There are many options for collecting application metrics, but you will need to use a MetricsConsumer if you are interested in getting at the storm internal metrics, or metrics of spouts or bolts that use the storm metrics system (e.g. KafkaSpout). I use both a storm metrics and another system (twitter ostrich based) - I tend to prefer using ostrich if possible, but it's not because of reliability. Ultimately the metrics from both end up in the same place.
If any component crashes (including a MetricsConsumer), it should be bringing down the worker on which it is running, not the whole topology. On Thu, Jul 23, 2015 at 9:08 AM Hong Jeon <[email protected]> wrote: > If the whole topology crashes when MetricsConsumer crashes, does this not > make using MetricsConsumer to log metrics a liability? Are there safer > alternatives? What do people think of this? >
