Basculer Gitlab de MySQL à PostgreSQL

PostgreSQL est un moteur de base de données plein de qualités. Bien qu'il soit souvent un peu hasardeux et compliqué de basculer une base de données de MySQL à PostgreSQL en raison de petites différences de comportement entre les bases de données, Gitlab fournit une procédure bien rodée pour une telle migration.

En substance, il faut :

  • faire un export de la base de données avec l'outil d'export de Gitlab
  • modifier le fichier de données au sein de l'export
  • réimporter l'export

La procédure est détaillée ici : http://doc.gitlab.com/ce/update/mysql_to_postgresql.html

A toute fin utile, j'en conserve une copie (en date du 13 juillet 2015 ci-dessous) :

# Stop GitLab
sudo service gitlab stop

# Create the backup
cd /home/git/gitlab
sudo -u git -H bundle exec rake gitlab:backup:create RAILS_ENV=production

# Note the filename of the backup that was created. We will call it
# TIMESTAMP_gitlab_backup.tar below.

# Move the backup file we will convert to its own directory
sudo -u git -H mkdir -p tmp/backups/postgresql
sudo -u git -H mv tmp/backups/TIMESTAMP_gitlab_backup.tar tmp/backups/postgresql/

# Create a separate database dump with PostgreSQL compatibility
cd tmp/backups/postgresql
sudo -u git -H mysqldump --compatible=postgresql --default-character-set=utf8 -r gitlabhq_production.mysql -u root gitlabhq_production -p

# Clone the database converter
sudo -u git -H git clone https://github.com/gitlabhq/mysql-postgresql-converter.git -b gitlab

# Convert gitlabhq_production.mysql
sudo -u git -H mkdir db
sudo -u git -H python mysql-postgresql-converter/db_converter.py gitlabhq_production.mysql db/database.sql

# Compress database backup
sudo -u git -H gzip db/database.sql

# Replace the MySQL dump in TIMESTAMP_gitlab_backup.tar.

# Warning: if you forget to replace TIMESTAMP below, tar will create a new file
# 'TIMESTAMP_gitlab_backup.tar' without giving an error.

sudo -u git -H tar rf TIMESTAMP_gitlab_backup.tar db/database.sql.gz

# Done! TIMESTAMP_gitlab_backup.tar can now be restored into a Postgres GitLab
# installation. Remember to recreate the indexes after the import.

et le fichier db_converter.py contient :

#!/usr/bin/env python

"""
Fixes a MySQL dump made with the right format so it can be directly
imported to a new PostgreSQL database.

Dump using:
mysqldump --compatible=postgresql --default-character-set=utf8 -r databasename.mysql -u root databasename
"""

import re
import sys
import os
import time
import subprocess


def parse(input_filename, output_filename):
    "Feed it a file, and it'll output a fixed one"

    # State storage
    if input_filename == "-":
        num_lines = -1
    else:
        num_lines = int(subprocess.check_output(["wc", "-l", input_filename]).strip().split()[0])
    tables = {}
    current_table = None
    creation_lines = []
    enum_types = []
    foreign_key_lines = []
    index_lines = []
    sequence_lines = []
    cast_lines = []
    num_inserts = 0
    started = time.time()

    # Open output file and write header. Logging file handle will be stdout
    # unless we're writing output to stdout, in which case NO PROGRESS FOR YOU.
    if output_filename == "-":
        output = sys.stdout
        logging = open(os.devnull, "w")
    else:
        output = open(output_filename, "w")
        logging = sys.stdout

    if input_filename == "-":
        input_fh = sys.stdin
    else:
        input_fh = open(input_filename)


    output.write("-- Converted by db_converter\n")
    output.write("START TRANSACTION;\n")
    output.write("SET standard_conforming_strings=off;\n")
    output.write("SET escape_string_warning=off;\n")
    output.write("SET CONSTRAINTS ALL DEFERRED;\n\n")

