The most interesting part of this is the k8s integration to find your peers and connect to them. I've wanted this for a variety of projects; I wonder if this is worth extracting into a separate library?
You can do service discovery with a k/v store like etcd, or consul, which is how Kubernetes does service discovery. In fact, Consul very much positions itself as a service discovery tool.
I'm not sure how it would be possible to do it in a "library" (without external centralized storage).
In 2003 Republicans threw a bunch of dark money to recall a democratically elected governor, Gray Davis, and replace him with a son of a nazi, Arnold Schwarzaneger, a bad actor, as governor. Really. Republicans, especially out of state Mormons opposed to gay marriage), spent a few hundred million dollars when they saw the chance to invalidate our hippy vote and .. replace our chosen governor with .. An aryan son of a nazi. I'm not creative enough to make this up.
Turns out it only takes about $200m to invalidate an election and replace a real lawmaker with Kindergarten Cop.
It was the second biggest theft in our political history, second only to that crying alcoholic fratboy Brett Kavanaugh, who was installed by the Republicans refusing to do their job for a year and a half and process Obama's final Supreme Court nominee, Merrick Garland.
It shows intent, and also tries to give more context behind the
simultaneously humorous and negative connotation behind the name. The name itself is a reference to an understanding that in the USA money, violence, and corruption required to be in politics.
The son of a nazi reference really needs to be there, as it highlights the Nazi element of the Republican party. They didn't just steal the governorship from a bunch of hippies, they replaced a democratically elected governor with "the aryan ideal", accent and racist lineage et al, to prove a point: Californians are second class Americans, white christians only.
The Gubernator does not take votes. The Gubernator terminates democracy.
This looks very cool, but I don't understand why they didn't build their entire caching infrastructure this way. Or to put it another way, why not build all caching like this, and then use your standard cache for rate limits?
I think this could be used for both incoming and outgoing requests. From what I gathered reading the docs and API, this doesn't actually forward your request to the API, it just keeps track of rate's and limits, leaving the actual limiting up to you. Its a microservice that keeps track of limits for you, and is highly distributed.
Basically, and I can be totally wrong, what it looks like is you send a rate limit request stating I'm doing 1 hit, I can do 100 hits a minute, am I over limit? Gubernator responds with either a status 0 (no, under limit), how many requests you have left, and the time the rate limit will reset. Or it will return 1 (yes, over limit), the time the limit resets, and you then have to hold your request until the reset time.
So you could use this as a self imposed rate limit for external services. I think the confusion is from the announcement, it sounded like this was a proxy type service that would hold your request until the limit was met. From what I read, it looks like you ask for the rate limit status, then its up to you to actually limit your requests.
You can use haproxy+squid for this purpose. Basically implement the rate limiting on haproxy and send requests to squid “backend” that is a forward proxy to the internet.
Squid is useful as a forward proxy for a number of reasons, like caching. Squid is often used for outbound proxying when you are anonymizing requests, i.e. you might have squid bound to multiple network interfaces.
If all you need is rate limiting, you can probably skip squid.
I'd usually use a queue + worker for this purpose.
You will have to work out how to deal with old requests, backlogs, etc.. But then, if you are considering effectively forcibly making requests time out to internal servers, you will also need a strategy for handling that, retries, etc.
It maintains zero configuration state. It is not completely stateless.
Consumption of available requests is maintained statefully in RAM. If your Gubernator goes down, the quotas are effectively topped back up when Gubernator returns from the dead, upon next request from that client.
22 comments
[ 2.9 ms ] story [ 54.9 ms ] threadI'm not sure how it would be possible to do it in a "library" (without external centralized storage).
Etymology of Latin "guberno": From a non-Indo-European Mediterranean language; akin to Ancient Greek κυβερνάω (kubernáō).
Turns out it only takes about $200m to invalidate an election and replace a real lawmaker with Kindergarten Cop.
It was the second biggest theft in our political history, second only to that crying alcoholic fratboy Brett Kavanaugh, who was installed by the Republicans refusing to do their job for a year and a half and process Obama's final Supreme Court nominee, Merrick Garland.
Governor + Terminator == Gubernator.
The son of a nazi reference really needs to be there, as it highlights the Nazi element of the Republican party. They didn't just steal the governorship from a bunch of hippies, they replaced a democratically elected governor with "the aryan ideal", accent and racist lineage et al, to prove a point: Californians are second class Americans, white christians only.
The Gubernator does not take votes. The Gubernator terminates democracy.
Basically, and I can be totally wrong, what it looks like is you send a rate limit request stating I'm doing 1 hit, I can do 100 hits a minute, am I over limit? Gubernator responds with either a status 0 (no, under limit), how many requests you have left, and the time the rate limit will reset. Or it will return 1 (yes, over limit), the time the limit resets, and you then have to hold your request until the reset time.
So you could use this as a self imposed rate limit for external services. I think the confusion is from the announcement, it sounded like this was a proxy type service that would hold your request until the limit was met. From what I read, it looks like you ask for the rate limit status, then its up to you to actually limit your requests.
If all you need is rate limiting, you can probably skip squid.
Consumption of available requests is maintained statefully in RAM. If your Gubernator goes down, the quotas are effectively topped back up when Gubernator returns from the dead, upon next request from that client.