StackLight LMA high availability

StackLight LMA high availability

High availability in StackLight LMA is achieved through the deployment of three Prometheus servers, Prometheus Relay service, and InfluxDB Relay service.

Warning

InfluxDB, including InfluxDB Relay and remote storage adapter, is deprecated in the Q4`18 MCP release and will be removed in the next release.

To ensure high availability for Prometheus, StackLight LMA deploys three Prometheus servers at the same time. Each Prometheus server uses the same configuration file, monitors the same endpoints, and has the same alerts defined. The Alertmanager service deduplicates the fired alerts, so you will receive one alert instead of three.

For external components such as Grafana, the Prometheus Relay service handles Prometheus API calls, sends them to all discovered Prometheus servers, merges the results, and returns them to Grafana to visualize the data from all Prometheus servers. Therefore, if one Prometheus servers is down, Grafana will contain the data from the remaining Prometheus servers.

The following diagram illustrates the Prometheus HA.

../_images/prometheus-ha.png

High availability for Prometheus long-term storage is achieved by scraping the same data in an independent way. In case one Prometheus server fails, the other two will contain the data. To keep the time series gapless, Prometheus Relay acts as a proxy and merges the results from three underlay Prometheus servers.

High availability for the InfluxDB service is achieved using the InfluxDB Relay service that listens to HTTP writes and sends the data to each InfluxDB server through the HTTP write endpoint. InfluxDB Relay returns a success response once one of the InfluxDB servers returns it. If any InfluxDB server returns a 4xx response or if all servers return a 5xx response, it will be returned to the client. If some but not all servers return a 5xx response, it will not be returned to the client.

This approach allows sustaining failures of one InfluxDB or one InfluxDB Relay service while these services will still perform writes and serve queries. InfluxDB Relay buffers failed requests in memory to reduce the number of failures during short outages or periodic network issues.