Skip to content

[Bioc-devel] Check time on macOS (MIRit)

8 messages · Jacopo Ronchi, Vincent Carey, Kasper Daniel Hansen +2 more

#
Dear developers,

I am struggling to reduce R CMD check times for my package, MIRit, on
macOS. The check on linux takes 4 minutes, while on macOS it needs 9/10
minutes. I have severely reduced the size of examples and tests, which now
use smaller datasets and cached resources.

On linux, unit tests require 30 seconds more or less, and examples 40
seconds. On the other hand, macOS tests take 100 seconds, and 75 for the
examples. The issue is that sometimes the check runs in less than 10
minutes, sometimes it needs a little bit more.

However I am out of ideas on how to further reduce examples and tests...
The vignette is really extensive, but since SPB uses the "--no-vignette"
option this should not influence heavily on this behavior, at least to my
understanding.

Sorry for bothering you,
Best regards,
Jacopo
#
Thanks Jacopo.  You shouldn't have to struggle.  We are aware of
disparities of platform
performance and are working to improve the build process.  It will take
time.  I believe that
achievement of acceptable timings on the linux platform should be the key
goal at
this time.

On Wed, Feb 7, 2024 at 5:18?AM Jacopo Ronchi <jacopo.ronchi at unimib.it>
wrote:

  
    
#
Dear Vincent, thanks for your answer. So, at least for now, can I ignore
the warning on macOS for the submission of MIRit?

Il mer 7 feb 2024, 12:26 Vincent Carey <stvjc at channing.harvard.edu> ha
scritto:
#
We need a bit of internal discussion before committing to this notion.  I
will get back to the list on this.

On Wed, Feb 7, 2024 at 6:34?AM Jacopo Ronchi <jacopo.ronchi at unimib.it>
wrote:

  
    
#
Ok, understood. Thanks

Il mer 7 feb 2024, 12:46 Vincent Carey <stvjc at channing.harvard.edu> ha
scritto:
#
To be a bit more clear and vague at the same time: currently, some of our
macOS builders exhibit weird slowdowns when the build system is run
automatically compared to running builds "by hand". This is pretty weird,
and hopefully we can figure out the cause and fix it. Specifically, builds
on kjohnson3 is (at the moment) weirdly slow.

Best,
Kasper

On Wed, Feb 7, 2024 at 7:24?AM Jacopo Ronchi <jacopo.ronchi at unimib.it>
wrote:

  
    
#
Hi Kasper,

According to
https://github.com/Bioconductor/BBS/blob/9ebb2cf20e70e4bebf24a2a824a67193de821b08/3.19/bioc-mac-arm64/kjohnson3/config.sh#L17-L20
the machine has 24 vCPUs.
The automatic/weekly build utilizes 12-16 out of those at any given time!
With the respective RAM and disk usage.

By "by hand" do you mean to run the build/check of a given package on a
"calm" machine ?
It would be expected that it will be faster if it has all/most resources
just for it.

Martin

On Wed, Feb 7, 2024 at 5:12?PM Kasper Daniel Hansen <
kasperdanielhansen at gmail.com> wrote:

            

  
  
#
Hi Martin,

A very short summary is that on kjohnson3 a build/check on package launched
by cron takes significantly longer than if one runs the same command
manually in a terminal.

There's some further discussion on this at
https://community-bioc.slack.com/archives/CEQ04GKEC/p1706816310942079 and
and Jen has some nice documentation of their investigation at
https://github.com/Bioconductor/BBS/issues/387

However since the original inquiry in this thread is about a package in
submission, and so the relevant tests are run by the single package builder
on merida1, I'm not sure it's pertinent to this particular issue.

Cheers,
Mike

On Wed, 7 Feb 2024 at 16:34, Martin Grigorov <martin.grigorov at gmail.com>
wrote: