When a query is running in the Presto machine [1], it uses almost all of the available memory. When this happens, Parquet2Hive doesn't have enough, and it fails with: > Failure to parse dataset, 'NoneType' object has no attribute 'group' What we probably want to do is to limit the memory used by Presto. [1]: hadoop@ec2-54-218-5-112.us-west-2.compute.amazonaws.com

The current Presto configuration uses a large chunk of the available memory on that instance [1]. The celery queue of re:dash can also consume a large amount of memory and I am not sure if that's bounded somehow. When that happens nearly no memory is left for other processes. We should deploy the re:dash service on its own independent instance; that would also make it easier to redeploy Presto with a new configuration. [1] https://github.com/vitillo/emr-bootstrap-presto/blob/master/ansible/files/telemetry.sh#L54

reassigning to Travis in light of whd's availability.

This will be taken care of by https://bugzilla.mozilla.org/show_bug.cgi?id=1279147