| *** ralonsoh_out is now known as ralonsoh | 05:52 | |
| ralonsoh | hi folks, I'm still investigating this issue. I'm checking an issue I have with the RPC service | 05:52 |
|---|---|---|
| ralonsoh | the RPC client is instantiated in the L3 agent init method | 05:53 |
| ralonsoh | but the agent class is passed as a oslo_service.Service to the launcher | 05:54 |
| ralonsoh | when the launcher is executed, the process is forked | 05:54 |
| ralonsoh | the RPC object doesn't like this | 05:54 |
| ralonsoh | https://github.com/openstack/oslo.messaging/blob/master/oslo_messaging/_drivers/impl_rabbit.py#L1082-L1085 | 05:55 |
| ralonsoh | hberaud[m], ^ | 06:10 |
| ralonsoh | (I don't see Daniel here yet) | 06:11 |
| ralonsoh | ok, sorry for the noise | 07:56 |
| ralonsoh | we had a problem in the L3 agent: the RPC clients were called during the init method. Instead they should be instantiated during the init_host() method | 07:56 |
| ralonsoh | looks much better now | 07:57 |
| gibi | ralonsoh: yeah anything threading related should not happen before the fork as fork will copy global state to the child process. | 14:53 |
| gibi | nova hit similar things with global thread pool inited before fork | 14:54 |
| gibi | now we re-init the pool after fork to get away with it as we cannot easily avoid using the pool pre-fork | 14:55 |
| *** dalees1 is now known as dalees | 19:01 | |
| *** Callum0278 is now known as Callum027 | 19:01 | |
Generated by irclog2html.py 4.0.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!