discourse/script/import_scripts/telligent.rb

493 lines
15 KiB
Ruby

require_relative 'base'
require 'tiny_tds'
# Import script for Telligent communities
#
# Users are currently imported from a temp table. This will need some
# work the next time this import script is used, because that table
# won't exist. Also, it's really hard to find all attachments, but
# the script tries to do it anyway.
class ImportScripts::Telligent < ImportScripts::Base
BATCH_SIZE ||= 1000
LOCAL_AVATAR_REGEX ||= /\A~\/.*(?<directory>communityserver-components-(?:selectable)?avatars)\/(?<path>[^\/]+)\/(?<filename>.+)/i
REMOTE_AVATAR_REGEX ||= /\Ahttps?:\/\//i
EMBEDDED_ATTACHMENT_REGEX ||= /<a href="\/cfs-file(?:\.ashx)?\/__key\/(?<directory>[^\/]+)\/(?<path>[^\/]+)\/(?<filename1>.+)">(?<filename2>.*?)<\/a>/i
CATEGORY_LINK_NORMALIZATION = '/.*?(f\/\d+)$/\1'
TOPIC_LINK_NORMALIZATION = '/.*?(f\/\d+\/t\/\d+)$/\1'
def initialize
super()
@client = TinyTds::Client.new(
host: ENV["DB_HOST"],
username: ENV["DB_USERNAME"],
password: ENV["DB_PASSWORD"],
database: ENV["DB_NAME"]
)
end
def execute
add_permalink_normalizations
import_users
import_categories
import_topics
import_posts
mark_topics_as_solved
end
def import_users
puts "", "Importing users..."
user_conditions = <<~SQL
(
EXISTS(SELECT 1
FROM te_Forum_Threads t
WHERE t.UserId = u.UserID) OR
EXISTS(SELECT 1
FROM te_Forum_ThreadReplies r
WHERE r.UserId = u.UserID)
)
SQL
last_user_id = -1
total_count = count(<<~SQL)
SELECT COUNT(1) AS count
FROM temp_User u
WHERE #{user_conditions}
SQL
batches do |offset|
rows = query(<<~SQL)
SELECT TOP #{BATCH_SIZE} *
FROM
(
SELECT
u.UserID,
u.Email,
u.UserName,
u.CommonName,
u.CreateDate,
p.PropertyName,
p.PropertyValue
FROM temp_User u
LEFT OUTER JOIN temp_UserProperties p ON (u.UserID = p.UserID)
WHERE u.UserID > #{last_user_id} AND #{user_conditions}
) x
PIVOT (
MAX(PropertyValue)
FOR PropertyName
IN (avatarUrl, bio, Location, webAddress, BannedUntil, UserBanReason)
) y
ORDER BY UserID
SQL
break if rows.blank?
last_user_id = rows[-1]["UserID"]
next if all_records_exist?(:users, rows.map { |row| row["UserID"] })
create_users(rows, total: total_count, offset: offset) do |row|
{
id: row["UserID"],
email: row["Email"],
username: row["UserName"],
name: row["CommonName"],
created_at: row["CreateDate"],
bio_raw: html_to_markdown(row["bio"]),
location: row["Location"],
website: row["webAddress"],
post_create_action: proc do |user|
import_avatar(user, row["avatarUrl"])
suspend_user(user, row["BannedUntil"], row["UserBanReason"])
end
}
end
end
end
# TODO move into base importer (create_user) and use consistent error handling
def import_avatar(user, avatar_url)
return if avatar_url.blank? || avatar_url.include?("anonymous")
if match_data = avatar_url.match(LOCAL_AVATAR_REGEX)
avatar_path = File.join(ENV["FILE_BASE_DIR"],
match_data[:directory].gsub("-", "."),
match_data[:path].split("-"),
match_data[:filename])
if File.exists?(avatar_path)
@uploader.create_avatar(user, avatar_path)
else
STDERR.puts "Could not find avatar: #{avatar_path}"
end
elsif avatar_url.match?(REMOTE_AVATAR_REGEX)
UserAvatar.import_url_for_user(avatar_url, user) rescue nil
end
end
def suspend_user(user, banned_until, ban_reason)
return if banned_until.blank?
if banned_until = DateTime.parse(banned_until) > DateTime.now
user.suspended_till = banned_until
user.suspended_at = DateTime.now
user.save!
