Dan, We use an externalized config with the PredictionIO buildpack for Heroku.
If you create a `spark-defaults.conf` file to hold the properties, it will be picked-up by Spark executors, like those passed via `pio train -- --xxxxx`. The file should be inside the PredictionIO distribution at: vendors/spark-hadoop/conf/spark-defaults.conf The config entries should look like: > spark.es.nodes=10.0.0.1,10.0.0.2 AFAIK, the ES config values should also be set for PredictionIO itself, in `pio-env.sh` like: > PIO_STORAGE_SOURCES_ELASTICSEARCH_HOSTS=10.0.0.1,10.0.0.2 *Mars ( <> .. <> ) > On Jun 7, 2017, at 08:42, Dan Guja <[email protected]> wrote: > > Hi > > Is there a way to externalize the es.nodes property during train phase ? > > I've tried with this: > pio train -- ---conf es.nodes=<es.address> > > But apparently everything that does not start with spark prefix is ignored. > > I also tried this: > pio train -- ---conf spark.es.nodes=<es.address> > > No luck either > > > Thank you >
