diff --git a/README.md b/README.md index 2b798126..50687b0b 100644 --- a/README.md +++ b/README.md @@ -15,20 +15,20 @@ Containers: * Prometheus (metrics database) `http://:9090` * AlertManager (alerts management) `http://:9093` -* Granafa (visualize metrics) `http://:3000` +* Grafana (visualize metrics) `http://:3000` * NodeExporter (host metrics collector) * cAdvisor (containers metrics collector) -While Granafa supports authentication, the Prometheus and AlertManager services have no such feature. +While Grafana supports authentication, the Prometheus and AlertManager services have no such feature. You can remove the ports mapping from the docker-compose file and use NGINX as a reverse proxy providing basic authentication for Prometheus and AlertManager. ## Setup Grafana -Navigate to `http://:3000` and login with user ***admin*** password ***changeme***. You can change the password from Granafa UI or +Navigate to `http://:3000` and login with user ***admin*** password ***changeme***. You can change the password from Grafana UI or by modifying the [user.config](https://github.com/stefanprodan/dockprom/blob/master/user.config) file. -From the Granafa menu, choose ***Data Sources*** and click on ***Add Data Source***. -Use the following values to add the Prometheues container as data source: +From the Grafana menu, choose ***Data Sources*** and click on ***Add Data Source***. +Use the following values to add the Prometheus container as data source: * Name: Prometheus * Type: Prometheus @@ -36,7 +36,7 @@ Use the following values to add the Prometheues container as data source: * Access: proxy Now you can import the dashboard temples from the [grafana](https://github.com/stefanprodan/dockprom/tree/master/grafana) directory. -From the Granafa menu, choose ***Dashboards*** and click on ***Import***. +From the Grafana menu, choose ***Dashboards*** and click on ***Import***. ***Docker Host Dashboard*** @@ -44,7 +44,7 @@ From the Granafa menu, choose ***Dashboards*** and click on ***Import***. The Docker Host Dashboard shows key metrics for monitoring the resource usage of your server: -* Server uptime, CPU idle percent, numner of CPU cores, available memory, swap and storage +* Server uptime, CPU idle percent, number of CPU cores, available memory, swap and storage * System load average graph, running and blocked by IO processes graph, interrupts graph * CPU usage graph by mode (guest, idle, iowait, irq, nice, softirq, steal, system, user) * Memory usage graph by distribution (used, free, buffers, cached) @@ -204,7 +204,7 @@ ALERT jenkins_high_memory The AlertManager service is responsible for handling alerts sent by Prometheus server. AlertManager can send notifications via email, Pushover, Slack, HipChat or any other system that exposes a webhook interface. -A compleat list of integrations can be found [here](https://prometheus.io/docs/alerting/configuration). +A complete list of integrations can be found [here](https://prometheus.io/docs/alerting/configuration). You can view and silence notifications by accessing `http://:9093`.