StaffActionLogger.new(Discourse.system_user).log_user_suspend(user, ban_reason)
end
end
def import_categories
@new_parent_categories = {}
@new_parent_categories[:archives] = create_category({ name: "Archives" }, nil)
@new_parent_categories[:spotlight] = create_category({ name: "Spotlight" }, nil)
@new_parent_categories[:optimizer] = create_category({ name: "SQL Optimizer" }, nil)
puts "", "Importing parent categories..."
parent_categories = query(<<~SQL)
SELECT
GroupID,
Name, HtmlDescription,
DateCreated, SortOrder
FROM cs_Groups g
WHERE (SELECT COUNT(1)
FROM te_Forum_Forums f
WHERE f.GroupId = g.GroupID) > 1
ORDER BY SortOrder, Name
SQL
create_categories(parent_categories) do |row|
{
id: "G#{row['GroupID']}",
name: clean_category_name(row["Name"]),
description: html_to_markdown(row["HtmlDescription"]),
position: row["SortOrder"]
}
end
puts "", "Importing child categories..."
child_categories = query(<<~SQL)
SELECT
ForumId, GroupId,
Name, Description,
DateCreated, SortOrder
FROM te_Forum_Forums
ORDER BY GroupId, SortOrder, Name
SQL
create_categories(child_categories) do |row|
parent_category_id = parent_category_id_for(row)
if category_id = replace_with_category_id(row, child_categories, parent_category_id)
add_category(row['ForumId'], Category.find_by_id(category_id))
url = "f/#{row['ForumId']}"
Permalink.create(url: url, category_id: category_id) unless Permalink.exists?(url: url)
nil
else
{
id: row['ForumId'],
parent_category_id: parent_category_id,
name: clean_category_name(row["Name"]),
description: html_to_markdown(row["Description"]),
position: row["SortOrder"]
}
end
end
end
def parent_category_id_for(row)
name = row["Name"].downcase
if name.include?("beta")
@new_parent_categories[:archives].id
elsif name.include?("spotlight")
@new_parent_categories[:spotlight].id
elsif name.include?("optimizer")
@new_parent_categories[:optimizer].id
elsif row.key?("GroupId")
category_id_from_imported_category_id("G#{row['GroupId']}")
else
nil
end
end
def replace_with_category_id(row, child_categories, parent_category_id)
name = row["Name"].downcase
if name.include?("data modeler") || name.include?("benchmark")
category_id_from_imported_category_id("G#{row['GroupId']}")
elsif only_child?(child_categories, parent_category_id)
parent_category_id
end
end
def only_child?(child_categories, parent_category_id)
count = 0
child_categories.each do |row|
count += 1 if parent_category_id_for(row) == parent_category_id
end
count == 1
end
def clean_category_name(name)
CGI.unescapeHTML(name)
.sub(/(?:\- )?Forum/i, "")
.strip
end
def import_topics
puts "", "Importing topics..."
last_topic_id = -1
total_count = count("SELECT COUNT(1) AS count FROM te_Forum_Threads")
batches do |offset|
rows = query(<<~SQL)
SELECT TOP #{BATCH_SIZE}
t.ThreadId, t.ForumId, t.UserId,
t.Subject, t.Body, t.DateCreated, t.IsLocked, t.StickyDate,
a.ApplicationTypeId, a.ApplicationId, a.ApplicationContentTypeId, a.ContentId, a.FileName
FROM te_Forum_Threads t
LEFT JOIN te_Attachments a
ON (a.ApplicationId = t.ForumId AND a.ApplicationTypeId = 0 AND a.ContentId = t.ThreadId AND
a.ApplicationContentTypeId = 0)
WHERE t.ThreadId > #{last_topic_id}
ORDER BY t.ThreadId
SQL
break if rows.blank?
