背景

Prometheus是业内有名的开源监控工具，我所在的公司也是采用Prometheus+Grafana方式构建监控系统，并且不只是监控运维层面的数据，业务层面的服务状态也通过Java代码的客户端micrometer向Prometheus提交数据并在Grafana上配置出图，方便运营和技术同学进行查看。

在有业务数据场景下监控体系的高可用性显得尤为重要，为此特意调研各个时序数据库，力求实现简单方便的高可用方案。

在调研时看到了influxdb，虽然influxdb3.0还未发布集群开源方案，但通过订阅的方式仍然可以做到高可用的方式。

比起使用TDengine集群的方式我认为本方案更节省机器资源和维护成本。

架构

解释：虚线代表Prometheus master节点的Prometheus服务故障之后的线路。

IP地址	主机名	操作系统	软件	网卡名
10.0.0.10	grafana	Ubuntu 20.04.4 LTS arm64架构	grafana_10.3.1	ens160
10.0.0.11	prometheus01	Ubuntu 20.04.4 LTS arm64架构	node_exporter-1.7.0, prometheus 2.45.3, keepalived	ens160
10.0.0.12	influxdb-master	Ubuntu 20.04.4 LTS arm64架构	influxdb-1.8.10	ens160
10.0.0.13	prometheus02	Ubuntu 20.04.4 LTS arm64架构	node_exporter-1.7.0, prometheus 2.45.3, keepalived	ens160
10.0.0.14	influxdb-slave	Ubuntu 20.04.4 LTS arm64架构	influxdb-1.8.10	ens160

解释：

在10.0.0.11和10.0.0.12上分别部署node_exporter-1.7.0, prometheus 2.45.3, keepalived，并且将Prometheus01的权重调高，Prometheus02处于stop状态；
Grafana连接Prometheus的VIP地址；
Prometheus01 读写设置为influxdb_master节点； 
Prometheus02 读写设置为influxdb_slave节点；
编写Keepalived脚本 实现当Prometheus01故障时 自动开启Prometheus02；
influxdb_master和influxdb_slave通过subscription的方式实现数据实时传送。

操作步骤

1.修改主机名

hostnamectl set-hostname grafana
hostnamectl set-hostname prometheus01
hostnamectl set-hostname influxdb-master
hostnamectl set-hostname prometheus02
hostnamectl set-hostname influxdb-slave

2.设置时区以及同步时间

全部都要机器都要操作

# 设置时区
timedatectl set-timezone Asia/Shanghai
# 安装基础软件
apt install -y lrzsz net-tools ntpdate
# 同步时间
/usr/sbin/ntpdate ntp1.aliyun.com
crontab -l > crontab_conf ; echo "*/5 * * * * /usr/sbin/ntpdate ntp1.aliyun.com >/dev/null 2>&1" >> crontab_conf && crontab crontab_conf && rm -f crontab_conf
timedatectl set-local-rtc 1

3.安装prometheus

只是安装Prometheus暂时不要启动,后面对配置文件更改后再启动prometheus01,02不启动；

wget https://github.com/prometheus/prometheus/releases/download/v2.45.3/prometheus-2.45.3.linux-arm64.tar.gz
mv prometheus-2.45.3.linux-arm64.tar.gz /etc/
cd /etc
tar -zxvf prometheus-2.45.3.linux-arm64.tar.gz
mv prometheus-2.45.3.linux-arm64 prometheus
rm -rf prometheus-2.45.3.linux-arm64.tar.gz
cd prometheus
mkdir data
mv prometheus promtool  /usr/local/bin/cat > /etc/systemd/system/prometheus.service << EOF
[Unit]
Description=Prometheus
Documentation=https://prometheus.io/
After=network.target
[Service]
ExecStart=/usr/local/bin/prometheus --config.file=/etc/prometheus/prometheus.yml --storage.tsdb.path=/etc/prometheus/data --web.listen-address=0.0.0.0:9090
WorkingDirectory=/etc/prometheus/
Restart=on-failure
[Install]
WantedBy=multi-user.target
EOF# 安装influxdb-client
apt install -y influxdb-client

