Target Postgres

A Singer postgres target, for use with Singer streams generated by Singer taps.

Features

• Creates SQL tables for Singer streams
• Denests objects flattening them into the parent object's table
• Denests rows into separate tables
• Adds columns and sub-tables as new fields are added to the stream JSON Schema
• Full stream replication via record version and ACTIVATE_VERSION messages.

Install

1. Add libpq dependency

# macos
brew install postgresql

# ubuntu
sudo apt install libpq-dev

1. install singer-target-postgres

pip install singer-target-postgres

Usage

1. Follow the Singer.io Best Practices for setting up separate tap and target virtualenvs to avoid version conflicts.

2. Create a config file at ~/singer.io/target_postgres_config.json with postgres connection information and target postgres schema.

   {
     "postgres_host": "localhost",
     "postgres_port": 5432,
     "postgres_database": "my_analytics",
     "postgres_username": "myuser",
     "postgres_password": "1234",
     "postgres_schema": "mytapname"
   }

3. Run target-postgres against a Singer tap.

   ~/.virtualenvs/tap-something/bin/tap-something \
     | ~/.virtualenvs/target-postgres/bin/target-postgres \
       --config ~/singer.io/target_postgres_config.json >> state.json

   If you are running windows, the following is equivalent:

   venvs\tap-exchangeratesapi\Scripts\tap-exchangeratesapi.exe | ^
   venvs\target-postgresql\Scripts\target-postgres.exe ^
   --config target_postgres_config.json
   

Config.json

The fields available to be specified in the config file are specified here.

Field	Type	Default	Details
postgres_host	["string", "null"]	"localhost"
postgres_port	["integer", "null"]	5432
postgres_database	["string"]	N/A
postgres_username	["string", "null"]	N/A
postgres_password	["string", "null"]	null
postgres_schema	["string", "null"]	"public"
postgres_sslmode	["string", "null"]	"prefer"	Refer to the libpq docs for more information about SSL
postgres_sslcert	["string", "null"]	"~/.postgresql/postgresql.crt"	Only used if a SSL request w/ a client certificate is being made
postgres_sslkey	["string", "null"]	"~/.postgresql/postgresql.key"	Only used if a SSL request w/ a client certificate is being made
postgres_sslrootcert	["string", "null"]	"~/.postgresql/root.crt"	Used for authentication of a server SSL certificate
postgres_sslcrl	["string", "null"]	"~/.postgresql/root.crl"	Used for authentication of a server SSL certificate
invalid_records_detect	["boolean", "null"]	true	Include false in your config to disable target-postgres from crashing on invalid records
invalid_records_threshold	["integer", "null"]	0	Include a positive value n in your config to allow for target-postgres to encounter at most n invalid records per stream before giving up.
disable_collection	["string", "null"]	false	Include true in your config to disable Singer Usage Logging.
logging_level	["string", "null"]	"INFO"	The level for logging. Set to DEBUG to get things like queries executed, timing of those queries, etc. See Python's Logger Levels for information about valid values.
persist_empty_tables	["boolean", "null"]	False	Whether the Target should create tables which have no records present in Remote.
max_batch_rows	["integer", "null"]	200000	The maximum number of rows to buffer in memory before writing to the destination table in Postgres
max_buffer_size	["integer", "null"]	104857600 (100MB in bytes)	The maximum number of bytes to buffer in memory before writing to the destination table in Postgres
batch_detection_threshold	["integer", "null"]	5000, or 1/40th max_batch_rows	How often, in rows received, to count the buffered rows and bytes to check if a flush is necessary. There's a slight performance penalty to checking the buffered records count or bytesize, so this controls how often this is polled in order to mitigate the penalty. This value is usually not necessary to set as the default is dynamically adjusted to check reasonably often.
state_support	["boolean", "null"]	True	Whether the Target should emit STATE messages to stdout for further consumption. In this mode, which is on by default, STATE messages are buffered in memory until all the records that occurred before them are flushed according to the batch flushing schedule the target is configured with.
add_upsert_indexes	["boolean", "null"]	True	Whether the Target should create column indexes on the important columns used during data loading. These indexes will make data loading slightly slower but the deduplication phase much faster. Defaults to on for better baseline performance.
before_run_sql	["string", "null"]	None	Raw SQL statement(s) to execute as soon as the connection to Postgres is opened by the target. Useful for setup like SET ROLE or other connection state that is important.
after_run_sql	["string", "null"]	None	Raw SQL statement(s) to execute as soon as the connection to Postgres is opened by the target. Useful for setup like SET ROLE or other connection state that is important.

Supported Versions

target-postgres only supports JSON Schema Draft4. While declaring a schema is optional, any input schema which declares a version other than 4 will be rejected.

target-postgres supports all versions of PostgreSQL which are presently supported by the PostgreSQL Global Development Group. Our CI config defines all versions we are currently supporting.

