discourse-data-explorer/lib/tasks/fix_query_ids.rake

132 lines
4.9 KiB
Ruby
Raw Normal View History

# frozen_string_literal: true
desc "Fix query IDs to match the old ones used in the plugin store (q:id)"
task "data_explorer:fix_query_ids" => :environment do
ActiveRecord::Base.transaction do
# Only queries with unique title can be fixed
movements = DB.query <<~SQL
SELECT deq.id AS from, (replace(plugin_store_rows.key, 'q:',''))::integer AS to
FROM plugin_store_rows
INNER JOIN data_explorer_queries deq ON deq.name = plugin_store_rows.value::json->>'name'
WHERE
(replace(plugin_store_rows.key, 'q:',''))::integer != deq.id AND
plugin_store_rows.plugin_name = 'discourse-data-explorer' AND
plugin_store_rows.type_name = 'JSON' AND
(SELECT COUNT(*) from data_explorer_queries deq2 WHERE deq.name = deq2.name) = 1
SQL
if movements.present?
# If there are new queries, they still may have conflict
# We just want to move their ids to safe space and we will not move them back
additional_conflicts =
DB.query(<<~SQL, from: movements.map { |m| m.from }, to: movements.map { |m| m.to })
SELECT id FROM data_explorer_queries
WHERE id IN (:to)
AND id NOT IN (:from)
SQL
additional_conflicts = additional_conflicts.map(&:id)
# Create temporary tables
DB.exec <<~SQL
CREATE TEMPORARY TABLE tmp_data_explorer_queries(
id INTEGER PRIMARY KEY,
name VARCHAR,
description TEXT,
sql TEXT,
user_id INTEGER,
last_run_at TIMESTAMP,
hidden BOOLEAN,
created_at TIMESTAMP,
updated_at TIMESTAMP
) ON COMMIT DROP
SQL
DB.exec <<-SQL
CREATE TEMPORARY TABLE tmp_data_explorer_query_groups(
id INTEGER PRIMARY KEY,
query_id INTEGER,
group_id INTEGER
) ON COMMIT DROP
SQL
movements.each do |movement|
# insert movements to temporary tables
DB.exec <<-SQL
INSERT INTO tmp_data_explorer_queries(id, name, description, sql, user_id, last_run_at, hidden, created_at, updated_at)
SELECT #{movement.to}, name, description, sql, user_id, last_run_at, hidden, created_at, updated_at
FROM data_explorer_queries
WHERE id = #{movement.from}
SQL
DB.exec <<-SQL
INSERT INTO tmp_data_explorer_query_groups(id, query_id, group_id)
SELECT id, #{movement.to}, group_id
FROM data_explorer_query_groups
WHERE query_id = #{movement.from}
SQL
end
# insert rest to temporary tables
already_moved_ids = movements.map(&:from) | additional_conflicts
DB.exec(<<-SQL, already_moved_ids: already_moved_ids)
INSERT INTO tmp_data_explorer_queries(id, name, description, sql, user_id, last_run_at, hidden, created_at, updated_at)
SELECT id, name, description, sql, user_id, last_run_at, hidden, created_at, updated_at
FROM data_explorer_queries
WHERE id NOT IN (:already_moved_ids)
SQL
DB.exec(<<-SQL, already_moved_ids: already_moved_ids)
INSERT INTO tmp_data_explorer_query_groups(id, query_id, group_id)
SELECT id, query_id, group_id
FROM data_explorer_query_groups
WHERE query_id NOT IN (:already_moved_ids)
SQL
# insert additional_conflicts to temporary tables
new_id =
DB.query("select greatest(max(id), 1) from tmp_data_explorer_queries").first.greatest + 1
additional_conflicts.each do |conflict_id|
DB.exec <<-SQL
INSERT INTO tmp_data_explorer_queries(id, name, description, sql, user_id, last_run_at, hidden, created_at, updated_at)
SELECT #{new_id}, name, description, sql, user_id, last_run_at, hidden, created_at, updated_at
FROM data_explorer_queries
WHERE id = #{conflict_id}
SQL
DB.exec <<~SQL
INSERT INTO tmp_data_explorer_query_groups(id, query_id, group_id)
SELECT id, #{new_id}, group_id
FROM data_explorer_query_groups
WHERE query_id = #{conflict_id}
SQL
new_id = new_id + 1
end
# clear original tables and copy data from temporary tables
DB.exec("DELETE FROM data_explorer_queries")
DB.exec("INSERT INTO data_explorer_queries SELECT * FROM tmp_data_explorer_queries")
DB.exec("DELETE FROM data_explorer_query_groups")
DB.exec("INSERT INTO data_explorer_query_groups SELECT * FROM tmp_data_explorer_query_groups")
# Update id sequences
DB.exec <<~SQL
SELECT
setval(
pg_get_serial_sequence('data_explorer_queries', 'id'),
(select greatest(max(id), 1) from data_explorer_queries)
);
SQL
DB.exec <<~SQL
SELECT
setval(
pg_get_serial_sequence('data_explorer_query_groups', 'id'),
(select greatest(max(id), 1) from data_explorer_query_groups)
);
SQL
end
end
end