Open rhessing opened 6 years ago
I used to think a lot on SSDB distributed.
I think the easiest way is to make a proxy, clients connect to the proxy, not to SSDB instance directly. The proxy knows any SSDB instance down and up, so it can make its decision to write to/read from which SSDB instance.
Distributed storage system is extremly complicated. So by this time, SSDB relies on users decisions.
So a configuration reload is not possible?
Not possible.
Currently I'm dockerizing SSDB to cope with autoscaling using master-master-master-etc replication. However, when I scale my service the new container will discover current containers. But the currently running containers also get notified of the new container and add it as a new master within the SSDB configuration. Then the ones running will restart SSDB to also slave from the new container.
Currently a SSDB restart will disconnect any running request which is not so nice. Can the configuration reload be done without a restart but a reload? Just like Apache/Nginx can?
This will enable horizontal scaling without service interruption :-)