jchampio/timescaledb-parallel-copy

timescaledb-parallel-copy

timescaledb-parallel-copy is a command line program for parallelizing PostgreSQL's built-in COPY functionality for bulk inserting data into TimescaleDB.

Getting started

You need the Go runtime (1.13+) installed, then simply go get this repo:

$ go install github.com/timescale/timescaledb-parallel-copy/cmd/timescaledb-parallel-copy@latest

Before using this program to bulk insert data, your database should be installed with the TimescaleDB extension and the target table should already be made a hypertable.

Using timescaledb-parallel-copy

If you want to bulk insert data from a file named foo.csv into a (hyper)table named sample in a database called test:

# single-threaded
$ timescaledb-parallel-copy --db-name test --table sample --file foo.csv

# 2 workers
$ timescaledb-parallel-copy --db-name test --table sample --file foo.csv \
    --workers 2

# 2 workers, report progress every 30s
$ timescaledb-parallel-copy --db-name test --table sample --file foo.csv \
    --workers 2 --reporting-period 30s

# Treat literal string 'NULL' as NULLs:
$ timescaledb-parallel-copy --db-name test --table sample --file foo.csv \
    --copy-options "NULL 'NULL' CSV"

Other options and flags are also available:

$ timescaledb-parallel-copy --help

Usage of timescaledb-parallel-copy:
  -batch-size int
        Number of rows per insert (default 5000)
  -columns string
        Comma-separated columns present in CSV
  -connection string
        PostgreSQL connection url (default "host=localhost user=postgres sslmode=disable")
  -copy-options string
        Additional options to pass to COPY (e.g., NULL 'NULL') (default "CSV")
  -db-name string
        Database where the destination table exists
  -file string
        File to read from rather than stdin
  -header-line-count int
        Number of header lines (default 1)
  -limit int
        Number of rows to insert overall; 0 means to insert all
  -log-batches
        Whether to time individual batches.
  -reporting-period duration
        Period to report insert stats; if 0s, intermediate results will not be reported
  -schema string
        Destination table's schema (default "public")
  -skip-header
        Skip the first line of the input
  -split string
        Character to split by (default ",")
  -table string
        Destination table for insertions (default "test_table")
  -truncate
        Truncate the destination table before insert
  -verbose
        Print more information about copying statistics
  -version
        Show the version of this tool
  -workers int
        Number of parallel requests to make (default 1)

Purpose

PostgreSQL native COPY function is transactional and single-threaded, and may not be suitable for ingesting large amounts of data. Assuming the file is at least loosely chronologically ordered with respect to the hypertable's time dimension, this tool should give you great performance gains by parallelizing this operation, allowing users to take full advantage of their hardware.

This tool also takes care to ingest data in a more efficient manner by roughly preserving the order of the rows. By taking a "round-robin" approach to sharing inserts between parallel workers, the database has to switch between chunks less often. This improves memory management and keeps operations on the disk as sequential as possible.

Contributing

We welcome contributions to this utility, which like TimescaleDB is released under the Apache2 Open Source License. The same Contributors Agreement applies; please sign the Contributor License Agreement (CLA) if you're a new contributor.

Running Tests

Some of the tests require a running Postgres database. Set the TEST_CONNINFO environment variable to point at the database you want to run tests against. (Assume that the tests may be destructive; in particular it is not advisable to point the tests at any production database.)

For example:

$ createdb gotest
$ TEST_CONNINFO='dbname=gotest user=myuser' go test -v ./...

jchampio / timescaledb-parallel-copy