Hi,
I’m getting this error while running a pipeline (gnomad_qc/subpop_analysis.py at subpops · broadinstitute/gnomad_qc · GitHub): Error summary: HailException: Hail off-heap memory exceeded maximum threshold: limit 3.70 GiB, allocated 3.70 GiB
Any suggestions on what I can try changing? Let me know if I should send a log.
Thanks!
do you have the full stack trace?
we’ve been working through a similar issue with Wenhan and may have a solution
Hi Hail Team!
I’m posting on this thread rather than a new post because I just got the same error using very similar code and the same MT, and it failed at the densify step gnomad_qc/compare_freq.py at ac4a9c2bc4dc0bcb38cd43a0e89fee801c9e6f45 · broadinstitute/gnomad_qc · GitHub.
Kristen said that her error was fixed when using Hail version: 0.2.84-2817a9851cdf
, however this was the version that I used for my test and I used the same cluster config that she did. One difference was that I added a split multi to the code.
Any thoughts on what I can change? I can send the log over email or Slack.
Thank you!
Hi @jkgoodrich,
Could you email the log to hail-team?
Yup, done. Thank you!
Closing the loop here. We were able to successfully run the pipeline here using the --off-heap-memory-fraction
argument to hailctl dataproc start
. I am continuing to investigate memory management in explode
s (used by sparse_split_multi
), since the split multi was the only difference between this failing pipeline and the one that Kristen ran sucessfully.