项目基于 mimirtool 进行封装，实现了对 prometheus 的指标基数优化；
其能实现根据指定的 grafana + prometheus rule 文件 分析出当前所使用的指标，同时分析 TSDB 侧来得出哪些 指标是未被使用的，分析生成优化relabel配置；
拿到结果后，我们就可以在 metric_relabel/write_relabel 阶段(根据需求)，来对相关无用指标进行 keep，从而实现 prometheus series的精简， 优化磁盘大小，优化内存，降低笛卡尔积，查询提速；
前提，准备好grafana api-key + 当前所用到的 prometheus rule/record 文件；
grafana: remote_url: http://10.0.0.100:30030 # 需要分析的grafana的http地址 api_token: eyJrIjoiVjdJdGkzczFERmJ2dTB1WkRZZGRJQ05GeVBDNUt4SmUiLCJuIjoibWltaXJ0b29sIiwiaWQiOjF9 # 需要申请grafana api-key prometheus: remote_url: http://10.0.0.100:30090 # 需要分析的prometheus地址 local_rule_file: ./rules/*.yaml # 需要分析的 prometheus rule 文件，包括rule/record, 支持通配符 mimirtool_dir: ./bin # mimirtool 二进制的目录，如果检测到没有，会去github下载二进制到此目录下 optimization_relabel_type: metric_relabel_configs # 最后需要生成的优化配置段 metric_relabel_configs/write_relabel_configs完成后执行 ls -l 即可在当前目录下看到生成的优化结果目录
root@DESKTOP-269NF9M:/mnt/d/project/prom-metric-analyze/cmd# ls -l total 0 ... -rwxrwxrwx 1 root1 root1 311 Mar 15 21:02 analyze.yaml drwxrwxrwx 1 root1 root1 4096 Mar 15 21:22 metrics_analyze_result # 优化结果目录 drwxrwxrwx 1 root1 root1 4096 Mar 14 23:39 rules # 进入 metrics_analyze_result 可看到当前根据 metric 前缀分别生成了相对应的 xxx_relabel_configs 配置 root@DESKTOP-269NF9M:/mnt/d/project/prom-metric-analyze/cmd/metrics_analyze_result# ll total 16 -rwxrwxrwx 1 root1 root1 151 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_aggregator -rwxrwxrwx 1 root1 root1 450 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_alertmanager -rwxrwxrwx 1 root1 root1 247 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_apiserver -rwxrwxrwx 1 root1 root1 389 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_cluster -rwxrwxrwx 1 root1 root1 122 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_code -rwxrwxrwx 1 root1 root1 633 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_container -rwxrwxrwx 1 root1 root1 302 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_coredns -rwxrwxrwx 1 root1 root1 91 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_go -rwxrwxrwx 1 root1 root1 583 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_instance -rwxrwxrwx 1 root1 root1 1369 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_kube -rwxrwxrwx 1 root1 root1 659 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_kubelet -rwxrwxrwx 1 root1 root1 285 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_kubeproxy -rwxrwxrwx 1 root1 root1 99 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_kubernetes -rwxrwxrwx 1 root1 root1 345 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_namespace -rwxrwxrwx 1 root1 root1 1518 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_node -rwxrwxrwx 1 root1 root1 123 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_others -rwxrwxrwx 1 root1 root1 160 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_process -rwxrwxrwx 1 root1 root1 2482 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_prometheus -rwxrwxrwx 1 root1 root1 148 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_rest -rwxrwxrwx 1 root1 root1 186 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_scheduler -rwxrwxrwx 1 root1 root1 191 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_storage -rwxrwxrwx 1 root1 root1 106 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_volume -rwxrwxrwx 1 root1 root1 154 Mar 15 21:22 suggest_for_metric_relabel_configs_with_prefix_workqueue随便打开一个文件看效果
# root@DESKTOP-269NF9M:cmd/metrics_analyze_result# cat suggest_for_metric_relabel_configs_with_prefix_kubelet metric_relabel_configs: - action: keep source_labels: [ __name__ ] regex: kubelet_runtime_operations_duration_seconds_bucket|kubelet_pod_worker_duration_seconds_bucket|kubelet_pleg_relist_duration_seconds_bucket|kubelet_cgroup_manager_duration_seconds_bucket|kubelet_runtime_operations_total|kubelet_pod_start_duration_seconds_bucket|kubelet_pleg_relist_interval_seconds_bucket|kubelet_running_containers|kubelet_runtime_operations_errors_total|kubelet_pod_worker_duration_seconds_count|kubelet_cgroup_manager_duration_seconds_count|kubelet_node_name|kubelet_pod_start_duration_seconds_count|kubelet_running_pods|kubelet_pleg_relist_duration_seconds_count可以看到，已经生成了针对 kubelet 前缀的 metric_relabel_configs keep规则，可以直接配置到 prometheus 指定配置段，从而实现精简并优化metric的效果