4.安装node_export

我这里只安装在Prometheus所在机器作为示例，实际情况请按需配置。

wget https://github.com/prometheus/node_exporter/releases/download/v1.7.0/node_exporter-1.7.0.linux-arm64.tar.gz
tar -xvzf node_exporter-1.7.0.linux-arm64.tar.gz
cp node_exporter-1.7.0.linux-arm64/node_exporter /usr/local/bin/node_exporter
rm -rf  node_exporter-1.7.0.linux-arm64*
cat > /etc/systemd/system/node_exporter.service << EOF
[Unit]
Description=node_exporter
Documentation=https://prometheus.io/
After=network.target[Service]
ExecStart=/usr/local/bin/node_exporter
Restart=on-failure
RestartSec=20[Install]
WantedBy=multi-user.target
EOFsystemctl daemon-reload
systemctl restart node_exporter
systemctl status node_exporter
systemctl enable node_exporter

5.安装Keepalived

prometheus01和prometheus02都需要进行安装并配置

apt -y install keepalived ipvsadm
systemctl enable keepalived

prometheus01机器Keepalived配置文件

cat /etc/keepalived/keepalived.conf

! Configuration File for keepalived
global_defs {router_id 1
}
vrrp_script check_prome {script "/etc/keepalived/check_prome.sh"interval 1weight -50
}
vrrp_instance VI_1 {state MASTERinterface ens160virtual_router_id 1mcast_src_ip 10.0.0.11priority 90advert_int 1authentication {auth_type PASSauth_pass 123456}track_script {check_prome}virtual_ipaddress {10.0.0.15}notify_master "/etc/keepalived/notify.sh master"notify_backup "/etc/keepalived/notify.sh backup"notify_fault "/etc/keepalived/notify.sh fault"
}

prometheus02机器Keepalived配置文件

cat /etc/keepalived/keepalived.conf

! Configuration File for keepalived
global_defs {router_id 2
}vrrp_instance VI_1 {state BACKUPinterface ens160virtual_router_id 1mcast_src_ip 10.0.0.13priority 80advert_int 1authentication {auth_type PASSauth_pass 123456}virtual_ipaddress {10.0.0.15}notify_master "/etc/keepalived/notify.sh master"notify_backup "/etc/keepalived/notify.sh backup"notify_fault "/etc/keepalived/notify.sh fault"
}

notify_master 当前节点成为主节点时触发脚本

notify_backup 当前节点成为备节点时触发脚本

notify_fault 当前节点转为"失败"状态时触发的脚本

6.设置Keepalived脚本

Keepalived主备切换时执行脚本_51CTO博客_keepalived主备切换

prometheus01 prometheus02机器都需要设置

prometheus_check.sh脚本

vim /etc/keepalived/check_prome.sh

chmod +x /etc/keepalived/check_prome.sh

记得创建脚本后赋予执行权限

#!/bin/bash
processes_num=$(ps -ef |grep [p]rometheus|wc -l)
if [ $processes_num = 0 ]
thenexit 1
elseexit 0
fi

notify.sh脚本

vim /etc/keepalived/notify.sh

chmod +x /etc/keepalived/notify.sh

记得创建脚本后赋予执行权限

master机器脚本

#!/bin/bash
case $1 inmaster)systemctl start prometheusinflux -host 10.0.0.12  -username root -password 2024.168 -database prometheus -execute """CREATE SUBSCRIPTION "prometheus" ON "prometheus"."prometheus" DESTINATIONS ALL 'http://prometheus:prometheus@2024.@10.0.0.14:8086';""";;backup)systemctl stop prometheusinflux -host 10.0.0.12 -username root -password 2024.168 -database prometheus -execute """DROP SUBSCRIPTION "prometheus" ON "prometheus"."prometheus";""";;fault)systemctl stop prometheusinflux -host 10.0.0.12  -username root -password 2024.168 -database prometheus -execute """DROP SUBSCRIPTION "prometheus" ON "prometheus"."prometheus";""";;*)echo "不支持该参数，请检查输入的参数是否正确"
esac

slave机器脚本

#!/bin/bash
case $1 inmaster)systemctl start prometheusinflux -host 10.0.0.14 -username root -password 2024.168 -database prometheus -execute """CREATE SUBSCRIPTION "prometheus" ON "prometheus"."prometheus" DESTINATIONS ALL 'http://prometheus:prometheus@2024.@10.0.0.12:8086';""";;backup)systemctl stop prometheusinflux -host 10.0.0.14 -username root -password 2024.168 -database prometheus -execute """DROP SUBSCRIPTION "prometheus" ON "prometheus"."prometheus";""";;fault)systemctl stop prometheusinflux -host 10.0.0.14 -username root -password 2024.168 -database prometheus -execute """DROP SUBSCRIPTION "prometheus" ON "prometheus"."prometheus";""";;*)echo "不支持该参数，请检查输入的参数是否正确"
esac