    for i, line in enumerate(input_fh):
        time_taken = time.time() - started
        percentage_done = (i+1) / float(num_lines)
        secs_left = (time_taken / percentage_done) - time_taken
        logging.write("\rLine %i (of %s: %.2f%%) [%s tables] [%s inserts] [ETA: %i min %i sec]" % (
            i + 1,
            num_lines,
            ((i+1)/float(num_lines))*100,
            len(tables),
            num_inserts,
            secs_left // 60,
            secs_left % 60,
        ))
        logging.flush()
        line = line.decode("utf8").strip().replace(r"\", "WUBWUBREALSLASHWUB").replace(r"\0", "").replace(r"\'", "''").replace("WUBWUBREALSLASHWUB", r"\")
        # Ignore comment lines
        if line.startswith("--") or line.startswith("/*") or line.startswith("LOCK TABLES") or line.startswith("DROP TABLE") or line.startswith("UNLOCK TABLES") or not line:
            continue

        # Outside of anything handling
        if current_table is None:
            # Start of a table creation statement?
            if line.startswith("CREATE TABLE"):
                current_table = line.split('"')[1]
                tables[current_table] = {"columns": []}
                creation_lines = []
            # Inserting data into a table?
            elif line.startswith("INSERT INTO"):
                output.write(re.sub(r"([^'])'0000-00-00 00:00:00'", r"\1NULL", line.encode("utf8")) + "\n")
                num_inserts += 1
            # ???
            else:
                print "\n ! Unknown line in main body: %s" % line

        # Inside-create-statement handling
        else:
            # Is it a column?
            if line.startswith('"'):
                useless, name, definition = line.strip(",").split('"',2)
                try:
                    type, extra = definition.strip().split(" ", 1)

                    # This must be a tricky enum
                    if ')' in extra:
                        type, extra = definition.strip().split(")")

                except ValueError:
                    type = definition.strip()
                    extra = ""
                extra = re.sub("CHARACTER SET [\w\d]+\s*", "", extra.replace("unsigned", ""))
                extra = re.sub("COLLATE [\w\d]+\s*", "", extra.replace("unsigned", ""))

                # See if it needs type conversion
                final_type = None
                final_default = None
                set_sequence = None
                if type == "tinyint(1)":
                    type = "int4"
                    set_sequence = True
                    final_type = "boolean"

                    if "DEFAULT '0'" in extra:
                        final_default = "FALSE"
                    elif "DEFAULT '1'" in extra:
                        final_default = "TRUE"

                elif type.startswith("int("):
                    type = "integer"
                    set_sequence = True
                elif type.startswith("bigint("):
                    type = "bigint"
                    set_sequence = True
                elif type == "longtext":
                    type = "text"
                elif type == "mediumtext":
                    type = "text"
                elif type == "tinytext":
                    type = "text"
                elif type.startswith("varchar("):
                    size = int(type.split("(")[1].rstrip(")"))
                    type = "varchar(%s)" % (size * 2)
                elif type.startswith("smallint("):
                    type = "int2"
                    set_sequence = True
                elif type == "datetime":
                    type = "timestamp with time zone"
                    extra = extra.replace("NOT NULL", "")
                elif type == "double":
                    type = "double precision"
                elif type == "blob":
                    type = "bytea"
                elif type.startswith("enum(") or type.startswith("set("):

                    types_str = type.split("(")[1].rstrip(")").rstrip('"')
                    types_arr = [type_str.strip('\'') for type_str in types_str.split(",")]

                    # Considered using values to make a name, but its dodgy
                    # enum_name = '_'.join(types_arr)
                    enum_name = "{0}_{1}".format(current_table, name)

                    if enum_name not in enum_types:
                        output.write("CREATE TYPE {0} AS ENUM ({1}); \n".format(enum_name, types_str));
                        enum_types.append(enum_name)

