search

Bioconductor / bioc_docker

[DEPRECATED] Docker containers for Bioconductor
https://github.com/bioconductor/bioconductor_docker
Artistic License 2.0
49 stars 27 forks source link

ddalpha is victim of OOM killer #50

Closed sneumann closed 6 years ago

sneumann commented 6 years ago

Hi, the last builds of devel_metabolomics2 are killed during the automated builds. https://hub.docker.com/r/bioconductor/devel_metabolomics2/builds/bpstlewv4gnnrvyeclryhn8/ Reason is that linking ddalpha.so consumes large (>=6 GB) of RAM, and the hub.docker build machines probably have a limit set.

...
g++ -std=gnu++11 -shared -L/usr/local/lib/R/lib -L/usr/local/lib -o ddalpha.so AlphaProcedure.o Common.o DKnn.o HD.o Knn.o Mahalanobis.o OjaDepth.o Polynomial.o PotentialDepth.o ProjectionDepth.o SimplicialDepth.o TukeyDepth.o ZonoidDepth.o asa047.o ddalpha.o depth.fd.o init.o stdafx.o -lgfortran -lm -lquadmath -L/usr/local/lib/R/lib -lR installing to /usr/local/lib/R/site-library/ddalpha/libs
byte-compile and prepare package for lazy loading
Killed

Locally the devel_metabolomics2 builds just fine.

Ideas ?

Yours, Steffen

lshep commented 6 years ago

I think it is a limitation on dockerhub itself - I haven't found much besides this post on stackoverflow

https://stackoverflow.com/questions/31158913/is-there-any-limit-on-pull-number-in-docker-hub

sneumann commented 6 years ago

Hi, I think that link is about the resulting image size.

The issue here is that the gcc link step is killed, and guess that the dockerhub build has RAM limits set, E.g. 2GB.

Yours Steffen

I blame Android for the brevity and typos

pokotylo commented 6 years ago

Hi, I have recently fixed this issue in ddalpha. Now it does not consume that much memory during byte-compilation.