fastqc is out of RAM

[naumenko-sa]

Hi!

In some rare cases 250m RAM is not enough for fastqc:

[2019-10-22T18:21Z] Started analysis of sort-downsample.bam
[2019-10-22T18:21Z] Exception in thread "Thread-1" java.lang.OutOfMemoryError: GC overhead limit exceeded
[2019-10-22T18:21Z] 	at java.lang.String.toCharArray(String.java:2899)
[2019-10-22T18:21Z] 	at uk.ac.babraham.FastQC.Modules.PerSequenceGCContent.truncateSequence(PerSequenceGCContent.java:222)
[2019-10-22T18:21Z] 	at uk.ac.babraham.FastQC.Modules.PerSequenceGCContent.processSequence(PerSequenceGCContent.java:174)
[2019-10-22T18:21Z] 	at uk.ac.babraham.FastQC.Analysis.AnalysisRunner.run(AnalysisRunner.java:88)
[2019-10-22T18:21Z] 	at java.lang.Thread.run(Thread.java:748)
[2019-10-22T18:21Z] Uncaught exception occurred

Running fastqc from the command line with -t 2 allocates 500M and works:
s-andrews/FastQC#24

It looks like we don't have an options to control fastqc threads or memory from bcbio.yaml.

bcbio-nextgen/bcbio/qc/fastqc.py

Line 47 in 987656b

"-t", str(num_cores), "--extract", "-o", tx_tmp_dir, "-f", frmt, bam_file]

And we need it.

SN

[roryk]

Thanks, what was the FastQC command line that failed? I think we can add an explicit -Xmx500m to the FastQC call to give it more memory.

[naumenko-sa]

The VM got killed already. It was a regular fastqc command.

I think fastqc is a perl script that does not allow to set up java memory. Instead it is doing hardcoded mem=threads*250M

https://github.com/s-andrews/FastQC/blob/master/fastqc#L166

So, I will just use 2 threads as suggested in here s-andrews/FastQC#24

[roryk]

Ah, interesting!

[naumenko-sa]

In the default bcbio installation in bcbio/galaxy/bcbio_system.yaml we have

resources:
  default:
    cores: 16

so fastqc is being called as fastqc -t 16 ... and we never face the fastqc memory issue.

However, in some installations people do bcbio/galaxy/bcbio_system.yaml:

resources:
  default:
    cores: 1

So default is 1, and fastqc call is fastqc -t 1 ... - memory issue.

Reminding myself how cores work for a multicore job project.
There are:

• bcbio_nextgen.py -n NUMCORES (default = 1, typically 20-40)
• bcbio/galaxy/bcbio_system.yaml: resources/default/cores (default=16)
• project/config/bcbio.yaml: resources/default/cores

Probably, the easiest way to tackle this situation is to avoid setting cores: 1 in a system-wide config, min cores: 2.

SN

[chapmanb]

Sergey;
Great catch on the underlying issue. I pushed a small fix that floors the number of threads used at 2 to avoid hitting the low memory issues even if folks set cores: 1, so that it'll work in these cases as well. Hope this avoids the problems going forward.

[naumenko-sa]

Thanks, @chapmanb !