7.安装influxdb

wget https://dl.influxdata.com/influxdb/releases/influxdb-1.8.10_linux_arm64.tar.gz
tar xvfz influxdb-1.8.10_linux_arm64.tar.gz
sudo cp ./influxdb-1.8.10-1/usr/bin/* /usr/local/bin/
mv influxdb-1.8.10-1/etc/influxdb/  /etc/cat > /etc/systemd/system/influxd.service << EOF
[Unit]
Description=influxd
Documentation=https://docs.influxdata.com/influxdb/v1/install/?t=Linux
After=network.target[Service]
ExecStart=/usr/local/bin/influxd -config /etc/influxdb/influxdb.conf
Restart=on-failure
RestartSec=20[Install]
WantedBy=multi-user.target
EOF# 启动influxdb
systemctl daemon-reload
systemctl restart influxd
systemctl enable influxd
systemctl status influxd# 开启用户认证
vim /etc/influxdb/influxdb.conf
auth-enabled = true# 重新启动
systemctl restart influxd

8.创建管理员账户

# 创建管理员账户
CREATE USER root WITH PASSWORD '2024.168' WITH ALL PRIVILEGES> show databases;
ERR: unable to parse authentication credentials
Warning: It is possible this error is due to not setting a database.
Please set a database with the command "use <database>".# 进行认证
> auth
username: root
password: 
> show databases;
name: databases
name
----
_internal

9.创建存储监控数据的库

不需要创建表 配置Prometheus后会自动创建表;

influxdb-master 和 influxdb-slave 都需要创建

influx
auth
create database prometheus;
# 创建默认的保留策略
use prometheus
CREATE RETENTION POLICY "prometheus" ON "prometheus" DURATION 1h REPLICATION 1 DEFAULT
show databases;

回显

> show databases;
name: databases
name
----
prometheus
_internal

创建prometheus账户并设置权限

# 创建Prometheus账户并设置权限
CREATE USER prometheus WITH PASSWORD 'prometheus@2024.'
GRANT ALL ON "prometheus" TO "prometheus"
SHOW USERS

Influxdb-master手动创建订阅

use prometheus
CREATE SUBSCRIPTION "prometheus" ON "prometheus"."prometheus" DESTINATIONS ALL 'http://prometheus:prometheus@2024.@10.0.0.14:8086'

10.配置Prometheus01

vim /etc/prometheus/prometheus.yml

global:scrape_interval: 15s evaluation_interval: 15s 
scrape_configs:- job_name: "prometheus"static_configs:- targets: ["10.0.0.11:9100", "10.0.0.13:9100"]
remote_write:- url: "http://10.0.0.12:8086/api/v1/prom/write?db=prometheus&u=prometheus&p=prometheus@2024."remote_timeout: 30squeue_config:capacity: 100000max_shards: 1000max_samples_per_send: 1000batch_send_deadline: 5smin_backoff: 30msmax_backoff: 100ms
remote_read:- url: "http://10.0.0.12:8086/api/v1/prom/read?db=prometheus&u=prometheus&p=prometheus@2024."remote_timeout: 10sread_recent: true

11.配置Prometheus02

vim /etc/prometheus/prometheus.yml

global:scrape_interval: 15s evaluation_interval: 15s 
scrape_configs:- job_name: "prometheus"static_configs:- targets: ["10.0.0.11:9100", "10.0.0.13:9100"]
remote_write:- url: "http://10.0.0.14:8086/api/v1/prom/write?db=prometheus&u=prometheus&p=prometheus@2024."remote_timeout: 30squeue_config:capacity: 100000max_shards: 1000max_samples_per_send: 1000batch_send_deadline: 5smin_backoff: 30msmax_backoff: 100ms
remote_read:- url: "http://10.0.0.14:8086/api/v1/prom/read?db=prometheus&u=prometheus&p=prometheus@2024."remote_timeout: 10sread_recent: true

