目录
ETCD 是一个高可用的分布式键值数据库,可用于服务发现。ETCD 采用 raft 一致性算法,基于 Go 语言实现。
01.集群部署基本信息
etcd_version: v3.4.6
etcd_base_dir: /var/lib/etcdetcd_data_dir: "/var/lib/etcd/default.etcd"
etcd_listen_port: "2379"
etcd_peer_port: "2380"
etcd_bin_dir: /srv/kubernetes/bin
etcd_conf_dir: /srv/kubernetes/conf
etcd_pki_dir: /srv/kubernetes/pki
部署主机:
10.40.58.153
10.40.58.154
10.40.58.116
02.ETCD凭证
创建 ETCD
凭证签发请求文件:
cat > etcd-csr.json <<EOF
{
"CN": "etcd",
"hosts": [
"10.40.61.116",
"10.40.58.153",
"10.40.58.154",
"127.0.0.1"
],
"key": {
"algo": "rsa",
"size": 2048
},
"names": [
{
"C": "CN",
"L": "BeiJing",
"O": "kubernetes",
"OU": "System",
"ST": "BeiJing"
}
]
}
EOF
其中hosts请填写部署etcd的主机IP地址, 如果使用域名请将域名一并填写
创建 ETCD
凭证和私钥:
cfssl gencert \
-ca=ca.pem \
-ca-key=ca-key.pem \
-config=ca-config.json \
-profile=kubernetes \
etcd-csr.json | cfssljson -bare etcd
结果将生成以下两个文件
etcd-key.pem
etcd.pem
03.部署
下载软件包
使用如下命令初始化etcd的运行环境及下载etcd程序并安装
mkdir -p /var/lib/etcd/default.etcd
mkdir -p /srv/kubernetes/bin
mkdir -p /srv/kubernetes/pki
mkdir -p /srv/kubernetes/conf
ETCD_VER=v3.4.6
GITHUB_URL=https://github.com/etcd-io/etcd/releases/download
DOWNLOAD_URL=${GITHUB_URL}
rm -f /tmp/etcd-${ETCD_VER}-linux-amd64.tar.gz
rm -rf /tmp/etcd-download-test && mkdir -p /tmp/etcd-download-test
curl -L ${DOWNLOAD_URL}/${ETCD_VER}/etcd-${ETCD_VER}-linux-amd64.tar.gz -o /tmp/etcd-${ETCD_VER}-linux-amd64.tar.gz
tar xzvf /tmp/etcd-${ETCD_VER}-linux-amd64.tar.gz -C /tmp/etcd-download-test --strip-components=1
cp /tmp/etcd-download-test/{etcd,etcdctl} /srv/kubernetes/bin
配置文件维护
节点1
NODENAME="py-modelo2o08cn-p005"
THISIPADDRESS="10.40.61.116"
CLUSTER="py-modelo2o08cn-p005=https://10.40.61.116:2380,\
py-modelo2o08cn-p003=https://10.40.58.153:2380,\
py-modelo2o08cn-p004=https://10.40.58.154:2380"
节点2
NODENAME="py-modelo2o08cn-p003"
THISIPADDRESS="10.40.58.153"
CLUSTER="py-modelo2o08cn-p005=https://10.40.61.116:2380,\
py-modelo2o08cn-p003=https://10.40.58.153:2380,\
py-modelo2o08cn-p004=https://10.40.58.154:2380"
节点3
NODENAME="py-modelo2o08cn-p004"
THISIPADDRESS="10.40.58.154"
CLUSTER="py-modelo2o08cn-p005=https://10.40.61.116:2380,\
py-modelo2o08cn-p003=https://10.40.58.153:2380,\
py-modelo2o08cn-p004=https://10.40.58.154:2380"
分别登录没台节点机器, 使上面的环境变量生效, 执行下面的命令
cat > /srv/kubernetes/conf/etcd.yaml <<EOF
name: ${NODENAME}
wal-dir:
data-dir: /var/lib/etcd/default.etcd
max-snapshots: 10
max-wals: 10
snapshot-count: 10
listen-peer-urls: https://${THISIPADDRESS}:2380
listen-client-urls: https://${THISIPADDRESS}:2379,https://127.0.0.1:2379
advertise-client-urls: https://${THISIPADDRESS}:2379
initial-advertise-peer-urls: https://${THISIPADDRESS}:2380
initial-cluster: ${CLUSTER}
initial-cluster-token: kube-etcd-cluster
initial-cluster-state: new
client-transport-security:
cert-file: /srv/kubernetes/pki/etcd.pem
key-file: /srv/kubernetes/pki/etcd-key.pem
client-cert-auth: true
trusted-ca-file: /srv/kubernetes/pki/ca.pem
auto-tls: false
peer-transport-security:
cert-file: /srv/kubernetes/pki/etcd.pem
key-file: /srv/kubernetes/pki/etcd-key.pem
client-cert-auth: true
