pthread_setaffinity_np failed Error while running pepper

[LYC-vio]

Hi,

I'm trying to run HapDup on my assembly from Flye. However, an error has occurred while running Pepper:
RuntimeError: /onnxruntime_src/onnxruntime/core/platform/posix/env.cc:173 onnxruntime::{anonymous}::PosixThread::PosixThread(const char*, int, unsigned int (*)(int, Eigen::ThreadPoolInterface*), Eigen::ThreadPoolInterface*, const onnxruntime::ThreadOptions&) pthread_setaffinity_np failed, error code: 0 error msg:
there's also a warning before this runtime error:
/usr/local/lib/python3.8/dist-packages/torch/onnx/symbolic_opset9.py:2095: UserWarning: Exporting a model to ONNX with a batch_size other than 1, with a variable length with LSTM can cause an error when running the ONNX model with a different batch size. Make sure to save the model with a batch size of 1, or define the initial states (h0/c0) as inputs of the model. warnings.warn("Exporting a model to ONNX with a batch_size other than 1, " +
Do you have any idea why this happens?

The commands that I use are like:

reads=NA24385_ONT_Promethion.fastq
outdir=`pwd`
assembly=${outdir}/assembly.fasta
hapdup_sif=../HapDup/hapdup_0.4.sif

time minimap2 -ax map-ont -t 30 ${assembly} ${reads} | samtools sort -@ 4 -m 4G > assembly_lr_mapping.bam
samtools index -@ 4 assembly_lr_mapping.bam

time singularity exec --bind ${outdir} ${hapdup_sif}\
	hapdup --assembly ${assembly} --bam ${outdir}/assembly_lr_mapping.bam --out-dir ${outdir}/hapdup -t 64 --rtype ont

Thank you

[fenderglass]

@kishwarshafin any ideas?

[kishwarshafin]

@LYC-vio ,

Yes, I think you are using an older version that uses quantized by default. Can you please see if you have the option to do --no_pepper_quantized in the pipeline that you are using? Since r0.7 I turned it off by default.

[fenderglass]

@kishwarshafin the current pepper version is 0.6, and I am in the process of updating to 0.7. Thanks!

@LYC-vio I'll post here once the new release with pepper 0.7 is ready.

[fenderglass]

The new hapdup 0.5 show now work for you - it contains the updated PEPPER 0.7.

[LYC-vio]

Hi,
Sorry for the late reply. I found out that this error maybe caused by assigning less cpus than indicated by -t. After increasing the number of CPUs the error no longer appears. And then it stuck in the MODEL QUANTIZATION ENABLED step. I think you are right, something may have gone wrong with the --no_pepper_quantized, I'll try again with hapdup 0.5.

Thank you very much for your help

[LYC-vio]

Hi,

Thank you for updating to hapdup 0.5. However, I still get the MODEL QUANTIZATION ENABLED message. After checking the pepper-variant (r0.7) code I found:

    parser.add_argument(
        "--quantized",
        default=True,
        action='store_true',
        help="PEPPER: Use quantization for inference while on CPU inference mode. Speeds up inference. Default is True."
    )
    parser.add_argument(
        "--no_quantized",
        dest='quantized',
        default=False,
        action='store_false',
        help="Do not use quantization for inference while on CPU inference mode. Speeds up inference."
    )

I think the quantization was still turned on.

By the way, would you please provide me some information about why this quantization takes such a long time? (it stuck in this step for days even with 64 threads)

Thank you very much

[fenderglass]

@kishwarshafin could you take a look?

[kishwarshafin]

@fenderglass , sorry I made the no quantization default in the wrapper that calls all "PEPPER-Margin-DeepVariant". Can you please add --no_quantized to your script in the pepper's parameters? Sorry for the confusion.

[fenderglass]

@kishwarshafin thanks!

@LYC-vio please try this docker image, let me know if it fixes the issue: mkolmogo/hapdup:0.5-iss10

[LYC-vio]

@fenderglass @kishwarshafin
Thank you very much! It works smoothly now (hapdup:0.5-iss10)!

[fenderglass]

Glad to hear, thanks!