search

rwdavies / STITCH

STITCH - Sequencing To Imputation Through Constructing Haplotypes
http://www.nature.com/ng/journal/v48/n8/abs/ng.3594.html
GNU General Public License v3.0
74 stars 19 forks source link

Error when using specific number of cores #35

Closed fbartusch closed 4 years ago

fbartusch commented 4 years ago

Hello,

Beside finding out good working combinations of nCores and memory on our cluster, we experienced a problem when using Stitch 1.6.3 with 20 cores, K=20.

[...]
[2020-04-05 11:23:20] Start EM
[2020-04-05 11:23:20] Number of samples: 397
[2020-04-05 11:23:20] Number of SNPs: 755291
[2020-04-05 11:23:20] Start of iteration 1
Error in priorSum_m[, s] : incorrect number of dimensions
Calls: STITCH -> completeSampleIteration -> calculate_updates
In addition: Warning message:
In mclapply(sampleRanges, mc.cores = nCores, FUN = subset_of_complete_iteration,  :
  scheduled cores 3, 4, 5, 8, 9, 11, 13, 16, 17, 19, 20 did not deliver results, all values of the jobs will be affected
Execution halted

Do you know if stitch works just with a specific number of cores?

Kind regards, Felix

rwdavies commented 4 years ago

Can you check and confirm it didn't run out of RAM while running? Out of RAM in R parallel computing often crashes in inelegant ways

fbartusch commented 4 years ago

Indeed, the job used more than the requested 90gb memory. We will try to reduce the memory footprint using suggestions in the README and other issues.