12.启动Prometheus01

systemctl daemon-reload
systemctl start prometheus
systemctl enable prometheus
systemctl restart prometheus
systemctl status prometheus

启动keepalived并检查VIP

# 启动
systemctl start  keepalived
# 检查VIP
root@prometheus01:~# ip a
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00inet 127.0.0.1/8 scope host lovalid_lft forever preferred_lft foreverinet6 ::1/128 scope host valid_lft forever preferred_lft forever
2: ens160: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc fq_codel state UP group default qlen 1000link/ether 00:0c:29:af:7d:17 brd ff:ff:ff:ff:ff:ffinet 10.0.0.11/24 brd 10.0.0.255 scope global ens160valid_lft forever preferred_lft foreverinet 10.0.0.15/32 scope global ens160valid_lft forever preferred_lft foreverinet6 fe80::20c:29ff:feaf:7d17/64 scope link valid_lft forever preferred_lft forever

13.安装grafana

sudo apt-get install -y adduser libfontconfig1 musl
wget https://dl.grafana.com/oss/release/grafana_10.3.1_arm64.deb
sudo dpkg -i grafana_10.3.1_arm64.deb
# 启动Grafana
sudo /bin/systemctl daemon-reload
sudo /bin/systemctl enable grafana-server
sudo /bin/systemctl start grafana-server
# 默认账户名密码
admin/admin
# 配置数据源时设置VIP地址 切记！！！# 导入Linux监控面板
8919

14.查看influxdb数据

use prometheus
show measurements

回显

> select * from go_info;
name: go_info
time                __name__ instance       job        value version
----                -------- --------       ---        ----- -------
1706866563610000000 go_info  127.0.0.1:9100 prometheus 1     go1.21.4
1706866578610000000 go_info  127.0.0.1:9100 prometheus 1     go1.21.4
1706866593610000000 go_info  127.0.0.1:9100 prometheus 1     go1.21.4
1706866608610000000 go_info  127.0.0.1:9100 prometheus 1     go1.21.4
1706866623610000000 go_info  127.0.0.1:9100 prometheus 1     go1.21.4
1706866638610000000 go_info  127.0.0.1:9100 prometheus 1     go1.21.4
1706866653612000000 go_info  127.0.0.1:9100 prometheus 1     go1.21.4
> 

15.实验验证

在压测磁盘的时候进行操作：

1、 关闭Prometheus01，查看VIP是否会转移到Prometheus02 并且自动设置Prometheus启动

2、启动Prometheus01，查看VIP是否回到了Prometheus01 并且Prometheus02的Prometheus服务关闭

验证操作过程中 显示磁盘读写的图表是否中断，全程图表没有中断则说明高可用架构已经实现。

# 压测命令
while true; do  dd if=/dev/nvme0n1p2  of=/testrw.dbf bs=4k && rm -rf /testrw.dbf; done

测试效果的另外一种验证方式

root@influxdb-slave:~# influx -username root -password 2024.168 -database prometheus -execute 'select * from node_cpu_seconds_total;' > 1.txt && wc -l 1.txt
771 1.txt
root@influxdb-slave:~# root@influxdb-master:~# influx -username root -password 2024.168 -database prometheus -execute 'select * from node_cpu_seconds_total;' > 1.txt && wc -l 1.txt
771 1.txt
root@influxdb-master:~# 

保存策略已经测试过了 是OK的 不过并不会特别严格：

​ 我创建了一个保留1小时的保存策略，从15点开始测试数据 但16点不会立刻将15点的数据给清理掉。我测试的现象是 16点30分左右 15点的数据已经清理了一部分了。

补充

influxdb可视化

文档：Chronograf 文档 (influxdata.com)

