Wednesday, 2022-10-05

NeilHanlonmetalink may be helpful to use clarkb; but unfortunately this is an unsolved issue in EL mirroring. even fedora encounters this from time to time02:19
NeilHanlonit maybe makes sense now to make some openinfra mirrors of rocky, perhaps02:22
opendevreviewMerged zuul/zuul-jobs master: Allow overriding of Bazel installer checksum
fricklerI've paused the rockylinux-9 image now to keep the builders from spinning uselessly, we can retry once in a while 06:10
opendevreviewMerged opendev/system-config master: Disable distro cloud image users more forcefully
fricklerlooking at AFS stats, I don't think we can set up a new mirror without doing another round of cleanup first
*** marios is now known as marios|ruck06:31
*** jpena|off is now known as jpena07:37
*** jpodivin_ is now known as jpodivin12:12
*** kopecmartin is now known as kopecmartin|sick12:56
*** marios|ruck is now known as marios13:31
*** dasm|off is now known as dasm13:43
*** Guest2330 is now known as diablo_rojo13:46
clarkbfrickler: agreed14:06
clarkbNeilHanlon: is there any way to see what is causing the underlying failure? shows what we are currently logging. I got discouraged beacuse trying to reproduce by hand by fetching the mirror list then fetching that package on the same build host seems to work fine14:07
NeilHanlonwe could try removing the mirrorlist and pointing directly to the Rocky Linux CDN. want me to throw in a change which does that for the rocky container image?14:44
NeilHanlonclarkb :^^ 14:44
clarkbNeilHanlon: ya that might be a good place to start. At elast then we'd know exactly what we are talking to?14:44
clarkband if it fails there we have something to pull on14:44
NeilHanlonyep. exactly my thought14:45
opendevreviewMerged openstack/diskimage-builder master: Fix wrong yum.conf name of CentOS 9 Stream
opendevreviewNeil Hanlon proposed openstack/diskimage-builder master: Change rockylinux-9 container to use baseurl instead of mirrorlist
NeilHanloni'm running some checks on the rocky mirrors as well to see if I can find the bad mirror(s)14:54
clarkbNeilHanlon: thansk!14:54
opendevreviewClark Boylan proposed openstack/project-config master: Update OpenDev tenants' default ansible to 6
*** marios is now known as marios|out15:17
clarkbI sent notice of ^ to the service-announce list15:22
clarkbI don't really expect problems but wanted to make sure people had pointers to the testing we've done and how to override this if necessary etc15:22
fungii'll give openstack-discuss a heads up with a pointer to that announcement16:03
fungiand sent16:06
corvusclarkb: you want to merge that tomorrow if no concerns?16:21
clarkbyup I think tomorrow works16:22
jrosser_fwiw the ansible apt module now tries to be much more specific about the versions of things it installs with ansible-core 2.13.x, which would be part of ansible-6 afaik16:37
fungijrosser_: i'm assuming that's a good thing?16:38
clarkbits weird it wouldn't defer that to apt though16:38
fungior are you calling it out as a potentially breaking behavior change for some jobs?16:38
*** jpena is now known as jpena|off16:38
jrosser_yes - potentially16:38
jrosser_this is at least fixed now
jrosser_where pinning was just ignored, which you'd think was something that the ansible apt module just defered to the system16:39
clarkbright, that would've always been my assumption. Apt module lets apt figure it out16:39
clarkbinteresting that it isn't16:39
jrosser_i think thats been the only breaking thing we've found from the OSA side in ansible-core 2.13.x16:40
clarkbwe found an issue in zuul-jobs around the interpretation of shebang lines in modules16:40
clarkbI filed a bug upstream askign the mto clarify the behavior in the documentation and was basically told no16:41
jrosser_ah yes right - i think we had one of those too but it only manifested itself on rocky-9 due to some interpreter discovery stuff16:41
clarkbtl;dr is that ansible uses the shebang to know what version of python to execute a module under not the specific binary16:41
clarkbwhich is a complete misuse of a shebang imo, but all I really wanted the mto do was document the behavior so we could deal with it16:41
fungirather than relying on the kernel's shebang parsing16:41
jrosser_i believe you don't need them at all16:42
clarkbjrosser_: correct that was our solution16:42
fungiwell, you need them if you want to execute them as scripts16:42
clarkbbtu also it isn't documented that they are optional. The docs say they are required16:42
clarkbI found it all very confusing, but apparently upstream doesn't think so so no docs updates to clarify16:42
fungii should say, if you want an ansible module to also be an executable script without the caller needing to specify the interpreter on their own16:43
clarkbat least google should find my issue now helping similarly confused individuals in the future16:43
jrosser_thats kind or orthogonal to some new ansible behviour where it will try many interpreters until it finds one where all the imports work16:43
fungianother workaround is to make your own executable wrapper script which imports the module, and then devs can use that for local testing outside ansible16:44
fungibut basically ansible and local invocation can no longer share the same entrypoint16:44
fungisince the shebang will be parsed by two different incompatible mechanisms16:45
fungiyou'd think it would have made more sense for ansible to just invent its own magic comment syntax independent of shebang, so that they can coexist independently in the same file16:46
jrosser_it's worth knowing about this as well
*** dviroel is now known as dviroel|afk21:17
*** dasm is now known as dasm|off21:34

Generated by 2.17.3 by Marius Gedminas - find it at!