# frozen_string_literal: true
# AnswerHub Importer
#
# Based on having access to a mysql dump.
# Pass in the ENV variables listed below before running the script.
require_relative "base"
require "mysql2"
require "open-uri"
class ImportScripts::AnswerHub < ImportScripts::Base
DB_NAME ||= ENV["DB_NAME"] || "answerhub"
DB_PASS ||= ENV["DB_PASS"] || "answerhub"
DB_USER ||= ENV["DB_USER"] || "answerhub"
TABLE_PREFIX ||= ENV["TABLE_PREFIX"] || "network1"
BATCH_SIZE ||= ENV["BATCH_SIZE"].to_i || 1000
ATTACHMENT_DIR = ENV["ATTACHMENT_DIR"] || ""
PROCESS_UPLOADS = ENV["PROCESS_UPLOADS"].to_i || 0
ANSWERHUB_DOMAIN = ENV["ANSWERHUB_DOMAIN"]
AVATAR_DIR = ENV["AVATAR_DIR"] || ""
SITE_ID = ENV["SITE_ID"].to_i || 0
CATEGORY_MAP_FROM = ENV["CATEGORY_MAP_FROM"].to_i || 0
CATEGORY_MAP_TO = ENV["CATEGORY_MAP_TO"].to_i || 0
SCRAPE_AVATARS = ENV["SCRAPE_AVATARS"].to_i || 0
def initialize
super
@client =
Mysql2::Client.new(host: "localhost", username: DB_USER, password: DB_PASS, database: DB_NAME)
@skip_updates = true
SiteSetting.tagging_enabled = true
SiteSetting.max_tags_per_topic = 10
end
def execute
puts "Now starting the AnswerHub Import"
puts "DB Name: #{DB_NAME}"
puts "Table Prefix: #{TABLE_PREFIX}"
puts
import_users
import_categories
import_topics
import_posts
import_groups
add_users_to_groups
add_moderators
add_admins
import_avatars
create_permalinks
end
def import_users
puts "", "creating users"
query =
"SELECT count(*) count
FROM #{TABLE_PREFIX}_authoritables
WHERE c_type = 'user'
AND c_active = 1
AND c_system <> 1;"
total_count = @client.query(query).first["count"]
puts "Total count: #{total_count}"
@last_user_id = -1
batches(BATCH_SIZE) do |offset|
query =
"SELECT c_id, c_creation_date, c_name, c_primaryEmail, c_last_seen, c_description
FROM #{TABLE_PREFIX}_authoritables
WHERE c_type = 'user'
AND c_active = 1
AND c_system <> 1
AND c_id > #{@last_user_id}
LIMIT #{BATCH_SIZE};"
results = @client.query(query)
break if results.size < 1
@last_user_id = results.to_a.last["c_id"]
create_users(results, total: total_count, offset: offset) do |user|
# puts user['c_id'].to_s + ' ' + user['c_name']
next if @lookup.user_id_from_imported_user_id(user["c_id"])
{
id: user["c_id"],
email: "#{SecureRandom.hex}@invalid.invalid",
username: user["c_name"],
created_at: user["c_creation_date"],
bio_raw: user["c_description"],
last_seen_at: user["c_last_seen"],
}
end
end
end
def import_categories
puts "", "importing categories..."
# Import parent categories first
query =
"SELECT c_id, c_name, c_plug, c_parent
FROM containers
WHERE c_type = 'space'
AND c_active = 1
AND c_parent = 7 OR c_parent IS NULL"
results = @client.query(query)
create_categories(results) do |c|
{ id: c["c_id"], name: c["c_name"], parent_category_id: check_parent_id(c["c_parent"]) }
end
# Import sub-categories
query =
"SELECT c_id, c_name, c_plug, c_parent
FROM containers
WHERE c_type = 'space'
AND c_active = 1
AND c_parent != 7 AND c_parent IS NOT NULL"
results = @client.query(query)
create_categories(results) do |c|
# puts c.inspect
{
id: c["c_id"],
name: c["c_name"],
parent_category_id: category_id_from_imported_category_id(check_parent_id(c["c_parent"])),
}
end
end
def import_topics
puts "", "importing topics..."