last_topic_id = rows[-1]["ThreadId"]
next if all_records_exist?(:post, rows.map { |row| import_topic_id(row["ThreadId"]) })
create_posts(rows, total: total_count, offset: offset) do |row|
user_id = user_id_from_imported_user_id(row["UserId"]) || Discourse::SYSTEM_USER_ID
post = {
id: import_topic_id(row["ThreadId"]),
title: CGI.unescapeHTML(row["Subject"]),
raw: raw_with_attachment(row, user_id),
category: category_id_from_imported_category_id(row["ForumId"]),
user_id: user_id,
created_at: row["DateCreated"],
closed: row["IsLocked"],
post_create_action: proc do |action_post|
topic = action_post.topic
Jobs.enqueue_at(topic.pinned_until, :unpin_topic, topic_id: topic.id) if topic.pinned_until
url = "f/#{row['ForumId']}/t/#{row['ThreadId']}"
Permalink.create(url: url, topic_id: topic.id) unless Permalink.exists?(url: url)
end
}
if row["StickyDate"] > Time.now
post[:pinned_until] = row["StickyDate"]
post[:pinned_at] = row["DateCreated"]
end
post
end
end
end
def import_topic_id(topic_id)
"T#{topic_id}"
end
def import_posts
puts "", "Importing posts..."
last_post_id = -1
total_count = count("SELECT COUNT(1) AS count FROM te_Forum_ThreadReplies")
batches do |offset|
rows = query(<<~SQL)
SELECT TOP #{BATCH_SIZE}
tr.ThreadReplyId, tr.ThreadId, tr.UserId, tr.ParentReplyId,
tr.Body, tr.ThreadReplyDate,
CONVERT(BIT,
CASE WHEN tr.AnswerVerifiedUtcDate IS NOT NULL AND NOT EXISTS(
SELECT 1
FROM te_Forum_ThreadReplies x
WHERE
x.ThreadId = tr.ThreadId AND x.ThreadReplyId < tr.ThreadReplyId AND x.AnswerVerifiedUtcDate IS NOT NULL
)
THEN 1
ELSE 0 END) AS IsFirstVerifiedAnswer,
a.ApplicationTypeId, a.ApplicationId, a.ApplicationContentTypeId, a.ContentId, a.FileName
FROM te_Forum_ThreadReplies tr
JOIN te_Forum_Threads t ON (tr.ThreadId = t.ThreadId)
LEFT JOIN te_Attachments a
ON (a.ApplicationId = t.ForumId AND a.ApplicationTypeId = 0 AND a.ContentId = tr.ThreadReplyId AND
a.ApplicationContentTypeId = 1)
WHERE tr.ThreadReplyId > #{last_post_id}
ORDER BY tr.ThreadReplyId
SQL
break if rows.blank?
last_post_id = rows[-1]["ThreadReplyId"]
next if all_records_exist?(:post, rows.map { |row| row["ThreadReplyId"] })
create_posts(rows, total: total_count, offset: offset) do |row|
imported_parent_id = row["ParentReplyId"] > 0 ? row["ParentReplyId"] : import_topic_id(row["ThreadId"])
parent_post = topic_lookup_from_imported_post_id(imported_parent_id)
user_id = user_id_from_imported_user_id(row["UserId"]) || Discourse::SYSTEM_USER_ID
if parent_post
post = {
id: row["ThreadReplyId"],
raw: raw_with_attachment(row, user_id),
user_id: user_id,
topic_id: parent_post[:topic_id],
created_at: row["ThreadReplyDate"],
reply_to_post_number: parent_post[:post_number]
}
post[:custom_fields] = { is_accepted_answer: "true" } if row["IsFirstVerifiedAnswer"]
post
else
puts "Failed to import post #{row['ThreadReplyId']}. Parent was not found."