Version	Current minor	Supported	First Release	Final Release
12	12.2	Yes	October 3, 2019	November 14, 2024
11	11.7	Yes	October 18, 2018	November 9, 2023
10	10.12	Yes	October 5, 2017	November 10, 2022
9.6	9.6.17	Yes	September 29, 2016	November 11, 2021
9.5	9.5.21	Yes	January 7, 2016	February 11, 2021
9.4	9.4.26	Yes	December 18, 2014	February 13, 2020
9.3	9.3.25	No	September 9, 2013	November 8, 2018

The above is copied from the current list of versions on Postgresql.org

Known Limitations

• Requires a JSON Schema for every stream.
• Only string, string with date-time format, integer, number, boolean, object, and array types with or without null are supported. Arrays can have any of the other types listed, including objects as types within items.
  • Example of JSON Schema types that work
    • ['number']
    • ['string']
    • ['string', 'null']
  • Exmaple of JSON Schema types that DO NOT work
    • ['string', 'integer']
    • ['integer', 'number']
    • ['any']
    • ['null']
• JSON Schema combinations such as anyOf and oneOf are not supported.
• JSON Schema $ref is partially supported:
  • NOTE: The following limitations are known to NOT fail gracefully
  • Presently you cannot have any circular or recursive $refs
  • $refs must be present within the schema:
    • URI's do not work
    • if the $ref is broken, the behaviour is considered unexpected
• Any values which are the string NULL will be streamed to PostgreSQL as the literal null
• Table names are restricted to:
  • 63 characters in length
  • can only be composed of _, lowercase letters, numbers, $
  • cannot start with $
  • ASCII characters
• Field/Column names are restricted to:
  • 63 characters in length
  • ASCII characters

Indexes

If the add_upsert_indexes config option is enabled, which it is by default, target-postgres adds indexes on the tables it creates for its own queries to be more performant. Specifically, target-postgres automatically adds indexes to the _sdc_sequence column and the _sdc_level_<n>_id columns which are used heavily when inserting and upserting.

target-postgres doesn't have any facilities for adding other indexes to the managed tables, so if there are more indexes required, they should be added by another downstream tool, or can just be added by an administrator when necessary. Note that these indexes incur performance overhead to maintain as data is inserted, These indexes can also prevent target-postgres from dropping columns in the future if the schema of the table changes, in which case an administrator should drop the index so target-postgres is able to drop the columns it needs to.

Note: Index adding is new as of version 0.2.1, and target-postgres does not retroactively create indexes for tables it created before that time. If you want to add indexes to older tables target-postgres is loading data into, they should be added manually.

Usage Logging

Singer.io requires official taps and targets to collect anonymous usage data. This data is only used in aggregate to report on individual tap/targets, as well as the Singer community at-large. IP addresses are recorded to detect unique tap/targets users but not shared with third-parties.

To disable anonymous data collection set disable_collection to true in the configuration JSON file.

Developing

target-postgres utilizes setup.py for package management, and PyTest for testing.

Documentation

See also:

• DECISIONS: A document containing high level explanations of various decisions and decision making paradigms. A good place to request more explanation/clarification on confusing things found herein.
• TableMetadata: A document detailing some of the metadata necessary for TargetPostgres to function correctly on the Remote

Docker

If you have Docker and Docker Compose installed, you can easily run the following to get a local env setup quickly.

$ docker-compose up -d --build
$ docker logs -tf target-postgres_target-postgres_1 # You container names might differ

As soon as you see INFO: Dev environment ready. you can shell into the container and start running test commands:

$ docker-compose exec target-postgres bash
(target-postgres) root@...:/code# pytest

The environment inside the docker container has a virtualenv set up and activated, with an --editable install of target-postgres inside it and your local code mounted as a Docker volume. If you make changes on your host and re-run pytest any changes should be reflected inside the container.

See the PyTest commands below!

DB

To run the tests, you will need a PostgreSQL server running.

NOTE: Testing assumes that you've exposed the traditional port 5432.

Make sure to set the following env vars for PyTest:

$ EXPORT POSTGRES_HOST='<your-host-name>' # Most likely 'localhost'
$ EXPORT POSTGRES_DB='<your-db-name>'     # We use 'target_postgres_test'
$ EXPORT POSTGRES_USER='<your-user-name'  # Probably just 'postgres', make sure this user has no auth

PyTest

To run tests, try:

$ python setup.py pytest

If you've bash shelled into the Docker Compose container (see above), you should be able to simply use:

$ pytest

Collaboration and Contributions

Join the conversation over at the Singer.io Slack and on the #target-postgres channel.

Try to adhere to the following for contributing:

• File New Issue -> Fork -> New Branch(If needed) -> Pull Request -> Approval -> Merge

Users can file an issue without submitting a pull request but be aware not all issues can or will be addressed.

Sponsorship

Target Postgres is sponsored by Data Mill (Data Mill Services, LLC) datamill.co.

Data Mill helps organizations utilize modern data infrastructure and data science to power analytics, products, and services.

---

Copyright Data Mill Services, LLC 2018