Recurrently getting "Kmer matching step died"

[apcamargo]

I'm recurrently getting the "Kmer matching step died" error during my assemblies. I wasn't able to pinpoint what may be causing it because I sometimes get it and sometimes don't even when I'm assembling the same data and using the same parameters.

PLASS Version: a983491

Ubuntu 18.04, 184 GB of memory, 96 × Intel(R) Xeon(R) Platinum 8275CL CPU @ 3.00GHz

strict_assembly.log

[martin-steinegger]

Do you use a grid engine and the job might have timed out? Your log outputs "killed" after the 6th iteration. So the OS or the grid engine might have killed the job.

[milot-mirdita]

Another idea: Could you try to run the assembly with less threads?
I think we assume about 4-8GB RAM per thread, at least that are the conditions in which we developed our software.

[apcamargo]

Thank you for the answers!

I'm not using a grid engine. This execution was performed in an AWS instance.

I'll try using less threads in my next assemblies. In any case, I could finish this assembly by starting it again in the same server (I needed three additional executions after this first one).

[apcamargo]

@milot-mirdita I reduced the number of cores so that I had ~5GB of memory per thread, but the problem persisted. I got the same error across multiple tries.

plass_assembly.log

[apcamargo]

Something else that I've noticed: After finishing the assembly of a sample I started assembling the second one. The "generate k-mers list" step was very slow and eventually died. After deleting the PLASS directory and re-compiling it, the "generate k-mers list" of the first step was quick again, until eventually getting slow and dying in the 6th step.

I noticed this behavior several times already. After re-compiling PLASS gets fast again and doesn't die until it has processed several steps.

[martin-steinegger]

@apcamargo I really do not know whats the issues. Some times it runs only one iteration and then six. Is it possible to share the data? I would try to reproduce the issues.

[apcamargo]

Sure @martin-steinegger! I'll email the data to you.