| RomanHros[m] | Hi, I have the following observation in active/standby and want to know if it is correct.... (full message at <https://matrix.org/oftc/media/v1/media/download/AeFY4NWaGppogB6vJZzsYlREm9d_hb9wBCmGB_kBwZqZvUELx-qF1_gMDpWynTv_py7IkVFf_YQpJjUetKFbLUpCefaX19CQAG1hdHJpeC5vcmcvZ0VvRGNBWWxFcENIcWFoUFZkdE5OTXV0>) | 05:54 |
|---|---|---|
| gthiemonge | RomanHros[m]: hey, yeah the role of the amphorae in the DB is almost purely informative and doesn't match the real roles of the amphorae (which can be updated several times in a second). Octavia lets keepalived decide which amp is the MASTER. before we fixed the nopreempt option, keepalived would have reassigned the MASTER role to a newly created MASTER amphora (after a failover for instance), | 06:37 |
| gthiemonge | and it triggered unnecessary traffic interruption | 06:37 |
| RomanHros[m] | gthiemonge: thank you, so my observation is right. There is no connection between roles in octavia db and keepalived roles | 06:40 |
| RomanHros[m] | Also, when you spawn fresh new LB, and we assume there are no disruptions, the roles match. But after the first LB failover, ha_ip is already on the BACKUP instance. And it will be still the same after additional failovers, because LB failover always marks first created instance as BACKUP. | 06:44 |
| RomanHros[m] | gthiemonge: I have another question. Octavia has this `controller_ip_port_list` which can be updated with `/v2/octavia/amphorae/{amphora_id}/config` API call. But I do not see this option for `admin_log_targets / tenant_log_targets`. What is good approach here? Maybe this can be implemented(probably rsysolg would need restart also)? In the docs I see only option | 08:36 |
| RomanHros[m] | https://docs.openstack.org/octavia/latest/admin/log-offloading.html#rsyslog-failover-considerations, but still IPs can change in the future. Probably there is no other way then failover right now? | 08:36 |
| gthiemonge | RomanHros[m]: admin_log_targets/tenant_log_targets are written into the rsyslog config file only during the init of the amphora, there's no way to update it and restart rsyslog ATM (while controller_ip_port_list is handled by oslo.config and the amphora-agent instance is reloaded gracefully) | 09:07 |
| gthiemonge | so yeah, failover is the only solution | 09:08 |
| RomanHros[m] | gthiemonge: thank you. Maybe this can be implemented in the octavia? So re-render the rsyslog config and restart rsyslog service? | 09:09 |
| gthiemonge | RomanHros[m]: yeah, I'll open a launchpad to track this work | 11:36 |
| *** zseguin_ is now known as zseguin | 23:53 | |
Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!