common-bandwidth-auto-switch

中文文档

why

Alibaba Cloud’s fancy 95-tier shared bandwidth plan guarantees a minimum of 200Mbps, with a minimum spend of 200*0.2=40Mbps. After some serious number crunching (and a bit of guesswork), I figured that using it somewhere between 40~50Mbps is the sweet spot.

Thanks to the 2020 Chinese New Year + Wuhan’s infamous coronavirus party crasher, website traffic plummeted faster than my motivation on a Monday morning. So, I shoved all the IPs into the shared bandwidth pool.

On 2020-03-05 at 13:20 (Thursday, because why not), Baidu decided to crawl our site like a caffeine-fueled spider, and bandwidth instantly shot up to 200Mbps.

I originally planned to yank the EIP out of the shared bandwidth. But that afternoon, freshly woken from a nap, operating at a solid 10% brainpower, I misread the screen and accidentally removed the SLB bound to the EIP, causing the site to go poof and become unreachable.

After licking my wounds, I vowed to develop an automatic program to manage and adjust the shared bandwidth EIP pool so that it can:

1. During low traffic periods, shove EIPs into the shared bandwidth to save money;
2. During bandwidth rush hour, yank EIPs out of the shared bandwidth to boost bandwidth utilization.

All in all, to squeeze out every last drop of cost optimization.

feature

When current shared bandwidth > desired threshold (maxBandwidth), automatically kick out the high-bandwidth EIPs from the shared bandwidth pool.

When current shared bandwidth < desired threshold ((maxBandwidth+minBandwidth)/2), automatically invite high-bandwidth EIPs back into the shared bandwidth pool.

The core algorithm is dynamic programming. Fancy, right?

Once bandwidth reaches equilibrium, you’ll realize this program is about as useful as a chocolate teapot — it barely triggers any scaling at all.

usage

RAM Permissions

APIs used:

vpc
DescribeEipMonitorData
DescribeEipAddresses

cms
DescribeCommonBandwidthPackages
AddCommonBandwidthPackageIp
RemoveCommonBandwidthPackageIp

cms
DescribeMetricList

Because you’re messing with VPC and shared bandwidth — aka high-risk operations — make sure your RAM permissions are on point.

If you’re feeling lazy about permissions, just slap on these:

1. AliyunVPCFullAccess
2. AliyunEIPFullAccess
3. AliyunCloudMonitorFullAccess
4. AliyunCommonBandwidthPackageFullAccess

Configuration File

Copy config-example.yaml to config.yaml and tweak it to your heart’s content.

cp config-example.yaml config.yaml
vi config.yaml

docker/docker-compose

# docker
docker run -it \
-v /root/common-bandwidth-auto-switch/config.yaml:/app/config/config.yaml \
zeusro/common-bandwidth-auto-switch:latest
# docker-compose
docker-compose up

kubernetes

Here’s an example of deploying with Kubernetes using a CronJob + Configmap combo, because why not?

kubectl create configmap cbwp-config --from-file=config.yaml=config.yaml

kubectl apply -f deploy/kubernetes/cronjob.yaml

Configuration loading order: tries config.yaml first; if missing, falls back to config-example.yaml.

rm -rf /

kubectl delete cronjob common-bandwidth-auto-switch
kubectl delete configmap cbwp-config

todo(NEVER DO)

1. Make it blocking, run periodically, and add health checks.
2. Run multiple instances.

Because who doesn’t love living on the edge?