mirror of
https://github.com/discourse/discourse.git
synced 2024-11-22 05:59:39 +08:00
508 lines
15 KiB
Ruby
508 lines
15 KiB
Ruby
# frozen_string_literal: true
|
|
|
|
# Yammer importer
|
|
# https://docs.microsoft.com/en-us/yammer/manage-security-and-compliance/export-yammer-enterprise-data#export-yammer-network-data-by-date-range-and-network
|
|
#
|
|
# You will need a bunch of CSV files:
|
|
#
|
|
# - Users.csv Groups.csv Topics.csv Groups.csv Files.csv Messages.csv
|
|
# (Others included in Yammer export are ignored)
|
|
|
|
require "csv"
|
|
require_relative "base"
|
|
require_relative "base/generic_database"
|
|
|
|
# Call it like this:
|
|
# RAILS_ENV=production bundle exec ruby script/import_scripts/yammer.rb DIRNAME
|
|
|
|
class ImportScripts::Yammer < ImportScripts::Base
|
|
BATCH_SIZE = 1000
|
|
NUM_WORDS_IN_TITLE = ENV["NUM_WORDS_IN_TITLE"].to_i || 20
|
|
SKIP_EMPTY_EMAIL = true
|
|
SKIP_INACTIVE_USERS = false
|
|
PARENT_CATEGORY_NAME = ENV["PARENT_CATEGORY_NAME"] || "Yammer Import"
|
|
IMPORT_GROUPS_AS_TAGS = true
|
|
MERGE_USERS = true
|
|
# import groups as tags rather than as categories
|
|
SiteSetting.tagging_enabled = true if IMPORT_GROUPS_AS_TAGS
|
|
PM_TAG = ENV["PM_TAG"] || "eht"
|
|
|
|
def initialize(path)
|
|
super()
|
|
|
|
@path = path
|
|
@db = ImportScripts::GenericDatabase.new(@path, batch_size: BATCH_SIZE, recreate: true)
|
|
end
|
|
|
|
def execute
|
|
create_developer_users
|
|
read_csv_files
|
|
|
|
import_categories
|
|
import_users
|
|
import_topics
|
|
import_pm_topics
|
|
import_posts
|
|
import_pm_posts
|
|
end
|
|
|
|
def create_developer_users
|
|
GlobalSetting
|
|
.developer_emails
|
|
.split(",")
|
|
.each { |e| User.create(email: e, active: true, username: e.split("@")[0]) }
|
|
end
|
|
|
|
def read_csv_files
|
|
puts "", "reading CSV files"
|
|
|
|
# consider csv_parse Tags.csv
|
|
# consider Admins.csv that has admins
|
|
|
|
u_count = 0
|
|
csv_parse("Users") do |row|
|
|
next if SKIP_INACTIVE_USERS && row[:state] != "active"
|
|
u_count += 1
|
|
@db.insert_user(
|
|
id: row[:id],
|
|
email: row[:email],
|
|
name: row[:name],
|
|
username: row[:name],
|
|
bio: "",
|
|
# job_title: row[:job_title],
|
|
# location: row[:location],
|
|
# department: row[:department],
|
|
created_at: parse_datetime(row[:joined_at]),
|
|
# deleted_at: parse_datetime(row[:deleted_at]),
|
|
# suspended_at: parse_datetime(row[:suspended_at]),
|
|
# guid: row[:guid],
|
|
# state: row[:state],
|
|
avatar_path: row[:user_cover_image_url],
|
|
# last_seen_at: ,
|
|
active: row[:state] == "active" ? 1 : 0,
|
|
)
|
|
end
|
|
|
|
category_position = 0
|
|
csv_parse("Groups") do |row|
|
|
@db.insert_category(
|
|
id: row[:id],
|
|
name: row[:name],
|
|
description: row[:description],
|
|
position: category_position += 1,
|
|
)
|
|
end
|
|
|
|
csv_parse("Files") do |row|
|
|
@db.insert_upload(
|
|
id: row[:file_id],
|
|
user_id: row[:uploader_id],
|
|
original_filename: row[:name],
|
|
filename: row[:path],
|
|
description: row[:description],
|
|
)
|
|
end
|
|
|
|
# get topics from messages
|
|
csv_parse("Messages") do |row|
|
|
next unless row[:thread_id] == row[:id]
|
|
next if row[:in_private_conversation] == "true"
|
|
next if row[:deleted_at].present?
|
|
# next if row[:message_type] == 'system'
|
|
title = ""
|
|
url = ""
|
|
description = ""
|
|
raw = row[:body]
|
|
reg = /opengraphobject:\[(\d*?) : (.*?) : title="(.*?)" : description="(.*?)"\]/
|
|
if row[:attachments]
|
|
row[:attachments].match(reg) do
|
|
url = Regexp.last_match(2)
|
|
title = Regexp.last_match(3) if Regexp.last_match(3)
|
|
description = Regexp.last_match(4)
|
|
raw += "\n***\n#{description}\n#{url}\n" if raw.exclude?(url)
|
|
end
|
|
row[:attachments].match(/uploadedfile:(\d*)$/) do
|
|
file_id = Regexp.last_match(1).to_i
|
|
up = @db.fetch_upload(file_id).first
|
|
path = File.join(@path, up["filename"])
|
|
filename = up["original_filename"]
|
|
user_id = user_id_from_imported_user_id(row["user_id"]) || Discourse.system_user.id
|
|
if File.exist?(path)
|
|
upload = create_upload(user_id, path, filename)
|
|
raw += html_for_upload(upload, filename) if upload&.persisted?
