directory 구조 변경

README.md

@@ -0,0 +1,11 @@
+# public-ds-cluster-env
+aws 환경 구축 시 사용하는 파일을 정리한다.
+
+## directory 구조
+1. docs: 문서파일 저장.
+2. architecture : aws 구성도
+3. aws_image : aws AMI 이미지 생성
+4. aws_kubernetes : aws kubernetes 생성 (kops)
+5. aws_network : aws network 구성 (terraform)
+6. aws_total : aws 백업 파일 (kops)
+7. script : support script

scripts/kops_env/kops_env.sh

@@ -0,0 +1,41 @@
+#!/bin/bash
+
+export KOPS_STATE_STORE=s3://clusters.spot.datasaker.io
+export KOPS_CLUSTER_NAME=k8s-spot.datasaker.io
+
+export VPC_ID="vpc-00ba2b0e9ad59f0ed "
+export Network_CIDR="172.24.0.0/19"
+export AMI_Image="ami-0abb33b73a78cae31"
+export Private_Subnet_ID_1="subnet-024f0deda82039fa4"
+export Private_Subnet_ID_2="subnet-050d942fa1c46540a"
+export Private_Subnet_ID_3="subnet-0946eb806af7377be"
+export Public_Subnet_ID_1="subnet-00c363356f133411d"
+export Public_Subnet_ID_2="subnet-07aa5e879a262014d"
+export Public_Subnet_ID_3="subnet-0073a61bc56a68a3e"
+
+kops_cmd="""
+kops create cluster \
+ --vpc "$VPC_ID" \
+ --cloud aws \
+ --ssh-public-key "$HOME/.ssh/id_rsa.pub" \
+ --topology private --kubernetes-version "1.25.2" \
+ --network-cidr "$Network_CIDR" \
+ --networking calico \
+ --container-runtime containerd \
+ --image $AMI_Image \
+ --zones ap-northeast-2a,ap-northeast-2b,ap-northeast-2c \
+ --master-count 3 \
+ --master-size t3.small \
+ --node-count 3 \
+ --node-size t3.small \
+ --node-volume-size 100 \
+ --subnets "$Private_Subnet_ID_1,$Private_Subnet_ID_2,$Private_Subnet_ID_3" \
+ --utility-subnets "$Public_Subnet_ID_1,$Public_Subnet_ID_2,$Public_Subnet_ID_3" \
+ -v 10
+"""
+
+kubeconfig="kops export kubecfg --admin --kubeconfig $HOME/.kube/config --name=${KOPS_CLUSTER_NAME} --state=${KOPS_STATE_STORE}"
+echo ${kubeconfig}
+echo
+echo
+echo ${kops_cmd}

scripts/spot_lambda/lambda_restart.sh

@@ -0,0 +1,102 @@
+if [ -z "$BASH_VERSION" ]; then exec bash "$0" "$@"; exit; fi
+
+#----------------------------------------------------------------------------------------------------------------
+# 변수 선언
+echo_line="=========================================================================="
+search_tag="spot"
+stop_function="spot_stop"
+start_function="spot_start"
+
+#----------------------------------------------------------------------------------------------------------------
+# aws 인스턴스 조회
+_get_aws_status (){
+    aws_query="Reservations[].Instances[].[ InstanceId, Tags[?Key=='Name'].Value|[0] ]"
+    aws_filter="Name=instance-state-name,Values=running"
+    current_ec2_num=`aws ec2 describe-instances --query "${aws_query}" --filter ${aws_filter} --output text | grep ${search_tag} | wc -l`
+}
+
+#----------------------------------------------------------------------------------------------------------------
+# log 출력
+_get_time_log (){
+    datetime=`date +'%Y-%m-%d %H:%M:%S'`
+    string=$1
+    echo "[${datetime}] ${string}"
+}
+
+#----------------------------------------------------------------------------------------------------------------
+# aws lambda 실행 후 인스턴스 조회하여 성공/실패 여부 체크
+
+_main (){
+    lambda_function=$1
+    echo ${echo_line}
+    if [[ ${lambda_fuction} == *"stop"*  ]]; then
+        what_func='stop'
+    else
+        what_func='start'
+    fi
+
+    _get_time_log "[INFO] aws lambda ${lambda_function} start"
+
+    aws lambda invoke --function-name ${lambda_function} --cli-binary-format raw-in-base64-out --payload '{ "key": "value" }' response.json > /tmp/aws_func_result
+    #echo "aws lambda invoke --function-name ${lambda_function} --cli-binary-format raw-in-base64-out --payload '{ "key": "value" }' response.json"
+
+    succ_flag=0
+    while read line
+    do
+        if [[ $line == *"200"* ]]; then
+          succ_flag=1
+          _get_time_log "[INFO] aws lambda ${lambda_function} success!"
+          break
+        else
+          succ_flag=0
+        fi
+    done < /tmp/aws_func_result
+
+    if [[ ${succ_flag} == 1 ]]; then
+        aws lambda invoke --function-name ${lambda_function} out --log-type Tail --query 'LogResult' --output text |  base64 -d > /tmp/aws_func_output
+        if [[ ${what_func} == 'stop' ]]; then
+            final_ec2_num=0
+        else
+            final_ec2_num=`grep "Starting" /tmp/aws_func_output | wc -l`
+        fi
+
+        num=0
+        while true
+        do
+            _get_aws_status
+            num=`echo $(( $num + 1 ))`
+            if [[ ${current_ec2_num} == ${final_ec2_num} ]]; then
+                _get_time_log "[INFO] aws ${search_tag} instance number: ${final_ec2_num} check success!"
+                break
+            elif [[ ${num} == 30 ]]; then
+                _get_time_log "[Error] time out!"
+                break
+            else
+                _get_time_log "[Retry: ${num}] aws ${search_tag} instance number: ${final_ec2_num} checking..."
+            fi
+            sleep 5
+        done
+    else
+        echo ${echo_line}
+        _get_time_log "[Error] aws lambda ${lambda_function} failed!"
+        exit
+    fi
+    echo ${echo_line}
+    #---------------
+
+}
+#----------------------------------------------------------------------------------------------------------------
+# 실행
+
+_main ${stop_function}
+
+_main ${start_function}
+
+#----------------------------------------------------------------------------------------------------------------
+# 임시 파일 삭제
+
+unlink out
+unlink response.json
+unlink /tmp/aws_func_output
+unlink /tmp/aws_func_result
+#----------------------------------------------------------------------------------------------------------------

