PGLoader

pgloader is a data loading tool for PostgreSQL, using the COPY command.

Its main avantage over just using COPY or \copy and over using a Foreign Data Wrapper is the transaction behaviour, where pgloader will keep a separate file of rejected data and continue trying to copy good data in your database.

The default PostgreSQL behaviour is transactional, which means that any erroneous line in the input data (file or remote database) will stop the bulk load for the whole table.

pgloader also implements data reformating, the main example of that being a transformation from MySQL dates 0000-00-00 and 0000-00-00 00:00:00 to PostgreSQL NULL value (because our calendar never had a year zero).

INSTALL

pgloader is now a Common Lisp program, tested using the SBCL and CCL implementation with Quicklisp.

apt-get install sbcl
wget http://beta.quicklisp.org/quicklisp.lisp
sbcl --load quicklisp.lisp
* (quicklisp-quickstart:install)
* (ql:add-to-init-file)

Now fetch pgloader sources into ~/quicklisp/local-projects/ so that you can do:

sbcl
* (ql:quickload :pgloader)
* (in-package :pgloader)
* (stream-database-tables "weetix")

Usage

TODO

Some notes about what I intend to be working on next.

internals & refactoring

• review pgloader.pgsql:reformat-row date-columns arguments
• review connection string handling for both PostgreSQL and MySQL
• provide a better toplevel API
• implement tests

command & control

• commands: LOAD and INI formats
• compat with SQL*Loader format

Here's a quick spec of the LOAD grammar:

LOAD FROM '/path/to/filename.txt'
          stdin
		  http://url.to/some/file.txt
		  mysql://[user[:pass]@][host[:port]]/dbname
	 [ COMPRESSED WITH zip | bzip2 | gzip ]

WITH workers = 2,
	 batch size = 25000,
	 batch split = 5,
     reject file = '/tmp/pgloader/<table-name>.dat'
	 log file = '/tmp/pgloader/pgloader.log',
	 log level = debug | info | notice | warning | error | critical,
	 truncate,
     fields [ optionally ] enclosed by '"',
     fields escaped by '"',
     fields terminated by '\t',
     lines terminated by '\r\n',
	 encoding = 'latin9',
	 drop table,
	 create table,
	 create indexes,
	 reset sequences;
	 
 SET guc-1 = 'value', guc-2 = 'value';
 
 PREPARE CLIENT WITH ( <lisp> )
 PREPARE SERVER WITH ( <sql> )
 
INTO postgresql://[user[:pass]@][host[:port]]/dbname?table-name
     [ WITH <options> SET <gucs> ]
     (
	   field-name data-type field-desc [ with column options ],
	   ...
	 )
USING (expression field-name other-field-name) as column-name,
      ...

INTO table-name  [ WITH <options> SET <gucs> ]
	 (
	   *
	 )

WHEN 

 FINALLY ON CLIENT DO ( <lisp> )
         ON SERVER DO ( <lisp> )

< data here if loading from stdin >

The accepted column options are:

terminated by ':'
nullif { blank | zero date }
date format "DD-Month-YYYY"

And we need a database migration command syntax too:

LOAD DATABASE FROM mysql://localhost:3306/dbname
              INTO postgresql://localhost/db
WITH drop tables,
	 create tables,
	 create indexes,
	 reset sequences,
     <options>
 SET guc = 'value', ...
CAST tablename.column to timestamptz drop default,
	 varchar to text,
	 int with extra auto_increment to bigserial,
	 datetime to timestamptz drop default,
	 date to date drop default

error management

• error management with a local buffer (done)
• error reporting (done)
• add input line number to log file?

data input

• import directly from MySQL, file based export/import (done)
• import directly from MySQL streaming (done)
• general CSV and Flexible Text source formats
• fixed cols input data format
• compressed input (gzip, other algos)
• fetch data from S3

data output

• PostgreSQL COPY Text format output for any supported input

convenience

• automatic creation of schema (from MySQL schema, or from CSV header)
• pre-fetch some rows to guesstimate data types?

performances

• some more parallelizing options
• support for partitionning in pgloader itself

reformating

Data reformating is now going to have to happen in Common Lisp mostly, maybe offer some other languages (cl-awk etc).

• raw reformating, before rows are split
• per column reformating
  • date (zero dates)
  • integer and "" that should be NULL
• user-defined columns (constants, functions of other rows)
• column re-ordering

Have a try at something approaching:

WITH data AS (
	COPY FROM ...
	RETURNING x, y
)
SELECT foo(x), bar(y)
  FROM data
 WHERE ...

A part of that needs to happen client-side, another part server-side, and the grammar has to make it clear what happens where. Maybe add a WHERE clause to the COPY or LOAD grammar for the client.

UI

• add a web controler with pretty monitoring
• launch new jobs from the web controler

crazy ideas

• MySQL replication, reading from the binlog directly
• plproxy (re-)sharding support
• partitioning support
• remote archiving support (with (delete returning *) insert into)