Question about inconsistent peak memory reporting

[tnmquann]

Hi,

First of all, thank you for developing YACHT - it has been working very well in my workflow, and the results have been consistent across samples.

While integrating YACHT into a Nextflow pipeline, I noticed an inconsistency in memory usage reporting between Nextflow trace reports and manual execution using the Unix time command. Specifically, some YACHT tasks in the Nextflow trace report show very high peak memory usage (peak_rss), reaching up to ~341 GB. This seems unrealistic given that I configured the tasks with a maximum of 120 GB RAM, and my cluster only supports up to ~125 GB per node.

To double-check, I reran the same commands manually using time with the following settings:
time -a -f "%U,%S,%E,%M,%P"
The observed memory usage appeared reasonable and aligned with the cluster limits. This makes me think that YACHT itself is likely behaving correctly, and the discrepancy might come from how memory usage is being reported rather than actual overconsumption.

For comparison, other tools in the same pipeline, such as sourmash and fastp, show consistent memory reporting between Nextflow trace and manual runs. This makes the issue seem specific to YACHT when executed within Nextflow.

I’m wondering whether this could be related to how YACHT manages subprocesses or aggregates memory usage internally, or if it’s more likely due to how Nextflow tracks peak_rss for this particular tool.

I’ve attached the raw Nextflow trace report and the corresponding time command outputs in case it helps with a more detailed investigation.

time_trace.txt
nextflow_trace.txt

Thanks again for your great work, and I look forward to your thoughts!