                    type = enum_name

                if final_type:
                    cast_lines.append("ALTER TABLE "%s" ALTER COLUMN "%s" DROP DEFAULT" % (current_table, name))
                    cast_lines.append("ALTER TABLE "%s" ALTER COLUMN "%s" TYPE %s USING CAST("%s" as %s)" % (current_table, name, final_type, name, final_type))
                    if final_default:
                        cast_lines.append("ALTER TABLE "%s" ALTER COLUMN "%s" SET DEFAULT %s" % (current_table, name, final_default))
                # ID fields need sequences [if they are integers?]
                if name == "id" and set_sequence is True:
                    sequence_lines.append("DROP SEQUENCE IF EXISTS %s_id_seq" % (current_table))
                    sequence_lines.append("CREATE SEQUENCE %s_id_seq" % (current_table))
                    sequence_lines.append("SELECT setval('%s_id_seq', max(id)) FROM %s" % (current_table, current_table))
                    sequence_lines.append("ALTER TABLE "%s" ALTER COLUMN "id" SET DEFAULT nextval('%s_id_seq')" % (current_table, current_table))
                # Record it
                creation_lines.append('"%s" %s %s' % (name, type, extra))
                tables[current_table]['columns'].append((name, type, extra))
            # Is it a constraint or something?
            elif line.startswith("PRIMARY KEY"):
                creation_lines.append(line.rstrip(","))
            elif line.startswith("CONSTRAINT"):
                foreign_key_lines.append("ALTER TABLE "%s" ADD CONSTRAINT %s DEFERRABLE INITIALLY DEFERRED" % (current_table, line.split("CONSTRAINT")[1].strip().rstrip(",")))
                foreign_key_lines.append("CREATE INDEX ON "%s" %s" % (current_table, line.split("FOREIGN KEY")[1].split("REFERENCES")[0].strip().rstrip(",")))
            elif line.startswith("UNIQUE KEY ""):
                index_name      = line.split('"')[1].split('"')[0]
                index_columns   = line.split("(")[1].split(")")[0]
                index_lines.append("CREATE UNIQUE INDEX "%s" ON %s (%s)" % (index_name, current_table, index_columns))
            elif line.startswith("UNIQUE KEY"):
                index_columns   = line.split("(")[1].split(")")[0]
                index_lines.append("CREATE UNIQUE INDEX ON %s (%s)" % (current_table, index_columns))
            elif line.startswith("KEY ""):
                index_name      = line.split('"')[1].split('"')[0]
                index_columns   = line.split("(")[1].split(")")[0]
                index_lines.append("CREATE INDEX "%s" ON %s (%s)" % (index_name, current_table, index_columns))
            elif line.startswith("KEY"):
                index_columns = line.split("(")[1].split(")")[0]
                index_lines.append("CREATE INDEX ON %s (%s)" % (current_table, index_columns))
            elif line.startswith("FULLTEXT KEY"):
                fulltext_keys = " || ' ' || ".join( line.split('(')[-1].split(')')[0].replace('"', '').split(',') )
                index_lines.append("CREATE INDEX ON %s USING gin(to_tsvector('english', %s))" % (current_table, fulltext_keys))
            # Is it the end of the table?
            elif line == ");":
                output.write("CREATE TABLE "%s" (\n" % current_table)
                for i, line in enumerate(creation_lines):
                    output.write("    %s%s\n" % (line, "," if i != (len(creation_lines) - 1) else ""))
                output.write(');\n\n')
                current_table = None
            # ???
            else:
                print "\n ! Unknown line inside table creation: %s" % line


    # Finish file
    output.write("\n-- Post-data save --\n")
    output.write("COMMIT;\n")
    output.write("START TRANSACTION;\n")

    # Write typecasts out
    output.write("\n-- Typecasts --\n")
    for line in cast_lines:
        output.write("%s;\n" % line)

    # Write FK constraints out
    output.write("\n-- Foreign keys --\n")
    for line in foreign_key_lines:
        output.write("%s;\n" % line)

    # Write sequences out
    output.write("\n-- Sequences --\n")
    for line in sequence_lines:
        output.write("%s;\n" % line)

    # Write indexes out
    output.write("\n-- Indexes --\n")
    for line in index_lines:
        output.write("%s;\n" % line)

    # Finish file
    output.write("\n")
    output.write("COMMIT;\n")
    print ""


if __name__ == "__main__":
    parse(sys.argv[1], sys.argv[2])