Fatal Python error: Segmentation fault

[biozzq]

Dear all,

Describe the issue:
I am using DeepVariant to detect genetic variants in my own data. However, it failed. Upon reviewing the log files, I noticed that, in addition to the previously mentioned warnings (#946), there is also an error message. The specific command I used is as follows:

Setup

• DeepVariant version: 1.8.0

Steps to reproduce:

• Command:

singularity exec ~/01.software/install/deepvariant_latest.sif run_deepvariant \
  --model_type=WGS \
  --ref=${ref} \
  --reads=/work/home/acu6cshe3p/data/03.mapping/${pre}.dup.sort.fix.bam \
  --output_vcf=/work/home/acu6cshe3p/data/01.reference/04.VCF/${pre}.vcf.gz \
  --output_gvcf=/work/home/acu6cshe3p/data/01.reference/03.mapping/${pre}.gvcf.gz \
  --num_shards=12 \
  --regions="/work/home/acu6cshe3p/data/01.reference/chr.bed" \
  --vcf_stats_report=true \
  --disable_small_model=false \
  --logging_dir=/work/home/acu6cshe3p/data/tmp/${pre}.logs \
  --haploid_contigs="X,Y" \
  --par_regions_bed="/work/home/acu6cshe3p/data/01.reference/GRCh38.PAR.bed" \
  --dry_run=false \
  --intermediate_results_dir /work/home/acu6cshe3p/data/tmp/${pre}

• Error trace: (if applicable)
  The processed log file from the software run is attached for your reference (A18.log). Since the raw log file is too large, I have processed it using the following command grep -v "BED file does not have a tabix index. Reading full bed file".
  Could you please help me determine the exact issue based on this error?

Thank you for your assistance.

Best wishes,
Zheng zhuqing

[kishwarshafin]

@biozzq

How much memory do you have for this sample? It seems like it's running out of memory in postprocess step:

Checking the headers of 12 files.
Done, the headers are compatible.
Fatal Python error: Segmentation fault

[biozzq]

Hi @kishwarshafin

Approximately 50G. I'll try increasing it to 100G.

Best wishes,
Zheng zhuqing

[biozzq]

Despite increasing the memory allocation to 200GB, the same error persists during execution. One potential approach I am considering is parallelizing the analysis by chromosome to reduce per-job memory demands.

[biozzq]

Despite parallelizing the analysis by chromosome and increasing the memory allocation to 200GB, the same error persists during execution. Could you please help me determine the exact issue based on these information?

Thank you for your assistance.

[lucasbrambrink]

Hi Zheng,

You may have encountered an issue with parallelism in postprocess_variants that will be fixed in the next release (coming soon!). In the meantime, you can try disabling parallelism of this step by setting --postprocess_cpus 0 (see flag spec).

Hopefully that'll resolve your issue!