Closed tarzanek closed 1 year ago
after checking logs
17:18:19 TASK [ansible-scylla-node : start scylla on seeders] ***************************
17:18:19 skipping: [XXX] => {"changed": false, "skip_reason": "Conditional result was False"}
17:18:24 TASK [ansible-scylla-node : Start scylla non-seeds nodes serially] *************
17:18:24 skipping: [xxxx] => (item=xxxx) => {"ansible_loop_var": "item", "changed": false, "item": "xxxx", "skip_reason": "Conditional result was False"}
so it means the service is started by SOMETHING else later in the game ?!?!?
maybe the agent? if its service depends on Scylla it will start it ... ????
yep ... agent starts it as dependency /lib/systemd/system/scylla-manager-agent.service
[Unit]
Description=Scylla Manager Agent
Wants=scylla-server.service
Wants=scylla-helper.slice
After=scylla-server.service
After=network-online.target
so https://github.com/scylladb/scylla-ansible-roles/blob/master/ansible-scylla-node/tasks/manager_agents.yml#L57 cannot be started if Scylla start is false ...
this needs fixing of the condition
reopening this condition in https://github.com/scylladb/scylla-ansible-roles/pull/185 didn't work needs more work
OK so there are handlers that trigger agent start:
https://github.com/scylladb/scylla-ansible-roles/blob/master/ansible-scylla-node/handlers/main.yml
they also shall not run if service start is false
OK, so https://github.com/scylladb/scylla-ansible-roles/blob/365a23dabb8fcd07a44654660847d37824e33a51/ansible-scylla-node/tasks/Debian.yml#L109 calls notify which then pushes the handler to the end of play
ok, above was added due to https://github.com/scylladb/scylla-ansible-roles/issues/55 so we need to make sure after upgrade agent is restarted
ok, now above worked :-)
It seems we lost ability to skip starting of Scylla nodes after provisioning
this needs to be brought back (e.g. replace dead nodes is very hard without this)