For custom metrics, you can take a look at Groupon’s spar metrics: https://github.com/groupon/spark-metrics
It is supported on spark 2.x. Alternatively, you can create a custom source (extending source trait), enable the sink and register the custom source to get the metrics. Some useful links for this approach: https://gist.github.com/ibuenros/9b94736c2bad2f4b8e23 https://kb.databricks.com/metrics/spark-metrics.html http://mail-archives.us.apache.org/mod_mbox/spark-user/201501.mbox/%3CCAE50=dq+6tdx9VNVM3ctBMWPLDPbUAacO3aN3L8x38zg=xb...@mail.gmail.com%3E I hope these help. -- Thanks & Regards, Akshay Haryani From: Aurélien Mazoyer <aurel...@aepsilon.com> Date: Monday, September 6, 2021 at 5:47 AM To: Haryani, Akshay <akshay.hary...@hpe.com> Cc: user@spark.apache.org <user@spark.apache.org> Subject: Re: Get application metric from Spark job Hi Akshay, Thank you for your reply. Sounds like a good idea, but I unfortunately have a 2.6 cluster. Do you know if there would be another solution that would run on 2.6 or if I have no other choice than migrating to 3? Regards, Aurélien Le jeu. 2 sept. 2021 à 20:12, Haryani, Akshay <akshay.hary...@hpe.com<mailto:akshay.hary...@hpe.com>> a écrit : Hi Aurélien, Spark has endpoints to expose the spark application metrics. These endpoints can be used as a rest API. You can read more about these here: https://spark.apache.org/docs/3.1.1/monitoring.html#rest-api<https://spark.apache.org/docs/3.1.1/monitoring.html#rest-api> Additionally, If you want to build your own custom metrics, you can explore spark custom plugins. Using a custom plugin, you can track your own custom metrics and plug it into the spark metrics system. Please note plugins are supported on spark versions above 3.0. -- Thanks & Regards, Akshay Haryani From: Aurélien Mazoyer <aurel...@aepsilon.com<mailto:aurel...@aepsilon.com>> Date: Thursday, September 2, 2021 at 8:36 AM To: user@spark.apache.org<mailto:user@spark.apache.org> <user@spark.apache.org<mailto:user@spark.apache.org>> Subject: Get application metric from Spark job Hi community, I would like to collect information about the execution of a Spark job while it is running. Could I define some kind of application metrics (such as a counter that would be incremented in my code) that I could retrieve regularly while the job is running? Thank you for help, Aurelien
