[CI][Crossbow] Nightly R with sanitizers build fails installing dependencies

[asfimport]

See https://circleci.com/gh/ursa-labs/crossbow/4006 for example. Failure looks like this:

...
g++ -fsanitize=address,undefined,bounds-strict -fno-omit-frame-pointer -std=gnu++11 -I"/usr/local/RDsan/lib/R/include" -DNDEBUG  -I'/usr/local/RDsan/lib/R/site-library/Rcpp/include' -I'/usr/local/RD/lib/R/library/BH/include' -I/usr/local/include  -I. -Ircon -fpic  -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2 -g -O0 -Wall -pedantic  -c TokenizerWs.cpp -o TokenizerWs.o

cc1plus: out of memory allocating 65536 bytes after a total of 31866880 bytes

cc1plus: out of memory allocating 65536 bytes after a total of 52080640 bytes
virtual memory exhausted: Cannot allocate memory
virtual memory exhausted: Cannot allocate memory
/usr/local/RDsan/lib/R/etc/Makeconf:175: recipe for target 'Collector.o' failed
make[1]: *** [Collector.o] Error 1
make[1]: *** Waiting for unfinished jobs....
/usr/local/RDsan/lib/R/etc/Makeconf:175: recipe for target 'CollectorGuess.o' failed
make[1]: *** [CollectorGuess.o] Error 1
/usr/local/RDsan/lib/R/etc/Makeconf:175: recipe for target 'Reader.o' failed
make[1]: *** [Reader.o] Error 1
/usr/local/RDsan/lib/R/etc/Makeconf:175: recipe for target 'TokenizerWs.o' failed
make[1]: *** [TokenizerWs.o] Error 1
make[1]: Leaving directory '/tmp/RtmpvSXO7m/R.INSTALL992bafde57/readr/src'
ERROR: compilation failed for package ‘readr’
* removing ‘/usr/local/RDsan/lib/R/site-library/readr’
Error: Failed to install 'decor' from GitHub:
  (converted from warning) installation of one or more packages failed,
  probably ‘readr’
Execution halted
ERROR: Service 'r-sanitizer' failed to build: The command '/bin/sh -c RDsan -e "remotes::install_github('romainfrancois/decor')"' returned a non-zero code: 1

Some thoughts:

• This dependency is not needed to run the tests--it is a developer dependency, only needed when you're editing the Rcpp code. So we could conditionally not install it if we're on CI.

• The docker code around this changed in ARROW-6918 last week, around the time this started failing. The installation of this package was moved to a different part of the job, and it appears that MAKEVARS=-j8 was added. Maybe that's causing more memory to be used concurrently? (Also, I would be shocked if CircleCI gave us a VM with 8 cores, so this seems wasteful and should probably be tuned to whatever the current system has.)

  cc @pitrou @fsaintjacques

Reporter: Neal Richardson / @nealrichardson
Assignee: Neal Richardson / @nealrichardson

PRs and other links:

• GitHub Pull Request #5986

_{Note: This issue was originally created as ARROW-6957. Please see the migration documentation for further details.}

[asfimport]

Antoine Pitrou / @pitrou:
The message "cc1plus: out of memory allocating 65536 bytes after a total of 31866880 bytes" clearly means that the VM doesn't have enough memory. Perhaps we should make the MAKEVARS configurable or something, and use a lower number on Circle CI.

If R didn't have an entirely braindead build system, it would be able to choose the optimal number of cores by itself. I chose 8 because it worked well for my machine, but of course it won't be optimal for another machine...

Perhaps a R expert can suggest a better way to choose the number of cores. You know those experts better than I do :)

 

[asfimport]

Neal Richardson / @nealrichardson:
Does docker provide a shell for the ENV command such that MAKEFLAGS=-j$(getconf _NPROCESSORS_ONLN) or similar would work (https://stackoverflow.com/a/23569003/11897522) ?

[asfimport]

Antoine Pitrou / @pitrou:
I don't know. Perhaps a Docker expert may be able to answer...

[asfimport]

Francois Saint-Jacques / @fsaintjacques:
$(nproc) should do it, it's part of coreutils very high chance it is installed. 

[asfimport]

Neal Richardson / @nealrichardson:
Issue resolved by pull request 5986
#5986