Searchfox release4 mingw indexing error on "yq read" due to snaps auto-updating
Categories
(Webtools :: Searchfox, defect)
Tracking
(Not tracked)
People
(Reporter: asuth, Assigned: kats)
Details
The use of "yq read" is specific to the mingw indexing. The indexer seems to be gone, so I'm assuming :kats or :emilio terminated the indexer? We haven't re-provisioned anything since October but I see that we install yq via a snap and although we explicitly remove unattended-upgrades as a package to address the situation where apt packages would self-update out from under us... I'm assuming that perhaps snaps can do the same thing...
Downloading mingw blame.
+ pushd /mnt/index-scratch/mingw-moz
/mnt/index-scratch/mingw-moz ~
+ '[' -d blame ']'
+ wget -nv https://s3-us-west-2.amazonaws.com/searchfox.repositories/mingw-blame.tar
2020-12-28 12:11:54 URL:https://s3-us-west-2.amazonaws.com/searchfox.repositories/mingw-blame.tar [178104320/178104320] -> "mingw-blame.tar" [1]
+ tar xf mingw-blame.tar
+ rm mingw-blame.tar
+ popd
~
+ date
Mon Dec 28 12:11:54 UTC 2020
++ curl -SsfL https://hg.mozilla.org/mozilla-central/raw-file/tip/taskcluster/ci/fetch/toolchains.yml
++ yq read - mingw-w64.fetch.revision
Error: unknown command "read" for "yq"
Run 'yq --help' for usage.
curl: (23) Failed writing body (0 != 13363)
|
Reporter | |
Comment 1•5 years ago
|
||
Actually, I see that we straight up call terminate_instances rather than stop_instances in terminate-indexer.py and always have... so I'm confused why the indexers ever ended up stopped before unless it was an edge case where the call to terminate the indexer was downgraded to a call to stop due to the runtime-attached EBS instance. Presumably bug 1471993 making us use DeleteOnTermination might have fixed that? (Or I fuzzily recall that perhaps we were erroring out before actually getting to the terminate call and so instead perhaps some other path was stopping the indexer like the failsafe timer?)
|
Assignee | |
Comment 2•5 years ago
•
|
||
Normally the shutdown call at the end of send-failure-email.py is what stops the instance. In this case I did (manually) terminate the indexer, since I didn't think there was much to investigate, but maybe I was mistaken.
|
|
Assignee | |
Comment 3•5 years ago
|
||
You were right, yq seems to have changed from under us and no longer takes a "read" command.
https://github.com/mozsearch/mozsearch-mozilla/pull/123 updates to the new syntax.
|
|
Assignee | |
Comment 4•5 years ago
|
||
I have new AMIs that use the yq from pip instead of the snap. Testing that out now.
|
|
Assignee | |
Comment 5•5 years ago
|
||
https://github.com/mozsearch/mozsearch/pull/391 for the provisioner changes
https://github.com/mozsearch/mozsearch-mozilla/pull/126 for the yq syntax change (again).
The new AMIs are currently in EC2 as {indexer,web-server}-18.04-nosnap and will need to be copied over to {indexer,web-server}-18.04 before merging that second PR.
|
|
Assignee | |
Comment 6•5 years ago
|
||
PRs merged and stuff is deployed.
Description
•