创建订阅

严格按照示例中的单引号和双引号来设置，否则容易报错。
本方案中只有第一次启动Prometheus01时需要手动设置订阅，其他情况下都会通过Keepalived的notify脚本来自动设置订阅。

influxb-master操作

use prometheus
CREATE SUBSCRIPTION "prometheus" ON "prometheus"."prometheus" DESTINATIONS ALL 'http://prometheus:prometheus@2024.@10.0.0.14:8086'

Influxb-slave操作

use prometheus
CREATE SUBSCRIPTION 'prometheus' ON 'prometheus'.'prometheus' DESTINATIONS ALL 'http://prometheus:prometheus@2024.@10.0.0.12:8086'

查看订阅

> SHOW SUBSCRIPTIONS
name: prometheus
retention_policy name       mode destinations
---------------- ----       ---- ------------
autogen          prometheus ALL  [http://10.0.0.13:8086]

> SHOW SUBSCRIPTIONS
name: prometheus
retention_policy name       mode destinations
---------------- ----       ---- ------------
autogen          prometheus ALL  [http://10.0.0.12:8086]

如果要删除订阅可以采用下面的命令

DROP SUBSCRIPTION "prometheus" ON "prometheus"."autogen"

数据备份

当有备份需要时进行设置，在本方案中并不采用这种方式来备份。

InfluxDB OSS 备份和还原 | InfluxDB v1.8 Documentation (cnosdb.com)

root@influxdb:~/backup# influxd backup -portable -database  prometheus -host 127.0.0.1:8088 .
2024/02/04 10:14:49 backing up metastore to meta.00
2024/02/04 10:14:49 backing up db=prometheus
2024/02/04 10:14:49 backing up db=prometheus rp=autogen shard=2 to prometheus.autogen.00002.00 since 0001-01-01T00:00:00Z
2024/02/04 10:14:49 backup complete:
2024/02/04 10:14:49     20240204T021449Z.meta
2024/02/04 10:14:49     20240204T021449Z.s2.tar.gz
2024/02/04 10:14:49     20240204T021449Z.manifest

数据恢复

当influxdb出现故障时 用于恢复数据

root@influxdb:~/backup# influxd restore -portable -host 10.0.0.13:8088 -db prometheus .
2024/02/04 10:20:40 Restoring shard 2 live from backup 20240204T021449Z.s2.tar.gz

如果要将数据还原到已经存在的数据库:

不能直接还原到已经存在的数据库中，如果尝试将restore命令运行到现有的数据库中，则会收到以下信息:

influxd restore -portable -db existingdb path-to-backup

2018/08/30 13:42:46 error updating meta: DB metadata not changed. database may already exist

restore: DB metadata not changed. database may already exist

正确的做法是：

1. 将现有的数据备份还原到临时数据库

   influxd restore -portable -db prometheus -newdb prometheus_bak path-to-backup
   

2. 侧向加载数据 (使用 SELECT ... INTO 语句) 放入现有目标数据库，并删除临时数据库

   > USE prometheus_bak
   > SELECT * INTO prometheus..:MEASUREMENT FROM /.*/ GROUP BY *
   > DROP DATABASE prometheus_bak
   

数据清理

监控数据我们只关注最近半年或者一年的，如果单纯的是监控数据可能保留的时间更短，我们可以根据需要定时清理不需要的数据。

数据清理我这里采用保留策略的方式，超过30天的监控数据自动清理；并设置为默认策略。

CREATE RETENTION POLICY "prometheus" ON "prometheus" DURATION 30d REPLICATION 1 DEFAULT

监控influxdb

influxdb不需要通过第三方工具暴露数据采集端口，influxdb本身已经提供了该端口，但influxdb自带的监控参数太少，我们采用influxdb_exporter的方式采集influxdb指标。

10.0.0.12:8086/metrics

Telegraf 是一个插件驱动的服务器代理，用于从数据库、系统和物联网传感器收集和发送指标和事件。Telegraf 是用 Go 编写的，编译成一个没有外部依赖关系的二进制文件，并且需要非常小的内存占用。

Telegraf 文档 (influxdata.com)