trusted-ca-file: /srv/kubernetes/pki/ca.pem
auto-tls: false
debug: true
logger: zap
log-outputs: [stderr]
EOF
如果需要etcd api v2 的支持, 请配置
enable-v2: true
使用system管理etcd
创建service文件
cat > /etc/systemd/system/etcd.service <<EOF
[Unit]
Description=Etcd Server
After=network.target
[Service]
WorkingDirectory=/var/lib/etcd
ExecStart=/srv/kubernetes/bin/etcd --config-file=/srv/kubernetes/conf/etcd.yaml
Type=notify
[Install]
WantedBy=multi-user.target
EOF
配置etcd开机启动
sudo /bin/systemctl daemon-reload
sudo /bin/systemctl enable etcd.service
用下面的命令启动或者停止etcd
sudo systemctl start etcd.service
sudo systemctl stop etcd.service
验证查看集群状态
为了方便使用配置etcdctl别名, 配置完成后退出重新登录主机
cat >> ~/.bashrc << EOF
alias etcdctl="/srv/kubernetes/bin/etcdctl \
--endpoints=https://10.40.58.153:2379,https://10.40.58.154:2379,https://10.40.61.116:2379 \
--cacert=/srv/kubernetes/pki/ca.pem \
--cert=/srv/kubernetes/pki/etcd.pem \
--key=/srv/kubernetes/pki/etcd-key.pem"
EOF
查看集群状况
etcdctl endpoint health
output
https://10.40.61.116:2379 is healthy: successfully committed proposal: took = 17.824976ms
https://10.40.58.154:2379 is healthy: successfully committed proposal: took = 18.437575ms
https://10.40.58.153:2379 is healthy: successfully committed proposal: took = 19.917812ms
04.架构及内部机制解析
raft一致性算法
-
节点数量
etcd集群节点的数量根据raft一致性算法中quorum的概念, 集群中必须存在(n+1)/2个节点才可以对外提供服务, 如果一个集群中有3个节点, 那么集群可以对外提供服务的最小节点数量为
(3+1)/2=2
个, 也就只允许其中一个节点故障 -
数据写入
etcd的集群由3个到5个节点组成, 使用raft一致性算法完成一致性协同, 算法选举leader并有leader完成数据的同步和数据的分发, 当leader发生故障时选举出新的leader并由新的leader完成数据的同步, 客户端连接到etcd的集群时只需要连接其中的任意一个节点就可以完成数据的读写
Api介绍
etcd 提供的接口可以划分为5组:
-
PUT(Key. Value)
/Delete(Key)
使用put接口提供Key和Value写入数据
使用Delete接口提供Key删除数据
-
GET(key)
/Get(keyFrom, keyEnd)
get(key) 查询指定key的value
get(keyFrom, keyEnd) 指定一个Key的范围进行查询
-
Watch(Key/keyPrefix)0
使用watch接口可以实时订阅etcd中数据更新
-
Transaction(if/then/else ops).commit()
etcd提供了事物支持, 可以指定条件满足时执行某些操作
-
Leases: Grant/Revoke/Keepalive
>
数据的版本控制
全局版本
-
etcd中有个term的概念, 代表的是整个集群leader的任期, 当leader发生切换时term的数值题+1
-
etcd中还有一个revision, revision代表全局数据的版本, 当数据发生变更如: 增、删、改、查都会是revision加一。 当集群的term不发生变化时revision递增
KeyValue数据版本
- Create_revision, 是keyvalue创建时的revision
- mod_revision, 对数据进行操作,数据发生变更时的revision
- version, 是一个计数器, 表示keyvalue被修改了多少次
验证
使用如下指令我们可以获得key的版本信息
etcdctl get name -w json |jq
{
"header": {
"cluster_id": 9796312800751810000,
"member_id": 13645171481868003000,
"revision": 15,
"raft_term": 54
},
"kvs": [
{
"key": "bmFtZQ==",
"create_revision": 15,
"mod_revision": 15,
"version": 1,
"value": "dG9t"
}
],
"count": 1
}
执行如下指令进行修改, 修改完成后进行对比
$ etcdctl put name alex
OK
$ etcdctl get name -w json |jq
{
"header": {
"cluster_id": 9796312800751810000,
"member_id": 13645171481868003000,
"revision": 16,
"raft_term": 54
},
"kvs": [
{
"key": "bmFtZQ==",
"create_revision": 15,
"mod_revision": 16,
"version": 2,
"value": "YWxleA=="
}
],
"count": 1
}
通过如上数据的对比我们可以得出以下结论,由于集群没有进行新的选举所以term的verison没有发生变化, 修改了一次数据以后全局的revision
加了1, 数据的create_revision
为数据写入时全局的revision, 由于进行了数据的修改所以mod_revision
递增+1, 数据修改了两次所以version
等于2
05.etcdctl命令详解
etcdctl是etcd的命令行工具, 可以使用此命令和etcd进行交互, 默认etcdctl使用v2
的API, 如果需要使用V3
的API请设置环境变量, 命令如下:
$ export ETCDCTL_API=3
数据的写入
$ etcdctl put key value
数据的查询
-
指定单个key查询
$ etcdctl get key
-
指定key的范围进行查询
写入数据
$ etcdctl put key0 value0 OK $ etcdctl put key1 value1 OK $ etcdctl put key2 value2 OK $ etcdctl put key3 value3 OK $ etcdctl put key4 value4 OK $ etcdctl put key5 value5 OK
查询数据
$ etcdctl get key0 key5 key0 value0 key1 value1 key2 value2 key3 value3 key4 value4
根据查询的结果我们可以知道查询的结果是不包含结束的key
-
查询指定前缀的key
etcdctl get --prefix key key value key0 value0 key1 value1 key2 value2 key3 value3 key4 value4 key5 value5
-
查看置顶revision的key
$ etcdctl get name --rev=16 name alex $ etcdctl get name --rev=15 name tom
-
查看集群中所有数据
etcdctl get / --prefix --keys-only
06.集群维护
节点增加
增加节点10.40.58.152
为新节点, 请先执行 下载软件包进行环境初始化
更新证书
修改证书的csr文件重新生成证书的公钥和私钥, hosts字段添加新的主机IP地址
cat > etcd-csr.json <<EOF
{
"CN": "etcd",
"hosts": [
"10.40.61.116",
"10.40.58.153",
"10.40.58.154",
"10.40.58.152"
],
"key": {
"algo": "rsa",
"size": 2048
},
"names": [
{
"C": "CN",
"L": "BeiJing",
"O": "kubernetes",
"OU": "System",
"ST": "BeiJing"
}
]
}
EOF
创建 admin
client 凭证和私钥:
cfssl gencert \
-ca=ca.pem \
-ca-key=ca-key.pem \
-config=ca-config.json \
-profile=kubernetes \
etcd-csr.json | cfssljson -bare etcd
结果将生成以下两个文件
etcd-key.pem
etcd.pem
复制 TLS 证书和密钥对
rsync -auv etcd-key.pem etcd.pem ca.pem root@10.40.58.152:/src/kubernetes/pki/
rsync -auv etcd-key.pem etcd.pem ca.pem root@10.40.58.153:/srv/kubernetes/pki
rsync -auv etcd-key.pem etcd.pem ca.pem root@10.40.58.154:/srv/kubernetes/pki
rsync -auv etcd-key.pem etcd.pem ca.pem /srv/kubernetes/pki/
重启etcd Cluster
$ sudo systemctl restart etcd.service
启动新实例
请根据如下命令生成etcd的配置文件, 完成后请根据使用system管理etcd配置systemd, 并启动节点
NODENAME="py-modelo2o08cn-p002"
THISIPADDRESS="10.40.58.152"
CLUSTER="py-modelo2o08cn-p005=https://10.40.61.116:2380,\
py-modelo2o08cn-p003=https://10.40.58.153:2380,\
py-modelo2o08cn-p004=https://10.40.58.154:2380,\
py-modelo2o08cn-p002=https://10.40.58.152:2380"
分别登录没台节点机器, 使上面的环境变量生效, 执行下面的命令
cat > /srv/kubernetes/conf/etcd.yaml <<EOF
name: ${NODENAME}
wal-dir:
data-dir: /var/lib/etcd/default.etcd
max-snapshots: 10
max-wals: 10
listen-peer-urls: https://${THISIPADDRESS}:2380
listen-client-urls: https://${THISIPADDRESS}:2379,https://127.0.0.1:2379
advertise-client-urls: https://${THISIPADDRESS}:2379
initial-advertise-peer-urls: https://${THISIPADDRESS}:2380
initial-cluster: ${CLUSTER}
initial-cluster-token: kube-etcd-cluster
initial-cluster-state: existing
client-transport-security:
cert-file: /srv/kubernetes/pki/etcd.pem
key-file: /srv/kubernetes/pki/etcd-key.pem
client-cert-auth: true
trusted-ca-file: /srv/kubernetes/pki/ca.pem
auto-tls: false
peer-transport-security:
cert-file: /srv/kubernetes/pki/etcd.pem
key-file: /srv/kubernetes/pki/etcd-key.pem
client-cert-auth: true
trusted-ca-file: /srv/kubernetes/pki/ca.pem
auto-tls: false
debug: true
logger: zap
log-outputs: [stderr]
EOF
添加实例
etcdctl member add py-modelo2o08cn-p002 --peer-urls=https://10.40.58.152:2380
节点删除
获取节点的ID
$ etcdctl member list