count_query =
"SELECT count(*) count
FROM #{TABLE_PREFIX}_nodes
WHERE c_visibility <> 'deleted'
AND (c_type = 'question'
OR c_type = 'kbentry');"
total_count = @client.query(count_query).first["count"]
@last_topic_id = -1
batches(BATCH_SIZE) do |offset|
# Let's start with just question types
query =
"SELECT *
FROM #{TABLE_PREFIX}_nodes
WHERE c_id > #{@last_topic_id}
AND c_visibility <> 'deleted'
AND (c_type = 'question'
OR c_type = 'kbentry')
ORDER BY c_id ASC
LIMIT #{BATCH_SIZE};"
topics = @client.query(query)
break if topics.size < 1
@last_topic_id = topics.to_a.last["c_id"]
create_posts(topics, total: total_count, offset: offset) do |t|
user_id = user_id_from_imported_user_id(t["c_author"]) || Discourse::SYSTEM_USER_ID
body = process_mentions(t["c_body"])
body = process_uploads(body, user_id) if PROCESS_UPLOADS == 1
markdown_body = HtmlToMarkdown.new(body).to_markdown
{
id: t["c_id"],
user_id: user_id,
title: t["c_title"],
category: category_id_from_imported_category_id(t["c_primaryContainer"]),
raw: markdown_body,
created_at: t["c_creation_date"],
post_create_action:
proc do |post|
tag_names = t["c_topic_names"].split(",")
DiscourseTagging.tag_topic_by_names(post.topic, staff_guardian, tag_names)
end,
}
end
end
end
def import_posts
puts "", "importing posts..."
count_query =
"SELECT count(*) count
FROM #{TABLE_PREFIX}_nodes
WHERE c_visibility <> 'deleted'
AND (c_type = 'answer'
OR c_type = 'comment'
OR c_type = 'kbentry');"
total_count = @client.query(count_query).first["count"]
@last_post_id = -1
batches(BATCH_SIZE) do |offset|
query =
"SELECT *
FROM #{TABLE_PREFIX}_nodes
WHERE c_id > #{@last_post_id}
AND c_visibility <> 'deleted'
AND (c_type = 'answer'
OR c_type = 'comment'
OR c_type = 'kbentry')
ORDER BY c_id ASC
LIMIT #{BATCH_SIZE};"
posts = @client.query(query)
next if all_records_exist? :posts, posts.map { |p| p["c_id"] }
break if posts.size < 1
@last_post_id = posts.to_a.last["c_id"]
create_posts(posts, total: total_count, offset: offset) do |p|
t = topic_lookup_from_imported_post_id(p["c_originalParent"])
next unless t
reply_to_post_id = post_id_from_imported_post_id(p["c_parent"])
reply_to_post = reply_to_post_id.present? ? Post.find(reply_to_post_id) : nil
reply_to_post_number = reply_to_post.present? ? reply_to_post.post_number : nil
user_id = user_id_from_imported_user_id(p["c_author"]) || Discourse::SYSTEM_USER_ID
body = process_mentions(p["c_body"])
body = process_uploads(body, user_id) if PROCESS_UPLOADS == 1
markdown_body = HtmlToMarkdown.new(body).to_markdown
{
id: p["c_id"],
user_id: user_id,
topic_id: t[:topic_id],
reply_to_post_number: reply_to_post_number,
raw: markdown_body,
created_at: p["c_creation_date"],
post_create_action:
proc do |post_info|
begin
if p["c_type"] == "answer" && p["c_marked"] == 1
post = Post.find(post_info[:id])
if post
user_id =
user_id_from_imported_user_id(p["c_author"]) || Discourse::SYSTEM_USER_ID
current_user = User.find(user_id)
solved = DiscourseSolved.accept_answer!(post, current_user)
# puts "SOLVED: #{solved}"
end
end
rescue ActiveRecord::RecordInvalid
puts "SOLVED: Skipped post_id: #{post.id} because invalid"
end
end,
}
end
end
end
def import_groups
puts "", "importing groups..."
query =
"SELECT c_id, c_name
FROM network6_authoritables
WHERE c_type='group'
AND c_id > 6;" # Ignore Anonymous, Users, Moderators, etc.
groups = @client.query(query)
create_groups(groups) do |group|
{ id: group["c_id"], name: group["c_name"], visibility_level: 1 }
end
end
def add_users_to_groups
puts "", "adding users to groups..."
query =
"SELECT c_id, c_name
FROM network6_authoritables
WHERE c_type='group'
AND c_id > 6;" # Ignore Anonymous, Users, Moderators, etc.
groups = @client.query(query)
members_query =
"SELECT *
FROM network6_authoritable_groups;"
group_members = @client.query(members_query)
total_count = groups.count
progress_count = 0
start_time = Time.now
group_members.map
groups.each do |group|
dgroup = find_group_by_import_id(group["c_id"])
next if dgroup.custom_fields["import_users_added"]
group_member_ids =
group_members
.map do |m|
user_id_from_imported_user_id(m["c_members"]) if m["c_groups"] == group["c_id"]
end
.compact
# add members
dgroup.bulk_add(group_member_ids)
# reload group
dgroup.reload
dgroup.custom_fields["import_users_added"] = true
dgroup.save
progress_count += 1
print_status(progress_count, total_count, start_time)
end
end
def add_moderators
puts "", "adding moderators..."
