New answers tagged

1

WebSocket is natively supported by the HTTP(s) load balancers in GCP. If you are using that load balancer and you are receiving the “upgrade_header_rejected” message is probably because the backend rejected the connection. It is also possible that the client tried to upgrade from a secure connection to an insecure one or that the header is an invalid one.


1

You shouldn't use elastic IPs on this kind of instances as you'll run out rather quickly. Remember there's a (soft) limit of 5 per region. About adding the instance to your load balancer, what you can do is register them into the target group upon launch. Adding something like aws elbv2 register-targets --target-group-arn YOUR_TG_ARN --targets "[ { 'Id':...


0

This can't be done with the ngx_http_realip_module itself, given that as specified in real_ip_recursive's description, only the last address in the header field (ie. your GFE's, when set to off) or the last untrusted address (ie. the client's address, when set to on) will be matched to replace $remote_addr, so even if you got the previous proxy's address ...


2

Also not sure why just hitting yobuzzer.com with the cname is not working, but dev.yobuzzer.com works You can't use CNAME on root domain (yobuzzer.com); you can only use CNAME on subdomain (www.yobuzzer.com, dev.yobuzzer.com, etc). Root domain should only use A or AAAA. Though, on Cloudflare, they can flatten CNAME on root domain, technically querying A ...


0

Perhaps https://www.ssllabs.com/ssltest/ can test/verify what you need?


0

You have to set timeout at the BalanceMember level in combination with setting failontimeout on the balancer level. See https://stackoverflow.com/questions/169453/bad-gateway-502-error-with-apache-mod-proxy-and-tomcat/64805009#64805009.


1

I was wondering how does the ingress controller make sure that the incoming requests for the K8S cluster is load-balanced between all the K8S nodes It defaults to round_robin and uses Endpoints instead of sending traffic through the Service IP which would, itself, do round-robin across all Nodes (which would be wasteful by sending traffic to a Node in which ...


3

The term you’re looking for is reverse proxy. What you do to achieve HA is that you set up a load balancer pair to listen to a common IP address, and then you use a protocol like carp or vrrp (for example using keepalived) that lets them agree on which node is active. Provided you keep all components and paths redundant enough you should be able to achieve ...


1

This is the wrong approach. AWS and other cloud providers do this, but they have IP pools to draw from. I can bet you don't. The approach you want is to have users point their domains with an A record to your public IP. A single IP for all clients. On your side you use hostname redirection to forward them to the appropriate appliacation/vm/container or ...


Top 50 recent answers are included