| opendevreview | Kaifeng Wang proposed openstack/ironic master: Remove instance image after deployment https://review.opendev.org/c/openstack/ironic/+/878680 | 06:12 |
|---|---|---|
| JayF | TheJulia: I wonder... could we do something like a more persistent connection for hb-only agents? | 14:33 |
| JayF | TheJulia: just pondering if we could keep the connection open and send more stuff, like pipelining. idk | 14:33 |
| * JayF melts back into the weekend | 14:33 | |
| TheJulia | Api would need to continuously re-poll most likely to make pipelining work, or the agent would just need to keep asking to trigger the backend | 14:39 |
| TheJulia | At least, that is the immediate thought | 14:39 |
| JayF | I'm just thinking more like, long polling | 14:40 |
| JayF | heartbeat -> [wait for command] -> once it gets a command execute it -> immediately heartbeat again -> [wait for command] | 14:41 |
| JayF | and if we configure the length of the poll is driven by the API server (it can send a "go away" response, for instance, to tell the agent to check in later b/c there is no work) | 14:42 |
| TheJulia | ... firewall folks will hate us :) | 22:51 |
| TheJulia | and services would need to be threaded instead of process per request | 22:53 |
| * TheJulia will never grok why people think just process per request is better | 22:53 | |
| JayF | I mean, in this sort of model you could always fall back. You don't *need* the long-polling to work. It's just a way to make the happy-path take less time (without setting heartbeat intervals to bully-the-api-service levels) | 23:03 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!