grafana全家桶-loki promtail收集k8s容器日志

loki是grafana旗下轻量级日志收集工具，为了减少loki对集群的影响，把loki的agent日志收集端promtail部署在k8s集群中，loki server部署在集群外面。这样简单做一个解耦，避免大量读写的应用影响到集群内业务服务。

一、promtail部署

helm配置文件，这里简单说一下promtail会自动从每台k8s node的容器目录去收集日志，默认是按照deployment的name进行拆分job，例如dp的名称是test和baga，job就会是test和baga
这里url写集群外的loki地址就行

cat >vaules.yaml<<EOF
config:clients:- url: http://172.28.199.199:3100/loki/api/v1/pushresources:limits:cpu: 300mmemory: 300Mirequests:cpu: 100mmemory: 100Mitolerations:- effect: NoSchedulekey: node-role.kubernetes.io/master
EOF

安装promtail

kubectl create namespace logging
helm upgrade --install promtail grafana/promtail -f values.yaml --namespace logging

二、安装配置loki

2.1、docker-compose文件

cat >docker-compose.yaml <<EOF
version: '3'services:loki:image: grafana/loki:latestports:- "3100:3100"- "9095:9095"volumes:- ./loki-data:/loki- ./loki-config.yaml:/etc/loki/local-config.yamluser: rootcommand: ["-config.file=/etc/loki/local-config.yaml"]
EOF

2.2、loki-config文件

cat >loki-config.yaml <<EOF
server:http_listen_port: 3100common:                             #通用配置path_prefix: /lokistorage:filesystem:chunks_directory: /loki/chunksrules_directory: /loki/rulesreplication_factor: 1ring:kvstore:store: inmemoryschema_config:configs:- from: 2020-10-24               # 架构配置的生效日期store: boltdb-shipper          # 存储类型object_store: filesystem       # 对象存储类型为文件系统schema: v11                    # 架构版本index:                         # 索引配置prefix: index_               # 索引前缀period: 24h                  # 索引周期storage_config:                                                # 存储相关boltdb_shipper:active_index_directory: /loki/boltdb-shipper-active        # 活动索引的目录cache_location: /loki/boltdb-shipper-cache                 # 缓存位置cache_ttl: 24h                                             # 缓存的过期时间shared_store: filesystem                                   # 共享存储类型为文件系统filesystem:directory: /loki/chunks                                    # 文件系统的目录compactor:working_directory: /loki/boltdb-shipper-compactor            # 紧缩器的工作目录shared_store: filesystem                                     # 共享存储类型为文件系统limits_config:reject_old_samples: true               # 是否拒绝旧样本reject_old_samples_max_age: 72h        # 72小时之前的样本被拒绝max_entries_limit_per_query: 9999      # 数值改为自己想要的最大查询行数ingestion_rate_mb: 32                  # mb单位，每个用户每秒的采样率限制  不设置客户端会报错ingestion_burst_size_mb: 64            # 每个用户允许的采样突发大小chunk_store_config:max_look_back_period: 72h              # 为避免查询超过保留期的数据，必须小于或等于下方的时间值table_manager:retention_deletes_enabled: true        # 保留删除开启retention_period: 72h                  # 超过72h的块数据将被删除
auth_enabled: false
EOF

经过三天后发现数据保留72h是有效的

2.3、启动loki server

$ docker-compose up -d
# 这里的loki-data目录是loki的data目录，映射出来了   ./loki-data:/loki
root@ip-172-28-199-199:/data/loki# ls
docker-compose.yaml  loki-config.yaml  loki-data

2.4、写入测试数据

在没有grafana时测试loki可用性

curl -X POST http://172.28.199.199:3100/loki/api/v1/push -H "Content-Type: application/json" -d '{"streams": [{"stream": {"job": "test-job","host": "example"},"values": [[ "'$(date +%s%N)'", "Hello Loki from test-job" ]]}]
}'

2.5、查看loki的job信息

$ curl "http://172.28.199.199:3100/loki/api/v1/label/job/values"
{"status":"success","data":["logging/promtail","sijia-test/vault-demo","test-job","vault/vault","vault/vault-agent-injector","velero/velero"]}

三、部署grafana

loki的ui就是grafana

3.1、docker-compose文件

直接把ldap也配置了

version: '3.7'services:grafana:image: grafana/grafana:latestcontainer_name: "grafana"ports:- "3000:3000"restart: alwaysvolumes:- "./grafana_data:/var/lib/grafana"- "./ldap.toml:/etc/grafana/ldap.toml"environment:- GF_AUTH_LDAP_ENABLED=true- GF_AUTH_LDAP_CONFIG_FILE=/etc/grafana/ldap.toml

3.2、ldap配置

ldap.toml，这是测试环境，所以用户权限都给Admin了

[[servers]]
# LDAP 服务器的地址和端口
host = "172.28.224.89"
port = 389
use_ssl = false
start_tls = false
ssl_skip_verify = false# 用于绑定 LDAP 服务器的用户凭证
bind_dn = "cn=admin,dc=xx,dc=com"
bind_password = 'xx.compassword'# 搜索过滤器和基础 DN，用于查找用户
search_filter = "(uid=%s)"
search_base_dns = ["dc=xx,dc=com"]# LDAP 属性映射
[servers.attributes]
name = "uid"
surname = "sn"
username = "cn"
member_of = "dn"
email = "email"# 组映射，所有用户默认分配 Admin 角色
[[servers.group_mappings]]
group_dn = "*"
org_role = "Admin"