ravi kanth <[email protected]> writes: > Wei-Chiu, > > Thanks for the response. Are there any workarounds for Hadoop 3.1.0? As > this is a critical production cluster it needs immediate attention.
My approach was a bit different. Instead of using Prometheus to pull the metrics, I used collectd [1] to collect system and JMX metrics, send that to Riemann [2] and store it in database (in my case OpenTSDB, but changing it to Prometheus pushgateway address would be couple of lines of code in Riemann). Inside Riemann, you can tag, shape and calculate metrics the way you like. There is also standalone app [3] you can run to watch for multiple JMX connections, but I haven't tried that. [1] https://collectd.org/ [2] http://riemann.io/ [3] https://github.com/twosigma/riemann-jmx > Thanks, > Ravi Best, Sanel > On Tue, Jun 2, 2020 at 5:59 PM Wei-Chiu Chuang <[email protected]> wrote: > >> Check out HADOOP-16398 >> <https://issues.apache.org/jira/browse/HADOOP-16398> >> It's a new feature in Hadoop 3.3.0 >> >> Akira might be able to help. >> >> On Tue, Jun 2, 2020 at 5:56 PM ravi kanth <[email protected]> wrote: >> >>> Hi Everyone, >>> >>> We have a production-ready cluster with 35 nodes that we are currently >>> using. We are currently using System metrics using Prometheus + Grafana to >>> visualize server. However, we are more interested in visualizing the Hadoop >>> & Yarn service level metrics. >>> >>> I came across hadoop JMX port which exposes all the needed metrics from >>> the service. However, I remained unsuccessful in tagging these metrics to >>> prometheus jmx agent. >>> >>> Is there anyone who successfully got the JMX monitoring of Hadoop >>> components work with Prometheus? Any help is greatly appreciated. >>> >>> Currently, we have scripts to parse the meaningful values out of JMX end >>> point of the namenode & datanodes. >>> >>> Thanks In advance, >>> >>> Ravi >>> >>> --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
