You are viewing a single comment's thread from:

RE: The Decentralizatoin Of Hive Infrastructure

in LeoFinance4 years ago

I have 1Gbit Ethernet yet I have a 1TB data cap. A witness node uses about 250-300GB of traffic/month from my observations, an API node far more than that.

I also live in an area where we get snow and occasional bad weather that can knock out the power for 1 minute to as long as 4 days.

More and more ISP are implementing data caps, Comcast just announced they will be enforcing data caps for everyone, previously it was only certain states.

While 99% of the time, my connection is fast and reliable, I avoid hosting anything off it due to these issues.

One of the problems issues I have been noticing lately is not all API nodes are running the latest version or correctly functioning, but to HAProxy they would return a heathy status and would continue to serve traffic.

Sort:  

Comcast, my home provider, just recently instituted data caps this month, but you can "upgrade" your plan back to unlimited for $30/month.

I think it's $100 for me, I have to look I don't think we have caps just yet, not till next month I think. I pay for 1Gbit Internet and have the same cap as someone with 50Mbps.

but to HAProxy they would return a heathy status and would continue to serve traffic.

No, that's not necessarily true, and it's not true the way we use it. Haproxy allows you to define your health check.

Do you have it setup to find issues like Anyx's node running an old version? That caused a lot of grief for a few days.

It's not setup that way now, because we use it to check the health of our own hiveminds, and we keep them at least functionally similar at all times (although we run different codes on them simultaneously sometimes, to see how they behave differently, especially when testing optimizations).

There are more advanced things that could be done though, such as having a haxproxy that is distributing its load do cross-checking on the responses from different nodes to API calls that should have the same answers from each node. But I think that's overkill for now.