Closed laurentS closed 4 years ago
bundle exec rake temp:update_params_yaml
by ssh'ing into the serverused do-release-upgrade
as advised pretty much everywhere. The upgrade itself looks ok, but on production server, it breaks as pdftk
is not available in bionic
anymore, https://github.com/mysociety/alaveteli/issues/5025 and fix in https://github.com/mysociety/alaveteli/pull/5232/files#diff-d5eb0653888644638720a6b98d9d9f1d don't work anymore because cosmic
is not supported anymore, so mirrors are down.
Fixed by installng pdftk from eoan
instead of cosmic
.
passenger
setup is completely different for bionic
vs xenial
, so nginx
wouldn't start either. Fixed by following https://www.phusionpassenger.com/docs/tutorials/deploy_to_production/installations/oss/ownserver/ruby/nginx/
opendkim refused to start on staging server because the default config was changed between xenial and bionic. Fixes are in the git log.
Just about everything else fell apart in various ways between the 2 servers, python, ruby, etc... I fixed most of them through manual trial and error.
Take away: If at all possible, install the newer OS on a fresh machine and move over to it instead of trying the ubuntu upgrade path which even for LTS versions is very much broken.
Upgrade itself super easy, except our xapian index update seemed to have crashed mid way a week before. removing production.new
and rerunning the update script fixed it.
geoip
is a problem, the v1 database is not available anymore, so rails geoip:download_data
does not run, and we're now unable to login to the site. Work around:
*.mmdb
file to the server, under <alaveteliroot>/vendor/data
config/general.yml
to have GEOIP_DATABASE: vendor/data/GeoLite2-Country.mmdb
nginx
This seems to be enough for the server to function normally.
The geoip
data download is now part of the rails-post-deploy
script, and fails for the same reason as above.
rails-post-deploy
scriptrails-deploy-before-down
script to comment out the line related to geoip
at the end of the filesudo -u alaveteli script/rails-post-deploy
manuallyPatched config/routes.rb
in our ansible setup as we overwrite the original one to use translated routes.
we're already on xapian 1.4.5 so we don't seem to need to move from flint, as it's not supported anymore
we're on postgresql 9.6, so no need to upgrade this
we're using ruby 2.5, which seems fine from the alaveteli release notes
first run of the CI process timed out on bundle install
, and the server went into a 100% cpu usage, losing ssh connectivity
after reboot, manual run of bundle install
took a while but completed
next CI run failed during rails-post-deploy
, which was fixed as per https://github.com/mysociety/alaveteli/issues/5844
This seems to be working ok now, closing.
En vue du passage vers les fonctions pro, on doit commencer par une mise à jour vers la dernière version. La 0.38.0.0 devrait être dispo dans les jours qui viennent. Je continue ce ticket en anglais, vu qu'il est probable que d'autres sites alaveteli puissent y trouver un intérêt.
This ticket keeps track of our upgrade process from 0.33.0.0 to 0.38.0.0, in preparation for the move to PRO features.