我读到的主要的一些亮点:
- 阿里云机器需求是高度不确定的(因为集群是多租户需求的总和),使用高斯过程来量化t+1时刻的不确定性
- 使用transformer模型中的attention机制来学习时间序列中复杂的依赖模式(非论文原创,之前一些论文也有这样的思路),设计了一种新颖的基于多尺度注意力算法来提取时间序列的特征
- 将scaler的缩放过程建模为强化学习中的MDP过程
- 使用蒙特卡洛方法,即采样有限视野的值来近似未来无限时间的最优成本策略
cat <<EOF | kubectl create -f - | |
apiVersion: apiextensions.k8s.io/v1beta1 | |
kind: CustomResourceDefinition | |
metadata: | |
name: tapps.apps.tkestack.io | |
spec: | |
group: apps.tkestack.io | |
version: v1 | |
names: | |
kind: TApp |
token=$(cat ~/.kube/config | grep token | awk -F: '{print $2}' | awk '{print $1}')
echo $token
server=$(cat ~/.kube/config | grep server | awk -F"server: " '{print $2}')
echo $server $token
curl -k -H "Authorization: Bearer ${token}" "${server}/apis/apps/v1/namespaces/default/statefulsets/ramists/scale"
{
package main | |
import ( | |
"bytes" | |
"flag" | |
"io/ioutil" | |
"log" | |
"net" | |
"net/http" | |
"net/http/httputil" |
wget https://get.helm.sh/helm-v3.2.1-linux-amd64.tar.gz
tar -zxvf helm-v3.2.1-linux-amd64.tar.gz
mv linux-amd64/helm /usr/local/bin/helm
helm repo add stable https://kubernetes-charts.storage.googleapis.com/
helm repo update
helm install grafana stable/grafana
kubectl port-forward --address 0.0.0.0 -n default svc/grafana 8081:80
# grafana password
kubectl get secret --namespace default grafana -o jsonpath="{.data.admin-password}" | base64 --decode ; echo
#!/bin/bash | |
set -e | |
set -u | |
#set -x | |
SSH_KEY="05:8b:8f:f5:09:35:ae:61:a9:3b:21:ea:1c:36:bf:0c" | |
USER_DATA_FILE="$(dirname $0)/vps.userdata" | |
### sub functions |
{
"a": "green",
"b": "white"
}
{
"a": "red",
"c": "purple"