v1.4 segfault consensus

[ptrebert]

Hi,
I have a single case of a segfault in the consensus phase:

  21699 Failed with 'Aborted'; backtrace (libbacktrace):
  21700  - ALIGNED 0.0238 at    133292-142920   
  21701 ERROR:  alignment from 133293 to 142920, but tiglen is only 132087
  21702 utgcns: utgcns/unitigConsensus.C:867: bool alignEdLib(dagAlignment&, tgPosition&, char*, uint32, char*, uint32, double, bool): Assertion `aln.end <= tiglen' failed.
  21703 
  21704 Failed with 'Aborted'; backtrace (libbacktrace):
  21705 
  21706 Failed with 'Segmentation fault'; backtrace (libbacktrace):
  21707 ./packages/part026.sh: line 19: 135788 Segmentation fault      /[...]/verkko/wd/.snakemake/conda/c410c39ac1cf6bea6ae75
  21707 0999a5ce20f_/lib/verkko/bin/utgcns -V -V -V -threads 8 -import ../7-consensus/packages/part026.cnspack -A ../7-consensus/packages/part026.fasta.WORKING -C 2 -norealign -max
  21707 coverage 50 -e 0.05 -em 0.20 -EM 0 -l 3000 -edlib
  21708 
  21709 Consensus did not finish successfully, exit code 0.

It's v1.4 plus two bug fixes (3119b39 and 4f6a54e). As a side note, the overall run was registered by Snakemake as having failed, despite the weird did not finish successfully, exit code 0.

Best,
Peter

[brianwalenz]

ERROR: alignment from 133293 to 142920, but tiglen is only 132087

It didn't just align incorrectly, it aligned to sequence that doesn't even exist. Neat trick, that!

Can you upload 7-consensus/packages/part026.cnspack so I can debug?

[ptrebert]

I can, but immediately only via Globus (same as for my other tickets). Does that work as well?

[brianwalenz]

That works.

[skoren]

To confirm, is this a Hi-C (or any non-trio) run? I think it might be related to b5d87b0 where we didn't tell consensus which reads are ONT in that case, leading to slower and potentially worse consensus. Can you see how many ONT reads you have in this case being used for consensus, there should be a ont_subset.id file in one of the 7-consensus folders.

[ptrebert]

Yes, this is a consensus run based on phasing paths derived from Strand-seq. The count of entries in the ont subset fasta is 6513.

[skoren]

OK, I confirmed if you change the -EM 0 to -EM 6513, this will complete w/o error. I'd suggest updating all the consensus shell scripts with that change and re-running them (which is what the bug fix above does) and you should be able to complete the asm.

[ptrebert]

Ok. Does that bug also affect the consensus in case of an unphased assembly run (no trio, no phasing paths)?

[skoren]

No, only those that use the --paths option.

[ptrebert]

that is good to know. I will apply b5d87b0 to my Verkko installation as well and rerun the consensus for my assemblies.

[ptrebert]

P.S.: any plans for a v1.4 bugfix release yet?

[skoren]

Yeah we plan to have one this month, I was waiting to see if #173 can also be fixed.