multisite replication data sync should occur on user defined network space

[Narinder Gupta]

If we configure the rados gw for multi-site replication data sync occurs on internal space only. While it is difficult for customers to enable the forwarding of their internal network between data centres. There should be separate binding for data replication and port should be defined where data replication happens so that respective ports can be used and open across the data centers for multi-site replication.

[James Page (james-page)]

I agree that a specific binding (probably just the space binding for the master/slave relation endpoint) would potentially be preferable rather than using the internal extra-binding.

This will require a bit of adaptation in the charm as the URL resolver currently only works with the (public,admin,internal) tuple of values.

Port - less so - that's actually the main service port for RGW so I don't think we should be changing that.

[Andrey Grebennikov (agrebennikov)]

James,
I've spent quite a bit of time on this.
Ideally we'd want to have master and slave binding addresses to be in the endpoints (if you still want it to be hardcoded), otherwise it should be configurable.
The port I think should be picked up from the charm config "port", this way even if the custom port is configured - the same one is used when configuring the zone and the zonegroup.

[Chris MacNaughton (chris.macnaughton)]

I've raised https://bugs.launchpad.net/charm-ceph-radosgw/+bug/1960520 as a more tactical fix to this, and am leaving this bug open as a feature request for a new network binding.