| opendevreview | Michal Nasiadka proposed openstack/kolla-ansible master: ovn: Add sb_relay_group_id to environment https://review.opendev.org/c/openstack/kolla-ansible/+/953528 | 05:46 |
|---|---|---|
| opendevreview | Michal Nasiadka proposed openstack/kolla-ansible master: fluentd: Add ovn logs to input https://review.opendev.org/c/openstack/kolla-ansible/+/942467 | 05:46 |
| opendevreview | Michal Nasiadka proposed openstack/kolla master: ovn: Fix log permissions https://review.opendev.org/c/openstack/kolla/+/952566 | 05:48 |
| opendevreview | Mariusz Karpiarz proposed openstack/kolla-ansible master: Fix prechecks failing for encrypted passwords https://review.opendev.org/c/openstack/kolla-ansible/+/953538 | 08:07 |
| opendevreview | Mariusz Karpiarz proposed openstack/kolla-ansible master: Fix prechecks failing for encrypted passwords https://review.opendev.org/c/openstack/kolla-ansible/+/953538 | 08:16 |
| mnasiadka | Morning | 08:18 |
| bbezak | mrn | 08:41 |
| opendevreview | Seunghun Lee proposed openstack/kolla-ansible master: Make RabbitMQ stream retention policy configurable https://review.opendev.org/c/openstack/kolla-ansible/+/953297 | 10:02 |
| opendevreview | Seunghun Lee proposed openstack/kolla-ansible master: Make RabbitMQ stream retention policy configurable https://review.opendev.org/c/openstack/kolla-ansible/+/953297 | 10:29 |
| *** darmach2 is now known as darmach | 11:33 | |
| F-C | HI people, I just ran into a strange behaviour with proxsql on a kolla-ansible 2024.2 deployment. HERE (https://opendev.org/openstack/kolla-ansible/src/commit/7956c7802281db83b6aa9c8525b227c8fd3c72d1/ansible/roles/loadbalancer/templates/proxysql/proxysql.yaml.j2#L81) is the config template for proxsql. It specificlaly weighs the first node with 100 and all others with 10. Why is that? | 11:52 |
| F-C | I'm asking because in the setup this leads to the DB becoming completely unavailable if the node with weight 100 goes down. I'm not sure why. From the docs of proxysql the weight is only meant for distributing queries when endpoints are active. But when setting weight manually to 10 on the remainin nodes, the DB is reachable again. | 11:54 |
| F-C | Also, why is the first node prioritized anyway in the proxysql config from kolla-ansible? | 11:54 |
| opendevreview | Mariusz Karpiarz proposed openstack/kolla-ansible master: Fix prechecks failing for encrypted passwords https://review.opendev.org/c/openstack/kolla-ansible/+/953538 | 12:06 |
| opendevreview | Seunghun Lee proposed openstack/kolla-ansible master: Make RabbitMQ stream retention policy configurable https://review.opendev.org/c/openstack/kolla-ansible/+/953297 | 12:48 |
| opendevreview | Jake Hutchinson proposed openstack/kayobe master: WIP: CI: Support Redfish baremetal emulation https://review.opendev.org/c/openstack/kayobe/+/952980 | 13:00 |
| jaydg | Hi! Is there anything I can do to get https://review.opendev.org/c/openstack/kolla/+/951090 moving again? | 13:32 |
| opendevreview | Amir Hossein Ahmadi proposed openstack/kolla-ansible master: swift: reintroduce Swift role and restore deployment https://review.opendev.org/c/openstack/kolla-ansible/+/953576 | 13:53 |
| mnasiadka | jaydg: done | 14:22 |
| opendevreview | Merged openstack/kayobe master: Support empty strings in dev-tools package lists https://review.opendev.org/c/openstack/kayobe/+/953274 | 16:11 |
| opendevreview | Pierre Riteau proposed openstack/kayobe stable/2025.1: Support empty strings in dev-tools package lists https://review.opendev.org/c/openstack/kayobe/+/953616 | 16:23 |
| opendevreview | Pierre Riteau proposed openstack/kayobe stable/2024.2: Support empty strings in dev-tools package lists https://review.opendev.org/c/openstack/kayobe/+/953617 | 16:24 |
| opendevreview | Pierre Riteau proposed openstack/kayobe stable/2024.1: Support empty strings in dev-tools package lists https://review.opendev.org/c/openstack/kayobe/+/953618 | 16:24 |
