Make record oriented data to columnar format.

Columnar formatted data is efficient for analytics queries, lightweight and ease to integrate with Data WareHouse middleware's. Conversion from record oriented data to columnar is sometimes realized by BigData stack like Hadoop ecosystem, and there's no easy way to do it lightly and quickly.

columnify is an easy conversion tool for columnar that enables to run single binary written in Go. It also supports some kinds of data format like JSONL(NewLine delimited JSON), Avro.

$ GO111MODULE=on go get github.com/reproio/columnify/cmd/columnify

$ ./columnify -h
Usage of columnify: columnify [-flags] [input files]
  -output string
        path to output file; default: stdout
  -recordType string
        data type, [avro|csv|jsonl|ltsv|msgpack|tsv] (default "jsonl")
  -schemaFile string
        path to schema file
  -schemaType string
        schema type, [avro|bigquery]

$ cat examples/record/primitives.jsonl
{"boolean": false, "int": 1, "long": 1, "float": 1.1, "double": 1.1, "bytes": "foo", "string": "foo"}
{"boolean": true, "int": 2, "long": 2, "float": 2.2, "double": 2.2, "bytes": "bar", "string": "bar"}

$ ./columnify -schemaType avro -schemaFile examples/primitives.avsc -recordType jsonl examples/primitives.jsonl > out.parquet

$ parquet-tools schema out.parquet
message Primitives {
  required boolean boolean;
  required int32 int;
  required int64 long;
  required float float;
  required double double;
  required binary bytes;
  required binary string (UTF8);
}

$ parquet-tools cat -json out.parquet
{"boolean":false,"int":1,"long":1,"float":1.1,"double":1.1,"bytes":"Zm9v","string":"foo"}
{"boolean":true,"int":2,"long":2,"float":2.2,"double":2.2,"bytes":"YmFy","string":"bar"}

• Apache Avro
• CSV
• JSONL(NewLine delimited JSON)
• LTSV
• Message Pack
• TSV

• Apache Parquet

• Apache Avro
• BigQuery Schema

• fluent-plugin-s3 parquet compressor

  • An example is examples/fluent-plugin-s3
  • It works as a Compressor of fluent-plugin-s3 write parquet file to tmp via chunk data.

columnify might consume lots of memory depending on a value specified by -parquetRowGroupSize. At least, it needs a memory of the row group size. Actually, it consumes more than double the row group size by default. The reason for that depends on Go's garbage collection behavior, and memory usage can decrease by triggering GC frequently. To adjust the frequency, set GOGC environment variable.

SetGCPercent sets the garbage collection target percentage: a collection is triggered when the ratio of freshly allocated data to live data remaining after the previous collection reaches this percentage. SetGCPercent returns the previous setting. The initial setting is the value of the GOGC environment variable at startup, or 100 if the variable is not set. A negative percentage disables garbage collection.

https://golang.org/pkg/runtime/debug/#SetGCPercent

Of course, frequent GC makes it increase execution time. Confirm which GOGC value (percent) is better in your environment.

Currently it has some limitations from schema/record types.

• Some logical types like Decimal are unsupported.
• If using -recordType = avro, it doesn't support a nested record has only 1 sub field.
• If using -recordType = avro, it converts bytes fields to base64 encoded value implicitly.

Columnifier reads input file(s), converts format based on given parameter, finally writes output files. Format conversion is separated by schema / record. The schema conversion accepts input schema, then converts it to targer's via Arrow's schema. The record conversion is similar to schema's but intermediate is simply map[string]interface{}, because Arrow record isn't available as an intermediate. columnify basically depends on existing modules but it contains additional modules like avro, parquet to fill insufficient features.

goreleaser is integrated in GitHub Actions. It's triggerd on creating a new tag. Create a new release with semvar tag(vx.y.z) on this GitHub repo, then you get archives for some environments attached on the release.