| opendevreview | Jiri Podivin proposed openstack/openstack-ansible-os_tempest master: Switching linters to pre-commit system https://review.opendev.org/c/openstack/openstack-ansible-os_tempest/+/858521 | 07:28 |
|---|---|---|
| moha7 | jrosser: "cinder does not need to be on the compute hosts for ceph", but the AIO_Ceph does it too, by setting`storage_hosts:` in /etc/openstack_deploy/conf.d/cinder.yml, doesn't it? | 10:07 |
| jrosser | moha7: because in the AIO there is only one host? so where else would it go :) | 10:08 |
| jrosser | you don’t need haproxy on all the compute hosts either, but also that just how it has to be in the AIO too | 10:09 |
| moha7 | So, you mean `storage_hosts: *controller_hosts` in my case? | 10:10 |
| moha7 | Indeed, should I set `storage_hosts` at all for a multinode env with a standalone Ceph cluster as Cinder storage? | 10:12 |
| jrosser | well you still need the cinder api and volume services to be running on the controllers | 10:12 |
| moha7 | not handled by `storage-infra_hosts:`? | 10:14 |
| jrosser | sorry I’m not in front of the code at all today | 10:14 |
| moha7 | Ah, ok | 10:14 |
| jrosser | either way cinder organises some remote storage to be used by the compute node, be it iscsi or rbd or whatever else | 10:15 |
| jrosser | part of the confusion might be that in the non ceph AIO cinder-volume is set up to export iscsi from a loopback lvm device, just for the purposes of a yes with the simplest backend. that has to be on the host rather than a container because the iscsi target stuff doesn’t work in an lxc | 10:19 |
| jrosser | *purposes of a test… | 10:20 |
| jrosser | for lvm backed storage cinder-volume actually provides the iscsi, but for ceph it just mediates the storage backend | 10:23 |
| moha7 | in /etc/openstack_deploy/env.d/cinder-volume.yml, does `is_metal: false` mean having a container for cinder-volume? | 10:24 |
| moha7 | my understanding: "Is it metal? No", then should be there a cinder volume container, but it does not exist among "lxc-ls --active"! | 10:26 |
| *** dviroel|rout is now known as dviroel | 11:19 | |
| moha7 | ^ | 14:15 |
| *** lowercas_ is now known as Arlion | 14:32 | |
| *** Arlion is now known as lowercase | 14:32 | |
| spatel | Folks, ceph question i have PERC H710 controller should i be using write-back or write-through cache? | 14:34 |
| moha7 | Update: The cinder volume container is there! My bad. | 14:48 |
| *** dviroel is now known as dviroel|lunch | 15:14 | |
| spatel | Any idea about this error - https://paste.opendev.org/show/bmAj7IrkhbRpMbe14zht/ | 16:08 |
| *** dviroel|lunch is now known as dviroel | 16:14 | |
| moha7 | I still don't where `storage_hosts` should point to? | 16:44 |
| moha7 | cinder in inventory.json: https://i.ibb.co/1zttDs2/Screenshot-2023-02-03-210536.png --> Is normal to have both `infra0X` and `infra0X_cinder_api_container` under the section`storage-infra_all`? Please have a look at `storage_all` section | 17:38 |
| moha7 | Seems yes. I compared it with the AIO inventory. | 17:40 |
| opendevreview | Damian Dąbrowski proposed openstack/openstack-ansible-specs master: Blueprint for separated haproxy service config https://review.opendev.org/c/openstack/openstack-ansible-specs/+/871187 | 18:19 |
| jrosser | damiandabrowski: ^ imho we should perhaps be limiting ourselves to be fixing (2) in your spec | 18:20 |
| opendevreview | Damian Dąbrowski proposed openstack/openstack-ansible-specs master: Blueprint for separated haproxy service config https://review.opendev.org/c/openstack/openstack-ansible-specs/+/871187 | 18:35 |
| opendevreview | Damian Dąbrowski proposed openstack/openstack-ansible-specs master: Blueprint for separated haproxy service config https://review.opendev.org/c/openstack/openstack-ansible-specs/+/871187 | 18:39 |
| damiandabrowski | omg, i can't get opendev rst preview to render this blueprint properly | 18:41 |
| damiandabrowski | jrosser: why you think so? | 18:42 |
| jrosser | my feeling is that the current changes are spreading data/responsibility all across the roles, rather than keeping them cleanly implemented | 18:43 |
| jrosser | for example i use the haproxy_server role for things nothing to do with OSA | 18:44 |
| jrosser | and now this talks about horizon and makes assumptions about port 443, which for my other use cases are not valid https://review.opendev.org/c/openstack/openstack-ansible-haproxy_server/+/871188/5/tasks/haproxy_ssl_letsencrypt.yml#67 | 18:45 |
| damiandabrowski | hmm, maybe we can discuss it on next meeting. My point was to define variables only in a groups where they are really used. For ex. with separated-haproxy-service-config it makes sense to define openstack_haproxy_horizon_stick_table only for horizon because this variable is not used anywhere else | 18:48 |
| damiandabrowski | but your point regarding horizon and port 443 may be valid, I'll have a look at it on Monday(but it has nothing to do with variable scope) | 18:49 |
| jrosser | i think that is a completely different problem to splitting up the way the haproxy deployment is done | 18:49 |
| jrosser | you could leave the vars almost completely as they are and still move the haproxy service config into each playbook | 18:50 |
| damiandabrowski | but how would you access variables like glance_backend_https then? | 18:50 |
| damiandabrowski | (if they are needed for haproxy service configuration) | 18:51 |
| damiandabrowski | it really sucks that now we need to define glance_backend_https globally(for all hosts) and point 1. from my blueprint tries to fix it | 18:52 |
| jrosser | i don't think that there is an answer to that | 18:58 |
| jrosser | there has to be some compromise here | 18:58 |
| jrosser | and these patches fundamentally change the "data is wired between roles using group_vars" concept in OSA | 18:58 |
| jrosser | and there is no escaping that some of these vars pretty much have to become globals | 18:58 |
| jrosser | also remember that OSA supports using an external lb such as an F5, so we have to keep everthing to do with haproxy seperable and optional | 19:02 |
| damiandabrowski | that's right, some of them do need to remain global(like haproxy_ssl) so I moved them to group_vars/all/haproxy.yml | 19:03 |
| damiandabrowski | but some of them(like openstack_haproxy_horizon_stick_table) are needed only for horizon, so i moved them to group_vars/horizon_all.yml | 19:03 |
| damiandabrowski | sorry but I really need to go :/ have a great weekend! | 19:03 |
| jrosser | ok, you too | 19:03 |
| *** dviroel is now known as dviroel|pto | 21:03 | |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!
