Hi,

On 8/4/20 12:24 PM, Vinod M V wrote:
> 
>           I am facing Memory usage with Prometheus service and
> Maintaining 30 days of data from Node exporter, Process exporter and JMX
> exporter for 95 servers in Prometheus Database. 
> 
>          Grafana and Prometheus are running on the same node. When
> loading node_exporter or JVM Grafana dashboards, memory usage will shoot
> up to more than 10 GB and sometime usage will hit max and Prometheus
> will be restarted. Moreover memory usage will not reduce even after
> closing JVM/Node exporter Grafana Dashboards. 
> 
>          Always need to restart Prometheus to release the memory. I am
> expecting 300 more servers to this configuration in future. 

95 servers don't sound that much. However, it really depends you many
metrics you are scraping from them. node_exporter should usually be now
problem. process_exporter's metric count may highly depend on your
configuration. jvm_exporter is probably what accounts for the largest
number of metrics. You can try to check if you can drop some of them
(e.g. by reworking the jmx_exporter config, or, if nothing helps, via
relabelling).

You may also want to look into your dashboards. Queries which load a lot
of metrics over longer timeframes might cause memory spikes. It can help
to set up recording rules which you can then use in your dashboard.

>         Can someone please suggest a solution for this memory issue ( or
> suggest a best database for save the metrics) and suggest a open source
> solution for High availability

Prometheus already contains a highly optimized TSDB. It can also be set
up as high availability setup. Just run two identically configured
Prometheus instances in order to ensure that at least one stays
available. You can then either point Grafana to just one of them
(especially if you want to avoid dashboard queries to overload both
instances) or you could load balance to both of your Prometheus instances.

If you are looking for alternatives, you can try Thanos, Cortex or
VictoriaMetrics.


Kind regards,
Christian

-- 
You received this message because you are subscribed to the Google Groups 
"Prometheus Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to prometheus-users+unsubscr...@googlegroups.com.
To view this discussion on the web visit 
https://groups.google.com/d/msgid/prometheus-users/abf47c43-12f7-975e-f711-4466ee2fb12b%40hoffmann-christian.info.

Reply via email to