RE: Hardware failover webserver cluster
Hi,
> perhaps that's a bit off topic here but can someone explain what I need to build a hardware failover nginx cluster?
Nope, but define hardware. A loadbalancer does not care what OS you are running, it does care what service it needs to balance. So, would "a debian server running loadbalancer software" be a "hardware load balancer" or are you talking abount a "real" hardware load balancer like a <insert brandname> device.
>The unclear is:
> How does the client know that the server 1 is down and use the other one?
> DNS failover has a delay because of the design I think.
A load balancer does NOT leave it up to the client to select the server, that is what the load balancer is for. The client sees just 1 ip number, that of the load balancer. After that the loadbalancer makes sure that the client gets redirected to one of the active servers. If the server/service suddenly dies the loadbalancer will not get a reply from the service/service when it wants to redirect the next client and remove that server from the active list and redirect the client to the next active server. All of that happens on "the other side" of the loadbalancer and is hidden from the client.
With https the SSL session will end on the loadbalancer and depending on the configuration will either go via straight http or a new https session to the real server. Usually the first because of performance, security should not be an issue on the internal network.
>client ---> SRV1
> |---> SRV2 (failover if SRV1 is down and load balancing)
This is called dns round robbin balancing and depending on who you asks is not "real loadbalancing". It does balance the load but when a server / service dies that server is still in the dns list and it needs a smart client to skip the dead server and go to the next in the list. Also in that case any session information will get lost unless there is some smart mechanism with which the servers in the cluster share session information. A real load balancer will usually recognize a client when it comes back for more information and send to the same server if there is any need for that.
> how did it work on google or somethink else?
>
[...]
;; ANSWER SECTION:
google.de. 300 IN A 173.194.32.255
google.de. 300 IN A 173.194.32.247
google.de. 300 IN A 173.194.32.248
There are 3 ip addresses. Those 3 addresses each point to a load balancer and not to a server. The client does not know that, it simply sees 1 ip number to which it will send the request and get the reply from.
> Is that load balance or failover?
It is a combination. There are 3 loadbalancers for failover in case a loadbalancer dies / has maintenance / whatever.
Bonno Bloksma
Reply to: