Page MenuHomeSoftware Heritage
Paste P326

Deploy the latest indexers m.o (new v0.0.55)
ActivePublic

Authored by ardumont on Oct 27 2018, 11:16 AM.
If it's urgent to redeploy indexers, here is what i foresee in the
current state of affairs,
- After review and acceptance, merge the diffs:
- D605, D615: swh-scheduler: bump version (git tag) and upload/deploy
new version (debian package)
- D606, D608, D609: swh-indexer: it might be necessary to require the
new swh-scheduler version in requirements-swh.txt
- D612: swh-indexer: rebase that patch on the previous merged diffs
and push, bump version (git tag) and upload/deploy new version
(debian package)
- D614: swh-site, production changes, merge, push and deploy on puppet
master (ask @olasd or @ftigeot)
- check on dev box that everything goes smoothly already
(swh-indexer/docs/dev-info.rst)
- softwareheritage-indexer (somerset): run migration script
sql/upgrade/128.sql (~> @olasd)
- softwareheritage-scheduler (prado): create new task_type
(indexer_mimetype, orchestrator, orchestrator_text). D615 should
demo the data.
For deploying, those nodes needs changes:
- saatchi (scheduler): need to upgrade the new python3-swh.indexer
dependency (IIRC puppet does not install the new python3-swh.indexer
version)
- uffizi (indexer-storage): need to upgrade the new
python3-swh.indexer.storage dependency (puppet does it)
- indexer-workers (indexer workers): need to upgrade the new
python3-swh.indexer and restart services (puppet does it)
That means, from louvre (hypervisor with clustershell configured):
ardumont@louvre sudo clush -b -w @windexer-workers -w saatchi -w uffizi "apt-get update; puppet agent --test; dpkg -l python3-swh.indexer"
(~ as per previous note, it's possible saatchi needs to be updated
manually regarding python3-swh.indexer)
I then stop all services from azure nodes
ardumont@louvre sudo clush -b -w @windexer-workers "systemctl stop
swh-worker@swh_indexer_mimetype";
Start only on 1 node, send some data for indexation and check that
everything goes well (or not). If failure, report (kibana helps), fix,
deploy...
When all is good, restart all services stop. And send data for
indexation.

Event Timeline

ardumont changed the title of this paste from Deploy the latest indexers (v0.0.55) to Deploy the latest indexers m.o (new v0.0.55).

Note: To deploy the new indexers, there are still work to do in another part of our stack, the puppet repository, notably the swh-site repository.
See for example:

  • the service [1]
  • its associated configuration [2] (don't forget there is a diff about it already P326)
  • the actual location deployment of the service [3]

I'll let you check for the orchestrator part as well.

[1] https://forge.softwareheritage.org/source/puppet-swh-site/browse/production/site-modules/profile/manifests/swh/deploy/worker/swh_indexer_mimetype.pp

[2] https://forge.softwareheritage.org/source/puppet-swh-site/browse/production/data/defaults.yaml$1966-1974

[3] https://forge.softwareheritage.org/source/puppet-swh-site/browse/production/data/location/azure_euwest.yaml$16-21

@moranegg to let you know ;)
@ftigeot might be interested on that part as well.