hashar added a comment.

  @awight potentially though `tmpfs` goes in memory and is expensive. A run 
with core + Wikibase and all the npm dependencies turn out to be around 9G iirc 
which is a lot to put in RAM.
  
  I was looking at a stall build today and it `iotop` definitely showed up the 
Docker daemon being IO busy. Ideally the container destroying should be done 
asynchronously rather than holding and preventing `docker run` from completing 
but that is how things are :/
  
  On the "hardware" side files are written to the instance "disk" which is 
backed up by Ceph. There is some IO throttling there. If we could find a way to 
make the underlying storage to be less synchronous or accept data loss, we 
would certainly save a bunch of time.  I have read some presentation about 
making CephFS (no idea whether we use that) which indicates all unlinks are 
synchronous, deleting the myriads of files installed by npm would definitely 
explain the slowness :-\

TASK DETAIL
  https://phabricator.wikimedia.org/T265615

EMAIL PREFERENCES
  https://phabricator.wikimedia.org/settings/panel/emailpreferences/

To: hashar
Cc: Aklapper, dduvall, bd808, jeena, Daimona, thcipriani, hashar, awight, 
kostajh, Lucas_Werkmeister_WMDE, 786, TheReadOnly, Suran38, Biggs657, 
Invadibot, Lalamarie69, maantietaja, Juan90264, Alter-paule, Beast1978, Un1tY, 
Akuckartz, Hook696, Kent7301, joker88john, CucyNoiD, Nandana, NebulousIris, 
Gaboe420, Giuliamocci, Cpaulf30, Lahi, Gq86, Af420, Bsandipan, 
GoranSMilovanovic, QZanden, LawExplorer, Lewizho99, Maathavan, _jensen, 
rosalieper, Liudvikas, Scott_WUaS, Wikidata-bugs, aude, Jdforrester-WMF, 
Mbch331, Jay8g
_______________________________________________
Wikidata-bugs mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to