Dynomite without Dyno about dynomite

[timiblossom]

@mfouilleul when a node goes down on on rack and traffics reroute to another rack, this gives the administrator time to recover the 1st rack by bringing up that down node with warm data. The probabability of having 2 down nodes, one on each rack, is small and we can always increase the number of racks to make this probability even smaller.

Yeah, having HA is expensive because we have to store redundant copies of data. If cost is the issue, you can use just one rack to front a database, then if a node is down, just let the traffics go directly to the database while you are fixing the down node(s). Since only a small traffics are allowed to hit the database, I think it should be ok for most cases.

[mfouilleul]

So my topology sounds good?

[timiblossom]

It sounds good. However, can traffics go back to RING1-DC1-RC1 after RING1-DC1-RC1 goes bad and you recover it?

[mfouilleul]

As soon as the RING1-DC1-RC1 is pingable haproxy routes the traffic to it. So we should care to make it pingable only if we recover the consistency (by playing with the /state commands).

I should test all this again and i think we will begin to setup a little rack soon.

Stay in touch.

Max

On 22 Mar 2015, at 08:46, Minh Do [email protected] wrote:

It sounds good. However, can traffics go back to RING1-DC1-RC1 after RING1-DC1-RC1 goes bad and you recover it?

—
Reply to this email directly or view it on GitHub.