On Sat, 2016-12-03 at 07:03 +0000, Fedora compose checker wrote: > > Installed system changes in test x86_64 Everything-boot-iso install_default: > System load changed from 0.04 to 0.19 > Previous test data: https://openqa.fedoraproject.org/tests/50142#downloads > Current test data: https://openqa.fedoraproject.org/tests/50240#downloads Aha! So welcome to the shiny new check-compose feature, check-compose fans! (*are* there any check-compose fans?) For a few weeks now, the 'default install' openQA tests have collected some system information data after completing the install process. It's not rocket science - they just log in and run: top -i -n20 -b rpm -qa --queryformat "%{NAME}\n" free df systemctl -t service --no-pager | grep -o ".*\.service" and store the output. check-compose now analyzes those results and includes what it thinks are interesting changes in the output between the 'previous' compose and the new one. It'll include a note when: * A filesystem's contents change in size by more than 10% * A non-tmpfs filesystem's size changes at all * Used memory or swap changes by more than 10% * The installed package set changes at all * The running service set changes at all * System load after the 60 second 'top' run changes by more than 0.1 * Average CPU usage during the 60 second 'top' run changes by more than 10% * Peak process count during the top run changes by more than 10% The 'top' ones are the trickiest ones to decide on, I'm still kinda fiddling with that. The idea is to catch if, say, some rogue process starts eating a bunch of CPU or I/O time when the system should be 'idle', but making sure it's a fair comparison is a bit difficult. I might have to fiddle around with exactly when the top run happens and stuff. Please excuse any not-terribly-useful output from the 'top' comparison. (In this case, I just changed the 'top' run from happening *after* all the other tests to happening *before* all the other tests; on 20161130.n.0 it ran after, on 20161202.n.2 it ran before, so it's not really an apples-to-apples comparison). When it prints any 'system information' lines for a particular test 'scenario', it'll print links to the download page for both the 'before' and 'after' tests, so you can look at the raw tool output and see if it's worth further investigation. Hope folks find this useful! We'll see how it goes. You can send bug reports, pull requests etc. to the check-compose project on Pagure: https://pagure.io/fedora-qa/check-compose -- Adam Williamson Fedora QA Community Monkey IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net http://www.happyassassin.net _______________________________________________ test mailing list -- test@xxxxxxxxxxxxxxxxxxxxxxx To unsubscribe send an email to test-leave@xxxxxxxxxxxxxxxxxxxxxxx