安装Telegraf

wget https://dl.influxdata.com/telegraf/releases/telegraf-1.29.4_linux_arm64.tar.gz
tar xf telegraf-1.29.4_linux_arm64.tar.gz
mv telegraf-1.29.4/usr/lib/telegraf/scripts/telegraf.service /etc/systemd/system/
mv telegraf-1.29.4/etc/telegraf /etc/
mv telegraf-1.29.4/usr/bin/telegraf /usr/bin/telegraf
# 配置system管理
cat > /etc/systemd/system/telegraf.service << EOF
[Unit]
Description=Telegraf
Documentation=https://github.com/influxdata/telegraf
After=network-online.target
Wants=network-online.target[Service]
ExecStart=/usr/bin/telegraf -config /etc/telegraf/telegraf.conf -config-directory /etc/telegraf/telegraf.d $TELEGRAF_OPTS
Restart=on-failure
RestartSec=20[Install]
WantedBy=multi-user.target
EOF# 配置
vim /etc/telegraf/telegraf.conf
[[outputs.prometheus_client]]listen = ":9273"[[inputs.influxdb]]urls = ["http://localhost:8086/debug/vars"username = "prometheus"password = "prometheus@2024."# 启动
systemctl daemon-reload
systemctl restart telegraf
systemctl enable telegraf
systemctl status telegraf

修改Prometheus01 Prometheus02配置文件添加以下内容：

  - job_name: "influxdb"static_configs:- targets: ["10.0.0.12:9273", "10.0.0.14:9273"]

grafana

引入监控面板：11334

方案调整

相互订阅

最开始的influxdb-master和influxdb-slave 是采用相互订阅的方式

发现相互订阅虽然数据是不会重复的，但是会出现相互发送数据会有CPU高的问题。

动态切换订阅

现在已经更改为采用keepalived的方式，在Prometheus写入数据之前 手动设置订阅influxdb-master实时写入influxdb-slave；并且在Keepalived脚本中设置：当master变为slave时 将删除订阅，由slave变成master的机器设置订阅。

经过测试 数据不会丢失，VIP在Prometheus01 Prometheus02中来回切换 图表也不会中断，并且数据始终一致。CPU压力很小。

数据的保留策略也已经生效：我是在11点40左右开始实验，下午1：43时 数据已经只能看到12点数据了。

查看方法：

 influx -username root -password 2024.168 -database prometheus -execute 'select * from node_cpu_seconds_total;' > 1.txt && wc -l 1.txt

获取第一条数据的时间戳：1707192006851000000

root@influxdb-master:~# head -n 10 1.txt 
name: node_cpu_seconds_total
time                __name__               cpu instance       job        mode    value
----                --------               --- --------       ---        ----    -----
1707192006851000000 node_cpu_seconds_total 0   10.0.0.13:9100 prometheus idle    5383.56
1707192006851000000 node_cpu_seconds_total 0   10.0.0.13:9100 prometheus iowait  2.43
1707192006851000000 node_cpu_seconds_total 0   10.0.0.13:9100 prometheus irq     0
1707192006851000000 node_cpu_seconds_total 0   10.0.0.13:9100 prometheus nice    33.53
1707192006851000000 node_cpu_seconds_total 0   10.0.0.13:9100 prometheus softirq 0.65
1707192006851000000 node_cpu_seconds_total 0   10.0.0.13:9100 prometheus steal   0
1707192006851000000 node_cpu_seconds_total 0   10.0.0.13:9100 prometheus system  18.14
root@influxdb-master:~# 

但为什么Grafana上仍然可以看到11点40的数据 我不清楚这是什么情况，还需要对influxdb做更多的学习。

时间戳在线转换：Unix 时间戳 - Epoch 转换器 (unixtimestamp.com)

参考文档

InfluxDB 中的 Prometheus 端点支持 |InfluxDB OSS v1 文档 (influxdata.com)

Manage subscriptions in InfluxDB | InfluxDB OSS v1 Documentation (influxdata.com)

从InfluxDB OSS复制数据 |InfluxDB Cloud （TSM） 文档 (influxdata.com)

influxdb2.x 主从数据实时同步_influxdb主从-CSDN博客

Prometheus 通过 Telegraf 将数据远程写入 InfluxDB 2.x 存储（InfluxDB 2.x 不同于 1.x）_prometheus telegraf-CSDN博客

centos7 keepalived+nginx实现vip漂移高可用 - 凯文队长 - 博客园 (cnblogs.com)