query =
"SELECT *
FROM network6_authoritable_groups
WHERE c_groups = 4;"
moderators = @client.query(query)
moderator_ids = moderators.map { |m| user_id_from_imported_user_id(m["c_members"]) }.compact
moderator_ids.each do |id|
user = User.find(id)
user.grant_moderation!
end
end
def add_admins
puts "", "adding admins..."
query =
"SELECT *
FROM network6_authoritable_groups
WHERE c_groups = 5 OR c_groups = 6;" # Super Users, Network Administrators
admins = @client.query(query)
admin_ids = admins.map { |a| user_id_from_imported_user_id(a["c_members"]) }.compact
admin_ids.each do |id|
user = User.find(id)
user.grant_admin!
end
end
def import_avatars
puts "", "importing user avatars"
query =
"SELECT *
FROM network6_user_preferences
WHERE c_key = 'avatarImage'"
avatars = @client.query(query)
avatars.each do |a|
begin
user_id = user_id_from_imported_user_id(a["c_user"])
user = User.find(user_id)
if user
filename = "avatar-#{user_id}.png"
path = File.join(AVATAR_DIR, filename)
next if !File.exist?(path)
# Scrape Avatars - Avatars are saved in the db, but it might be easier to just scrape them
if SCRAPE_AVATARS == 1
File.open(path, "wb") do |f|
f << open(
"https://#{ANSWERHUB_DOMAIN}/forums/users/#{a["c_user"]}/photo/view.html?s=240",
).read
end
end
upload = @uploader.create_upload(user.id, path, filename)
if upload.persisted?
user.import_mode = false
user.create_user_avatar
user.import_mode = true
user.user_avatar.update(custom_upload_id: upload.id)
user.update(uploaded_avatar_id: upload.id)
else
Rails.logger.error("Could not persist avatar for user #{user.username}")
end
end
rescue ActiveRecord::RecordNotFound
puts "Could not find User for user_id: #{a["c_user"]}"
end
end
end
def process_uploads(body, user_id)
if body.match(//)
# There could be multiple images in a post
images = body.scan(//)
images.each do |image|
filepath = File.basename(image).split('"')[0]
filepath = File.join(ATTACHMENT_DIR, filepath)
if File.exist?(filepath)
filename = File.basename(filepath)
upload = create_upload(user_id, filepath, filename)
image_html = html_for_upload(upload, filename)
original_image_html = ''
body.sub!(original_image_html, image_html)
end
end
end
# Non-images
if body.match(//)
# There could be multiple files in a post
files = body.scan(//)
files.each do |file|
filepath = File.basename(file).split('"')[0]
filepath = File.join(ATTACHMENT_DIR, filepath)
if File.exist?(filepath)
filename = File.basename(filepath)
upload = create_upload(user_id, filepath, filename)
file_html = html_for_upload(upload, filename)
original_file_html = ''
body.sub!(original_file_html, file_html)
end
end
end
body
end
def process_mentions(body)
raw = body.dup
# https://example.forum.com/forums/users/1469/XYZ_Rob.html
raw.gsub!(%r{(https://example.forum.com/forums/users/\d+/[\w_%-.]*.html)}) do
legacy_url = $1
import_user_id =
legacy_url.match(%r{https://example.forum.com/forums/users/(\d+)/[\w_%-.]*.html}).captures
user = @lookup.find_user_by_import_id(import_user_id[0])
if user.present?
# puts "/users/#{user.username}"
"/users/#{user.username}"
else
# puts legacy_url
legacy_url
end
end
# /forums/users/395/petrocket.html
raw.gsub!(%r{(/forums/users/\d+/[\w_%-.]*.html)}) do
legacy_url = $1
import_user_id = legacy_url.match(%r{/forums/users/(\d+)/[\w_%-.]*.html}).captures
# puts raw
user = @lookup.find_user_by_import_id(import_user_id[0])
if user.present?
# puts "/users/#{user.username}"
"/users/#{user.username}"
else
# puts legacy_url
legacy_url
end
end
raw
end
def create_permalinks
puts "", "Creating redirects...", ""
# https://example.forum.com/forums/questions/2005/missing-file.html
Topic.find_each do |topic|
pcf = topic.first_post.custom_fields
if pcf && pcf["import_id"]
id = pcf["import_id"]
slug = Slug.for(topic.title)
begin
Permalink.create(url: "questions/#{id}/#{slug}.html", topic_id: topic.id)
rescue StandardError
nil
end
print "."
end
end
end
def staff_guardian
@_staff_guardian ||= Guardian.new(Discourse.system_user)
end
# Some category parent id's need to be adjusted
def check_parent_id(id)
return nil if SITE_ID > 0 && id == SITE_ID
return CATEGORY_MAP_TO if CATEGORY_MAP_FROM > 0 && id == CATEGORY_MAP_FROM
id
end
end
ImportScripts::AnswerHub.new.perform