uec-component-listener needs to respect deregistered components

[Dustin Kirkland ]

uec-component-listener currently watches the network for new uec components broadcasting avahi messages.

It appears to me that if I manually deregister a node, but leave it running on the network (ie, eucalyptu-nc-publication is still running), eventually the component listener will automatically register the node again.

There's a few ways we could handle this (including documenting that the node must be removed from the network or the publication job stopped).

I think it would make sense if we created a blacklist of manually deregistered components (like nodes.list). This could be a flat file, or could be merged and denoted in the existing list. There's a lot of ways of doing this. I think we should discuss it and try to solve this in a manner that makes sense.

[Dustin Kirkland  (kirkland)]

triaged/medium, assigned to Thierry, as this bug requires at least his input. Ideally this is something that Thierry can think about and commit a fix. If it involves a lot of work, let's discuss it further.

[Dustin Kirkland  (kirkland)]

I think it would be good to solve this for Lucid, but I'll defer to Thierry's judgement on that.

[Thierry Carrez (ttx)]

Dustin: I'm not sure that's a bug :)

uec-component-listener watches for new announcements, so it won't re-register your node unless you stop and start the publication again (or restart the node)... which to me means that you want the node to be re-registered. It will also register your node again if you restart the uec-component-listener process.

I think the classic use case is the following:
You want to remove a node from your cluster. You will shutdown your node and unregister it. If you restart it and you have autoregistration enabled, for me it means you want it to be used again. Otherwise I don't see the point in restarting it. If you don't want node autoregistration to happen, you can disable it at the CC level.

Using a blacklist would just prevent a specific IP address from being autoregistered ever again. I'm not sure why I would want that.

Maybe I don't understand the problem you're trying to solve...

[Thierry Carrez (ttx)]

Dustin, could you give me your opinion on the previous comment ?

[Dustin Kirkland  (kirkland)]

Thierry,

I think your analysis is correct. Marking invalid. Thanks.