On 25 Sep 2012, at 13:14, Galder Zamarreño wrote:
Can't we rather enhance HR so that the shutting down servers would piggyback
the "restart" information to the clients? When a client receives a restart
response (similar to a topology change response), it restarts itself
without requiring any manual work. How does it sound?
I didn't understand what you have in mind, but nevertheless it still
sounds like a nice-to-have.
Neither I… and given how critical shutdown procedure is to make sure it's fast and does not block, I'm a bit dubious...
obviously I haven't explained this right then. Let me try again :-)
AFAIK ISPN-1407 [1] is needed in order to switch the HR clients connected to cluster A (running e.g. ISPN 5.2) to cluster B (running e.g. ISPN 5.3) when doing rolling upgrades.
This involves an sys admin connecting to every client and manually updating this list.
What I'm suggesting is a potentially more elegant approach to achieve this, by upgrading the HR protocol:
- sys admin connects(JMX) to cluster A and marks the site as "upgraded to cluster B"
- at this stage the HR endpoints on cluster A, when receiving requests from connected users, would also tell them(piggyback, same as we do with topology info) to connect to "cluster B" on next requests
- this way the sys admin does not have to update the list of servers for all the users (there might be lots of them), but only update it once on the server side
[1]ISPN-1407 - Make sure the server_list attribute of the Hot Rod Client can be updated dynamically