Outflux - Migrate InfluxDB to TimescaleDB

This repo contains code for exporting complete InfluxDB databases or selected measurements to TimescaleDB.

Table of Contents

1. Installation

• Installing from source
• Binary releases

2. How to use

• Before using it
• Connection params
• Schema Transfer
• Migrate
• Examples

3. Connection

• TimescaleDB connection params
• InfluxDB connection params

4. Known limitations

Installation

Binary releases

We provide binaries for GNU/Linux, Windows and MacOS with each release, these can be found under releases. To use outflux, download the binary, extract the compressed tarball and run the executable.

wget https://github.com/timescale/outflux/releases/download/v0.3.0/outflux_0.3.0_Linux_x86_64.tar.gz
tar xf outflux_0.3.0_Linux_x86_64.tar.gz
./outflux --help

Installing from source

Outflux is a Go project managed by go modules. You can download it in any directory and on the first build it will download it's required dependencies.

Depending on where you downloaded it and the go version you're using, you may need to set the GO111MODULE to auto, on or off. Learn about the GO111MODULE environment variable here.

# Fetch the source code of Outflux in any directory
$ git clone [email protected]:timescale/outflux.git
$ cd ./outflux

# Install the Outflux binary (will automaticly detect and download)
# dependencies.
$ cd cmd/outflux
$ GO111MODULE=auto go install

# Building without installing will also fetch the required dependencies
$ GO111MODULE=auto go build ./... 

How to use

Outflux supports InfluxDB 1.x.

Outflux should support using the 1.x query APIs for InfluxDB 2.x and 3.x. You will need to enable the 1.x APIs to use them. Consult the InfluxDB documentation for more details.

Before using it

It is recommended that you have and InfluxDB database with some data. For testing purposes you can check out the TSBS Data Loader Tool, which is part of the Time Series Benchmark Suite. It can generate large amounts of data to load into influx. Data can be generated with one command, just specify the format as 'influx', and then load it in with another command.

Connection params

Detailed information about how to pass the connection parameters to Outflux can be found at the bottom of this document at the Connection section.

Schema Transfer

The Outflux CLI has two commands. The first one is schema-transfer. This command discoverx the schema of an InfluxDB database, or specific measurements in an InfluxDB database, and (depending on the strategy selected) create or verify a TimescaleDB database that could hold the data.

The possible flags for the command can be seen by running:

$ cd $GOPATH/bin/
$ ./outflux schema-transfer --help

Usage is outflux schema-transfer database [measure1 measure2 ...] [flags], where database is the name of the InfluxDB database you wish to export, [measure1 ...] are optional and if specified will export only those measurements from the selected database. Additionally, you can specify the retention policy with the retention-policy flag.

For example outflux schema-transfer benchmark cpu mem will discover the schema for the cpu and mem measurements from the benchmark database.

Available flags for schema-transfer are:

Migrate

The second command of the Outflux CLI is migrate. The possible flags for the command can be seen by running:

$ cd $GOPATH/bin/
$ ./outflux migrate --help

Usage is outflux migrate database [measure1 measure2 ...] [flags], where database is the name of the InfluxDB database you wish to export, [measure1 measure2 ...] are optional and if specified will export only those measurements from the selected database.

The retention policy can be specified with the retention-policy flag. By default, the 'autogen' retention policy is used.

For example outflux migrate benchmark cpu mem will export the cpu and mem measurements from the benchmark database. On the other hand outflux migrate benchmark will export all measurements in the benchmark database.

Available flags are:

Examples

• Use environment variables for determining output db connection

$ PGPORT=5433
$ PGDATABASE=test
$ PGUSER=test
...
$ ./outflux schema-transfer benchmark

• Export the complete 'benchmark' database on 'localhost:8086' to the 'targetdb' database on localhost:5432. Use environment variable to set InfluxDB password

$ PGDATABASE=some_default_db
$ INFLUX_PASSWORD=test
...
$ outflux migrate benchmark \
> --input-user=test \
> --input-pass=test \
> --output-conn='dbname=targetdb user=test password=test' \

• Export only measurement 'cpu' from 'two_week' retention policy in the 'benchmark' database. Drop the existing '"two_week.cpu"' table in 'targetdb' if exists, create if not

$ outflux migrate benchmark two_week.cpu \
> --input-user=test \
> --input-pass=test \
> --output-conn='dbname=targetdb user=test pass=test'\
> --schema-strategy=DropAndCreate

• Export only the 1,000,000 rows from measurements 'cpu' and 'mem' from 'benchmark', starting from Jan 14th 2019 09:00

$ ./outflux migrate benchmark cpu mem \
> --input-user=test \
> --input-pass=test \
> --limit=1000000 \
> --from=2019-01-01T09:00:00Z

Connection

TimescaleDB connection params

The connection parameters to the TimescaleDB instance can be passed to Outflux in several ways. One is through the Postgres Environment Variables. Supported environment variables are: PGHOST, PGPORT, PGDATABASE, PGUSER, PGPASSWORD, PGSSLMODE, PGSSLCERT, PGSSLKEY, PGSSLROOTCERT, PGAPPNAME, PGCONNECT_TIMEOUT. If they are not specified defaults used are: host=localhost, dbname=postgres, pguser=$USER, and sslmode=disable.

The values of the enviroment variables can be OVERRIDEN by specifying the '--output-conn' flag when executing Outflux.

The connection string can be in the format URI or DSN format:

• example URI: "postgresql://username:password@host:port/dbname?connect_timeout=10"
• example DSN: "user=username password=password host=1.2.3.4 port=5432 dbname=mydb sslmode=disable"

InfluxDB connection params

The connection parameters to the InfluxDB instance can be passed also through flags or environment variables. Supported/Expected environment variables are: INFLUX_USERNAME, INFLUX_PASSWORD. These are the same environment variables that the InfluxDB CLI uses.

If they are not set, or if you wish to override them, you can do so with the --input-user and --input-pass. Also you can specify to Outflux to skip HTTPS verification when communicating with the InfluxDB server by setting the --input-unsafe-https flag to true.

Known limitations

Fields with different data types across shards

Outflux doesn't support fields that have the same name but different data types across shards in InfluxDB, UNLESS the field is an integer and float in the InfluxDB shards. InfluxDB can store the fields as integer (64bit integer), float (64bit float), string, and boolean. You can specify the multishard-int-float-cast flag. This will tell Outflux to cast the integer values to float values. A 64bit float can't hold all the int64 values, so this may result in scrambled data (for values > 2^53).

If the same field is of any of the other possible InfluxDB types, an error will be thrown, since the values can't be converted.

This is also an issue even if you select a time interval in which a field has a consistent type, but exists as a different type in a shard outside of that interval. This is because the SHOW FIELD KEYS FROM measurement_name doesn't accept a time interval for which you would be asking