GATK4_MARKDUPLICATES not respecting --TMP_DIR

[alexnater]

Description of the bug

There seems to be an issue with GATK's MarkDuplicates and probably other Picard tools in GATK, since they don't seem to fully respect the --TMP_DIR parameter. The tool still writes a folder CSPI.tmpXXXXXX to /tmp, causing problems on our HPC system. The solution is to set --java-options "-Djava.io.tmpdir=.", even though this should be equivalent to setting --TMP_DIR according to the documentation. This can't be done via a custom config, since it's an argument to gatk rather than MarkDuplicates, so it would be good to have that included in the GATK4_MARKDUPLICATES module.

Command used and terminal output

Relevant files

No response

System information

No response

[alexnater]

I found a workaround by setting java.io.tmpdir via _JAVA_OPTIONS in process.containerOptions. So this problem should be solved for us.