690eb5228cd49828, started, py-modelo2o08cn-p002, https://10.40.58.152:2380, https://10.40.58.152:2379, false
7064f95d4211e35b, started, py-modelo2o08cn-p003, https://10.40.58.153:2380, https://10.40.58.153:2379, false
b54dd19729976a3f, started, py-modelo2o08cn-p004, https://10.40.58.154:2380, https://10.40.58.154:2379, false
bd5d632ae4086bfd, started, py-modelo2o08cn-p005, https://10.40.61.116:2380, https://10.40.61.116:2379, false
删除member
$ etcdctl member remove 690eb5228cd49828
Member 690eb5228cd49828 removed from cluster 87f37e96d56c7453
查看集群的健康状态
$ etcdctl endpoint health
https://10.40.61.116:2379 is healthy: successfully committed proposal: took = 17.168768ms
https://10.40.58.154:2379 is healthy: successfully committed proposal: took = 21.879205ms
https://10.40.58.153:2379 is healthy: successfully committed proposal: took = 21.980464ms
数据备份
--snapshot-count
:指定有多少事务(transaction)被提交时,触发截取快照保存到磁盘,在v3.2之前的版本,默认的参数是10000条,3.2之后调整为100000条。 这个条目数量不能配置过高或者过低,过低会导致频繁的io压力,过高会导致占用高内存以及会导致etcd GC过慢。建议设置为10W-20W条。
--max-snapshots '5'
: 最大保留多少快照文件
配置snaphost, 添加如下内容到etcd.yaml文件中, 其中snapshot-count
为了测试设置为10
max-snapshots: 10
max-wals: 10
snapshot-count: 10
执行相关提交10次后查看日志发现snapshot已经保存
Apr 5 10:34:11 py-modelo2o08cn-p003 etcd: {"level":"info","ts":"2020-04-05T10:34:11.480+0800","caller":"etcdserver/server.go:2381","msg":"saved snapshot","snapshot-index":64}
备份的snap文件存在/var/lib/etcd/default.etcd/member/snap
下
数据恢复
本文的测试用例为3台节点其中一台节点所有原始文件全部丢失, 查看集群的健康状态如下:
其中10.40.61.116节点状态异常, 我们删掉了data-dir下的所有文件并停止了服务
$ etcdctl endpoint status
{"level":"warn","ts":"2020-04-05T14:38:36.309+0800","caller":"clientv3/retry_interceptor.go:62","msg":"retrying of unary invoker failed","target":"passthrough:///https://10.40.61.116:2379","attempt":0,"error":"rpc error: code = DeadlineExceeded desc = latest connection error: connection error: desc = \"transport: Error while dialing dial tcp 10.40.61.116:2379: connect: connection refused\""}
Failed to get the status of endpoint https://10.40.61.116:2379 (context deadline exceeded)
https://10.40.58.153:2379, 7064f95d4211e35b, 3.4.6, 20 kB, true, false, 5, 25, 25,
https://10.40.58.154:2379, b54dd19729976a3f, 3.4.6, 20 kB, false, false, 5, 25, 25,
创建快照
从正在运行的节点上执行如下执行
- 写入测试数据
$ etcdctl put b 1
- 生成快照
$ etcdctl snapshot save snapshot.db
- 将快照cp到
10.40.61.116
主机
$ rsync -auv snapshot.db root@10.40.61.116:/root
恢复快照
$ etcdctl snapshot restore snapshot.db
验证
$ etcdctl endpoint health
$ etcdctl get b
Q&A
Q:
$ /srv/kubernetes/bin/etcdctl --endpoints=https://10.40.58.153:2379,https://10.40.58.154:2379,https://10.40.61.116:2379 --cert-file=/srv/kubernetes/pki/etcd.pem --key-file=/srv/kubernetes/pki/etcd-key.pem --ca-file /srv/kubernetes/pki/ca.pem --debug ls /
Error: client: response is invalid json. The endpoint is probably not valid etcd cluster endpoint
$ curl -X GET https://10.40.61.116:2379/v2/members --cacert /root/certificated/ca.pem --cert /root/certificated/etcd.pem --key /root/certificated/etcd-key.pem
404 page not found
A:
上面报错的主要是因为当前集群没有开通api v2的支持, 请在/srv/kubernetes/conf/etcd.yaml
文件中添加v2支持
enable-v2: true
网友评论