segaling_repeat_masker still crashes about segalign

[RenzoTale88]

I'm coming across the same issue using 2xA100 GPUs running segaling_repeat_masker on GCA_946052875.1 genome:

        Chromosome block 2 interval 96/100 (2950000000:2960000000) with ref (720545186:1330545186) rc (70545185:80545185)
        Chromosome block 2 interval 97/100 (2960000000:2970000000) with ref (720545186:1330545186) rc (60545185:70545185)
        Chromosome block 2 interval 98/100 (2970000000:2980000000) with ref (720545186:1330545186) rc (50545185:60545185)
        Chromosome block 2 interval 99/100 (2980000000:2990000000) with ref (720545186:1330545186) rc (40545185:50545185)
        Chromosome block 2 interval 100/100 (2990000000:2999999981) with ref (720545186:1330545186) rc (30545204:40545185)

        Sending block 3 ...
        Chromosome block 3 interval 1/4 (3000000000:3010000000) with ref (0:330545186) rc (20545185:30545185)
        Chromosome block 3 interval 2/4 (3010000000:3020000000) with ref (0:330545186) rc (10545185:20545185)
        Error: cudaMemcpy of 4 bytes for num_anchors failed with error " invalid argument "
        terminate called after throwing an instance of 'thrust::system::system_error'
          what():  CUDA free failed: cudaErrorCudartUnloading: driver shutting down

        [2023-06-02T23:43:08+0100] [MainThread] [E] [toil.worker] Exiting the worker because of a failed job on host nodeXXX
<=========

[glennhickey]

@RenzoTale88 This version of SegAlign (github.com/gsneha26/SegAlign) is broken and, sadly, doesn't seem to be maintained anymore. You may have more luck with the fork that Cactus uses, which can be found here: https://github.com/ComparativeGenomicsToolkit/SegAlign (in addition to the Cactus release GPU docker images). Even this fork has some overflow issues, but it works on most data.

[RenzoTale88]

@glennhickey thanks for the quick reply, I have been trying also with the second repository, but came across the same issue. I can open a second ticket on the cactus github page if you prefer?

[glennhickey]

Did you try the commit that cactus uses? I only just merged it into master. If you did not use that commit (or the current master as of 10 minutes ago), you would see the same error.

[RenzoTale88]

Just cloned the repo and currently compiling it. I'll come back on this, thanks!

[RenzoTale88]

@glennhickey still coming across this issue:

        Chromosome block 2 interval 100/100 (2990000000:2999999981) with ref (794006738:1424006738) rc (114006756:124006737)

        Sending block 3 ...
        Chromosome block 3 interval 2/12 (3010000000:3020000000) with ref (0:424006738) rc (94006737:104006737)
        Chromosome block 3 interval 1/12 (3000000000:3010000000) with ref (0:424006738) rc (104006737:114006737)
        Chromosome block 3 interval 4/12 (3030000000:3040000000) with ref (0:424006738) rc (74006737:84006737)
        Chromosome block 3 interval 3/12 (3020000000:3030000000) with ref (0:424006738) rc (84006737:94006737)
        Chromosome block 3 interval 5/12 (3040000000:3050000000) with ref (0:424006738) rc (64006737:74006737)
        Chromosome block 3 interval 6/12 (3050000000:3060000000) with ref (0:424006738) rc (54006737:64006737)
        Chromosome block 3 interval 7/12 (3060000000:3070000000) with ref (0:424006738) rc (44006737:54006737)
        Chromosome block 3 interval 8/12 (3070000000:3080000000) with ref (0:424006738) rc (34006737:44006737)
        Chromosome block 3 interval 9/12 (3080000000:3090000000) with ref (0:424006738) rc (24006737:34006737)
        Chromosome block 3 interval 10/12 (3090000000:3100000000) with ref (0:424006738) rc (14006737:24006737)
        Chromosome block 3 interval 11/12 (3100000000:3110000000) with ref (0:424006738) rc (4006737:14006737)
        Error: cudaMemcpy of 4 bytes for num_anchors failed with error " invalid argument "
        terminate called after throwing an instance of 'thrust::system::system_error'
          what():  CUDA free failed: cudaErrorCudartUnloading: driver shutting down

        [2023-06-06T00:58:05+0100] [MainThread] [E] [toil.worker] Exiting the worker because of a failed job on host NODE

[gsneha26]

I will get it fixed by the end of the week.