More cleanup needed for /root/.npm/_cacache

[yarikoptic]

Follow up to #1648 which was addressed by #1768 but either it was not sufficient alone or just came back (a new npm install or alike step came without desired cleanup):

(dev3) 2 5281.....................................:Mon 19 Jul 2021 10:49:31 AM EDT:.
smaug:~datalad/datalad-registry/instance/cache
$> docker pull poldracklab/fmriprep
Using default tag: latest
latest: Pulling from poldracklab/fmriprep
0a01a72a686c: Pull complete
cc899a5544da: Pull complete
19197c550755: Pull complete
716d454e56b6: Pull complete
b5bf898e214a: Pull complete
42da0942cc0e: Pull complete
14f5757104e9: Pull complete
611fe4f705a5: Downloading [==============================================>    ]  2.744GB/2.964GB
ac0b78389510: Download complete
9e499ad45d56: Download complete
d9fe4a797d63: Pull complete
0096e55d11c8: Pull complete
8edcf0ca7093: Pull complete
4d1c726a5a26: Pull complete
6a13c4f0f499: Extracting [==================================================>]    147MB/147MB
4041fab309df: Download complete
76cb4738a4de: Download complete
9f64cb20b123: Download complete
59feaf5d0228: Download complete
bf2e5cea3f0a: Download complete
b4ba13fe1476: Download complete
31a70e08417a: Download complete
fdd6ba96cc6b: Download complete
39bb0bc52b4b: Download complete
8d4583aa23d3: Download complete
4ac066736182: Download complete
378885e5b53c: Download complete
failed to register layer: ApplyLayer exit status 1 stdout:  stderr: write /root/.npm/_cacache/content-v2/sha512/39/e9/cc5303f01ee98af6a201887d9e0f72ef032bfc23a8390fc0a013c1e5c23f8aa0e26dbfc6502037ce4eac8210199eccc2988b0183f1fa6759803c5bd54384: no space left on device

[effigies]

Could be misunderstanding, but it seems that you've run out of space where you're storing docker images? I'm not really sure how you would fail to have space in a docker image that you're downloading for the content in the image.

Also, do you still see this issue with nipreps/fmriprep?

[yarikoptic]

my point -- why images should carry /root/.npm/_cacache/ at all? Still trying to figure out what/where space is lacking since I still have GBs and free inodes whenever it crashes

[mgxd]

The LTS Dockerfile can definitely be better optimized - for example, the npm steps are spread out in 3 separate RUN commands.

Re: space issues: You might wanna run a docker system prune to clear out some space.

[yarikoptic]

FWIW, here is what I observe (on another box, where it managed to pull fresh (which took 15 minutes, primarily to extract I believe. box is on a 10Gbps, IO during extraction was about 150MBps read)

$> docker run -it --entrypoint bash poldracklab/fmriprep -c 'find /root/ | nl | tail'
  2579	/root/.npm/_cacache/index-v5/ff/d3
  2580	/root/.npm/_cacache/index-v5/ff/d3/eca629ba696cac5a91e13f61bb768d7cbf82072f798d17071572ab3943f2
  2581	/root/.npm/_cacache/tmp
  2582	/root/.npm/_locks
  2583	/root/.npm/anonymous-cli-metrics.json
  2584	/root/.conda
  2585	/root/.conda/environments.txt
  2586	/root/.conda/pkgs
  2587	/root/.conda/pkgs/urls
  2588	/root/.conda/pkgs/urls.txt

$> docker run -it --entrypoint bash poldracklab/fmriprep -c 'du -scm /root/'                
76	/root/
76	total

edit: so the likely culprit was inodes, not actual size

[oesteban]

Hey @yarikoptic, this probably holds after the change, but you should be using nipreps/fmriprep instead (I believe).

[effigies]

Should be fixed in next release.