end
end
end
end
def raw_with_attachment(row, user_id)
raw, embedded_paths, upload_ids = replace_embedded_attachments(row["Body"], user_id)
raw = html_to_markdown(raw) || ""
filename = row["FileName"]
return raw if filename.blank?
path = File.join(
ENV["FILE_BASE_DIR"],
"telligent.evolution.components.attachments",
"%02d" % row["ApplicationTypeId"],
"%02d" % row["ApplicationId"],
"%02d" % row["ApplicationContentTypeId"],
("%010d" % row["ContentId"]).scan(/.{2}/),
clean_filename(filename)
)
unless embedded_paths.include?(path)
if File.exists?(path)
upload = @uploader.create_upload(user_id, path, filename)
if upload.present? && upload.persisted? && !upload_ids.include?(upload.id)
raw << "\n" << @uploader.html_for_upload(upload, filename)
end
else
STDERR.puts "Could not find file: #{path}"
end
end
raw
end
def replace_embedded_attachments(raw, user_id)
paths = []
upload_ids = []
raw = raw.gsub(EMBEDDED_ATTACHMENT_REGEX) do
filename, path = attachment_path(Regexp.last_match)
if File.exists?(path)
upload = @uploader.create_upload(user_id, path, filename)
if upload.present? && upload.persisted?
paths << path
upload_ids << upload.id
@uploader.html_for_upload(upload, filename)
end
else
STDERR.puts "Could not find file: #{path}"
end
end
[raw, paths, upload_ids]
end
def clean_filename(filename)
CGI.unescapeHTML(filename)
.gsub(/[\x00\/\\:\*\?\"<>\|]/, '_')
.gsub(/_(?:2B00|2E00|2D00|5B00|5D00|5F00)/, '')
end
def attachment_path(match_data)
filename, path = join_attachment_path(match_data, filename_index: 2)
filename, path = join_attachment_path(match_data, filename_index: 1) unless File.exists?(path)
[filename, path]
end
# filenames are a total mess - try to guess the correct filename
# works for 70% of all files
def join_attachment_path(match_data, filename_index:)
filename = clean_filename(match_data[:"filename#{filename_index}"])
base_path = File.join(
ENV["FILE_BASE_DIR"],
match_data[:directory].gsub("-", "."),
match_data[:path].split("-")
)
path = File.join(base_path, filename)
return [filename, path] if File.exists?(path)
original_filename = filename.dup
filename = filename.gsub("-", " ")
path = File.join(base_path, filename)
return [filename, path] if File.exists?(path)
filename = filename.gsub("_", "-")
path = File.join(base_path, filename)
return [filename, path] if File.exists?(path)
[original_filename, File.join(base_path, original_filename)]
end
def mark_topics_as_solved
puts "", "Marking topics as solved..."
DB.exec <<~SQL
INSERT INTO topic_custom_fields (name, value, topic_id, created_at, updated_at)
SELECT 'accepted_answer_post_id', pcf.post_id, p.topic_id, p.created_at, p.created_at
FROM post_custom_fields pcf
JOIN posts p ON p.id = pcf.post_id
WHERE pcf.name = 'is_accepted_answer' AND pcf.value = 'true'
SQL
end
def html_to_markdown(html)
HtmlToMarkdown.new(html).to_markdown if html.present?
end
def add_permalink_normalizations
normalizations = SiteSetting.permalink_normalizations
normalizations = normalizations.blank? ? [] : normalizations.split('|')
add_normalization(normalizations, CATEGORY_LINK_NORMALIZATION)
add_normalization(normalizations, TOPIC_LINK_NORMALIZATION)
SiteSetting.permalink_normalizations = normalizations.join('|')
end
def add_normalization(normalizations, normalization)
normalizations << normalization unless normalizations.include?(normalization)
end
def batches
super(BATCH_SIZE)
end
def query(sql)
@client.execute(sql).to_a
end
def count(sql)
query(sql).first["count"]
end
end
ImportScripts::Telligent.new.perform