make hostnames explicit#617
make hostnames explicit#617whysthatso wants to merge 1 commit intoeikek:masterfrom whysthatso:patch-1
Conversation
this should solve #608
|
Hi @whysthatso , thank you a lot! I'm in favor for this change. Still want to ping @bjeanes , because he might have a different view? I would then also revert the change in docspell.conf, wdyt? (there is only the joex url that has a host-based url) I think it's nice to be able to simply start new joex containers. But tbh, if it's running on the same machine, one can also just set For reference, see PR #552 |
|
If you hardcode the hostname in docker compose, then you can ONLY run one of them. |
indeed, as i am under the impression that this would be the standard case, albeit this should be documented. |
I don't mind if this is merged, because I can still do whatever I want in my deployment. But, I think the preferred solution here would be something like tracking "last seen at" for each node and culling nodes that haven't reported in in some period. Could also have it randomly pick a node instead of work down the list (if I correctly understand that this is what is happening).
Perhaps, but being able to elastically scale up workers based on queue size is not possible using just
Yes, agreed. |
|
i'll try to add something that makes that parameter explicit in the documentation. |
That's what I figured. People usually have their own deployments and different setups etc and probably use their own compose files (or whatever alternative) anyways. I think we can't and shouldn't support all possible scenarios. Although I can see that this one seems rather like a "best practice". However, I doubt that people use more than one machine most of the time and with one machine, the
Absolutely! This will be fixed. It works as you said; the order of the list is left to the db, so not really random. But I want to notify all nodes, not just one. Currently there is nothing in place to cleanup/watch this list, which is a bug actually. I'm going to create a separate issue for this, was first thinking to use this one, but now I find it better to track separately. I'd say we merge this change for now, until the bug is fixed. Then we can adopt the docker setup again. -? |
|
That all seems sensible @eikek! Agreed that most people won't use multiple machines and that ON one machine |
|
Only want to confirm: the config is not auto-reloaded. so @whysthatso Would you mind also adopting |
|
Another reason to change Line 30 in bd32dff |
this should solve #608