers ]  100  %   elapsed:   0 min 26 sec   remaining:   0 min 0  sec   cpu: 100.0 % mem: [  29,   29,   29] MB
[DSK: nb solid kmers found : 439869125   ]  116  %   elapsed:  64 min 5  sec   remaining:   0 min 0  sec   cpu: 706.6 %   mem: [ 152, 50662, 50731] MB
bcalm_algo params, prefix:295-8.minia-k31-8c.unitigs.fa k:47 a:2 minsize:10 threads:8 mintype:1
DSK used 1 passes and 16 partitions
prior to queues allocation                      21:32:25     memory [current, maxRSS]: [ 159, 50731] MB
Starting BCALM2                                 21:32:25     memory [current, maxRSS]: [ 159, 50731] MB
[Iterating DSK partitions                ]  0    %   elapsed:   0 min 0  sec   remaining:   0 min 0  sec
Iterated 34589186 kmers, among them 2486370 were doubled
/var/spool/slurmd/job830979/slurm_script: line 11:  1446 Segmentation fault      ../minia-v3.2.0-bin-Linux/bin/minia -in 295-8_filenames -nb-cores 8 -max-memory 60000 -kmer-size 47 -out 295-8.minia-k31-8c

This has happened regardless of how much memory I allocate (here I gave it 60G).

But, if I use a smaller kmer size (say, 31), the job completes with 8 cores, no problem. For that job, the maxRSS was 22G.

Is the problem that the software runs out of memory, or something else?

Thanks,

Matt

Answer 1

[Rayan Chikhi]

Hi,

Segfault is unlikely a memory error but rather a bug in the program or something else. Do you have enough disk space? Would it be possible to share the input data?

Best, Rayan