scripts/spot_lambda/lambda_stop.sh

@@ -0,0 +1,100 @@
+if [ -z "$BASH_VERSION" ]; then exec bash "$0" "$@"; exit; fi
+
+#----------------------------------------------------------------------------------------------------------------
+# 변수 선언
+echo_line="=========================================================================="
+search_tag="spot"
+stop_function="spot_stop"
+start_function="spot_start"
+
+#----------------------------------------------------------------------------------------------------------------
+# aws 인스턴스 조회
+_get_aws_status (){
+    aws_query="Reservations[].Instances[].[ InstanceId, Tags[?Key=='Name'].Value|[0] ]"
+    aws_filter="Name=instance-state-name,Values=running"
+    current_ec2_num=`aws ec2 describe-instances --query "${aws_query}" --filter ${aws_filter} --output text | grep ${search_tag} | wc -l`
+}
+
+#----------------------------------------------------------------------------------------------------------------
+# log 출력
+_get_time_log (){
+    datetime=`date +'%Y-%m-%d %H:%M:%S'`
+    string=$1
+    echo "[${datetime}] ${string}"
+}
+
+#----------------------------------------------------------------------------------------------------------------
+# aws lambda 실행 후 인스턴스 조회하여 성공/실패 여부 체크
+
+_main (){
+    lambda_function=$1
+    echo ${echo_line}
+    if [[ ${lambda_fuction} == *"stop"*  ]]; then
+        what_func='stop'
+    else
+        what_func='start'
+    fi
+
+    _get_time_log "[INFO] aws lambda ${lambda_function} start"
+
+    aws lambda invoke --function-name ${lambda_function} --cli-binary-format raw-in-base64-out --payload '{ "key": "value" }' response.json > /tmp/aws_func_result
+    #echo "aws lambda invoke --function-name ${lambda_function} --cli-binary-format raw-in-base64-out --payload '{ "key": "value" }' response.json"
+
+    succ_flag=0
+    while read line
+    do
+        if [[ $line == *"200"* ]]; then
+          succ_flag=1
+          _get_time_log "[INFO] aws lambda ${lambda_function} success!"
+          break
+        else
+          succ_flag=0
+        fi
+    done < /tmp/aws_func_result
+
+    if [[ ${succ_flag} == 1 ]]; then
+        aws lambda invoke --function-name ${lambda_function} out --log-type Tail --query 'LogResult' --output text |  base64 -d > /tmp/aws_func_output
+        if [[ ${what_func} == 'stop' ]]; then
+            final_ec2_num=0
+        else
+            final_ec2_num=`grep "Starting" /tmp/aws_func_output | wc -l`
+        fi
+
+        num=0
+        while true
+        do
+            _get_aws_status
+            num=`echo $(( $num + 1 ))`
+            if [[ ${current_ec2_num} == ${final_ec2_num} ]]; then
+                _get_time_log "[INFO] aws ${search_tag} instance number: ${final_ec2_num} check success!"
+                break
+            elif [[ ${num} == 30 ]]; then
+                _get_time_log "[Error] time out!"
+                break
+            else
+                _get_time_log "[Retry: ${num}] aws ${search_tag} instance number: ${final_ec2_num} checking..."
+            fi
+            sleep 5
+        done
+    else
+        echo ${echo_line}
+        _get_time_log "[Error] aws lambda ${lambda_function} failed!"
+        exit
+    fi
+    echo ${echo_line}
+    #---------------
+
+}
+#----------------------------------------------------------------------------------------------------------------
+# 실행
+
+_main ${stop_function}
+
+#----------------------------------------------------------------------------------------------------------------
+# 임시 파일 삭제
+
+unlink out
+unlink response.json
+unlink /tmp/aws_func_output
+unlink /tmp/aws_func_result
+#----------------------------------------------------------------------------------------------------------------