search

Hoohm / dropSeqPipe

A SingleCell RNASeq pre-processing snakemake workflow
Creative Commons Attribution Share Alike 4.0 International
147 stars 47 forks source link

STAR: terminate called after throwing an instance of 'std::bad_alloc' #17

Closed pwl closed 6 years ago

pwl commented 6 years ago

I encountered this issue https://github.com/alexdobin/STAR/issues/103 when running dropSeqPipe in the meta mode. Would it be possible to add the --genomeChrBinNbits parameter to your pipeline? After running the same STAR command but with --genomeChrBinNbits 15 everything went fine.

pwl commented 6 years ago

Also, it's not a big deal but --runThreadN option ignores the --cores value when running in the meta mode https://github.com/Hoohm/dropSeqPipe/blob/c8e70ce39cf8ce46638e85906b43fb479e2688e3/rules/generate_meta.smk#L85

EDIT: now I see how it works (allocating tasks according to how many threads they are defined to use). Please ignore that comment:-)

Hoohm commented 6 years ago

Hello, I can expose this value, no problem. It will be done in the next release (maybe 1 or 2 weeks). In the meantime you can easily edit it manually in the rule.

Concerning the number of threads = 4 for the INDEX generation, I ran into performance issue when using a high number of threads a couple of months ago, hence the limit to 4. Maybe this has been fixed now in STAR. I'll test it out and maybe change it to a higher value.

Hoohm commented 6 years ago

Hello @pwl I changed it up a bit, in the next version will calculate the value with the proposed function in the issue.

pwl commented 6 years ago

@Hoohm thanks for your comments.

So far nothing changed in the generate_meta.smk, I guess you haven't pushed the changes yet?

I tested STAR with more threads (8) and it seems to run just fine but that's just my limited experience.

Hoohm commented 6 years ago

@pwl That is right, haven't pushed anything yet. I'm gonna test STAR again later, but that's not gonna be a priority and I guess it's not much a bottleneck right now since it only runs once per read size.

pwl commented 6 years ago

Nah, the index generation runs fine with 4 threads, so that's not critical by any means. I'll keep an eye on the new releases, currently I just keep a modified generate_meta.smk locally.

Oh, and by the way, I saw the changelog file and there are release numbers there but there are no git tags to reflect these releases (aside from v0.24). I know this is an unrelated issue but could you please bring back these tags with the upcoming releases?

Hoohm commented 6 years ago

Yea, I have to force myself to make releases. It's planned for the next release, 0.31

pwl commented 6 years ago

Thanks for all the help, I think I can close this issue now.