[mej10]

Do you know of any good ways of outputting log data directly into Parquet format without hadoop/spark?

[mason55]

I'm curious about this as well. I was evaluating Athena a few months ago and was surprised that there was no good way to get my data into Parquet or ORC format without spinning up an EMR cluster and loading into a Hive table of the desired format.

My guess is that in the past there was no real use for these formats unless you already had a Hadoop cluster running. If Amazon wants these "Hadoop as a service" concepts to take off it seems like it would be wise for them to make it easier to get data onto S3 in a better format than CSV.

[arnon]

Mostly what I'm hearing here is:

"You're using it wrong"

"You should start up a Hadoop cluster, create a table, load the CSV into that table, then export it as Parquet, and then load that into S3, so that Athena can scan it"

Wouldn't you be honestly just better off creating a table and loading the CSV into a columnar database directly, like on Amazon RDS (Even if it means you bring the server instance down after you're done)?

[vamin]

You might look into Secor: https://github.com/pinterest/secor

[gallamine]

I was able to use Python + Arrow to convert single files into Parquet.

[mej10]

I will check Arrow out, thanks!

[rsynnott]

You can use org.apache.parquet.hadoop.ParquetWriter and its superclasses. It doesn't require a Hadoop cluster. There may also be other ways.

[marklit]

If you mean not as a cluster Spark standalone can do conversions.