| opendevreview | Merged openstack/kayobe master: Bump dellemc.os10 to 1.2.7 https://review.opendev.org/c/openstack/kayobe/+/953207 | 16:26 |
| opendevreview | Merged openstack/kolla master: Fix glance-api: add missing lsscsi package https://review.opendev.org/c/openstack/kolla/+/951090 | 16:27 |
| opendevreview | Jake Hutchinson proposed openstack/kayobe master: WIP: CI: Support Redfish baremetal emulation https://review.opendev.org/c/openstack/kayobe/+/952980 | 16:54 |
| -opendevstatus- NOTICE: Gerrit is being restarted to pick up a configuration change. You may notice a short outage. | 17:37 | |
| opendevreview | Pierre Riteau proposed openstack/kayobe stable/2025.1: Bump dellemc.os10 to 1.2.7 https://review.opendev.org/c/openstack/kayobe/+/953660 | 18:10 |
| opendevreview | Maksim Malchuk proposed openstack/kolla stable/2025.1: Fix glance-api: add missing lsscsi package https://review.opendev.org/c/openstack/kolla/+/953667 | 18:35 |
| opendevreview | Maksim Malchuk proposed openstack/kolla stable/2024.2: Fix glance-api: add missing lsscsi package https://review.opendev.org/c/openstack/kolla/+/953668 | 18:36 |
| opendevreview | Maksim Malchuk proposed openstack/kolla stable/2024.1: Fix glance-api: add missing lsscsi package https://review.opendev.org/c/openstack/kolla/+/953669 | 18:36 |
| kevko | F-C: Well, it's like that because back in the day, when the load balancer was HAProxy, it used to send traffic to the first host, and in case of a failure, to another one. This behavior was copied and used for ProxySQL as well... Theoretically also because Galera is not an NDB cluster, and in theory, there's no guarantee that you have consistent | 19:50 |
| kevko | data. That's why the writer is always the first host and the others are backups... the weight ensures that the first one is used... but in case of failure, the second one takes over the role... I don't see why it shouldn't work... what's the specific problem you're facing? | 19:50 |
| kevko | F-C: Su | 19:52 |
| kevko | Galera Cluster uses synchronous replication at the transaction level, but it's not fully deterministic like MySQL NDB Cluster. In Galera, if a node crashes during a transaction, it might not have applied the write set yet, leading to temporary inconsistencies. NDB, on the other hand, uses a shared-nothing architecture with real-time distributed | 19:52 |
| kevko | data and two-phase commit, guaranteeing strict data consistency across all nodes | 19:52 |
| kevko | So ..that it is ..proxysql ensure that writer is one server ..and others are backup writer ...also weight is 100 for first node ..because historically it was haproxy first node ... | 19:53 |
| kevko | F-C: I didn't see an issue if first node gone down..even not in CI ...so can you be more descriptive ? | 19:55 |
| kevko | F-C: specific | 20:49 |
| F-C | kevko: Thanks for th reply! I havent fully investigated it. When the node with weight 100 is not available, everything that uses mariadb doesnt work (horizon login, api containers become unhealthy, grafana wont load). When i set the weight for the first node to 10 as well (so all nodes have equal weight) the services come back online. | 21:06 |
| F-C | It might be an issue with how proxysql checks the backend availability. This is a planned controller migration. The node is not completely down. we have just stopped all Kolla services, but the IP-Adress proxysql still has (havent reconfigured with new controller node yet), is still reachable, just not the port/service | 21:08 |
| opendevreview | Jake Hutchinson proposed openstack/kayobe master: WIP: CI: Support Redfish baremetal emulation https://review.opendev.org/c/openstack/kayobe/+/952980 | 21:22 |
Generated by irclog2html.py 4.0.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!