|
|
end
|
|
end
|
|
end
|
|
@db.insert_topic(
|
|
id: row[:id],
|
|
title: title,
|
|
raw: raw,
|
|
category_id: row[:group_id],
|
|
closed: row[:closed] == "TRUE" ? 1 : 0,
|
|
user_id: row[:sender_id],
|
|
created_at: parse_datetime(row[:created_at]),
|
|
)
|
|
end
|
|
|
|
# get pm topics
|
|
csv_parse("Messages") do |row|
|
|
next unless row[:thread_id] == row[:id]
|
|
next unless row[:in_private_conversation] == "true"
|
|
next if row[:deleted_at].present?
|
|
# next if row[:message_type] == 'system'
|
|
title = ""
|
|
url = ""
|
|
description = ""
|
|
raw = row[:body]
|
|
reg = /opengraphobject:\[(\d*?) : (.*?) : title="(.*?)" : description="(.*?)"\]/
|
|
if row[:attachments]
|
|
row[:attachments].match(reg) do
|
|
url = Regexp.last_match(2)
|
|
title = Regexp.last_match(3) if Regexp.last_match(3)
|
|
description = Regexp.last_match(4)
|
|
raw += "\n***\n#{description}\n#{url}\n" if raw.exclude?(url)
|
|
end
|
|
row[:attachments].match(/uploadedfile:(\d*)$/) do
|
|
file_id = Regexp.last_match(1).to_i
|
|
up = @db.fetch_upload(file_id).first
|
|
path = File.join(@path, up["filename"])
|
|
filename = up["original_filename"]
|
|
user_id = user_id_from_imported_user_id(row["user_id"]) || Discourse.system_user.id
|
|
if File.exist?(path)
|
|
upload = create_upload(user_id, path, filename)
|
|
raw += html_for_upload(upload, filename) if upload&.persisted?
|
|
end
|
|
end
|
|
end
|
|
@db.insert_pm_topic(
|
|
id: row[:id],
|
|
title: title,
|
|
raw: raw,
|
|
category_id: row[:group_id],
|
|
closed: row[:closed] == "TRUE" ? 1 : 0,
|
|
target_users: row[:participants].gsub("user:", ""),
|
|
user_id: row[:sender_id],
|
|
created_at: parse_datetime(row[:created_at]),
|
|
)
|
|
end
|
|
|
|
# get posts from messages
|
|
csv_parse("Messages") do |row|
|
|
next if row[:thread_id] == row[:id]
|
|
next if row[:deleted_at].present?
|
|
next if row[:in_private_conversation] == "true"
|
|
@db.insert_post(
|
|
id: row[:id],
|
|
raw: row[:body] + "\n" + row[:attachments],
|
|
topic_id: row[:thread_id],
|
|
reply_to_post_id: row[:replied_to_id],
|
|
user_id: row[:sender_id],
|
|
created_at: parse_datetime(row[:created_at]),
|
|
)
|
|
end
|
|
|
|
# get pm posts from messages
|
|
csv_parse("Messages") do |row|
|
|
next if row[:thread_id] == row[:id]
|
|
next if row[:deleted_at].present?
|
|
next unless row[:in_private_conversation] == "false"
|
|
@db.insert_pm_post(
|
|
id: row[:id],
|
|
raw: row[:body] + "\n" + row[:attachments],
|
|
topic_id: row[:thread_id],
|
|
reply_to_post_id: row[:replied_to_id],
|
|
user_id: row[:sender_id],
|
|
created_at: parse_datetime(row[:created_at]),
|
|
)
|
|
end
|
|
|
|
#@db.delete_unused_users
|
|
@db.sort_posts_by_created_at
|
|
end
|
|
|
|
def parse_datetime(text)
|
|
return nil if text.blank? || text == "null"
|
|
DateTime.parse(text)
|
|
end
|
|
|
|
def import_categories
|
|
puts "", "creating categories"
|
|
parent_category = nil
|
|
if !PARENT_CATEGORY_NAME.blank?
|
|
parent_category = Category.find_by(name: PARENT_CATEGORY_NAME)
|
|
parent_category =
|
|
Category.create(
|
|
name: PARENT_CATEGORY_NAME,
|
|
user_id: Discourse.system_user.id,
|
|
) unless parent_category
|
|
end
|
|
|
|
if IMPORT_GROUPS_AS_TAGS
|
|
@tag_map = {}
|
|
rows = @db.fetch_categories
|
|
rows.each { |row| @tag_map[row["id"]] = row["name"] }
|
|
else
|
|
rows = @db.fetch_categories
|
|
create_categories(rows) do |row|
|
|
{
|
|
id: row["id"],
|
|
name: row["name"],
|
|
description: row["description"],
|
|
position: row["position"],
|
|
parent_category_id: parent_category,
|
|
}
|
|
end
|
|
end
|
|
end
|
|
|
|
def batches
|
|
super(BATCH_SIZE)
|
|
end
|
|
|
|
def import_users
|
|
puts "", "creating users"
|
|
total_count = @db.count_users
|
|
puts "", "Got #{total_count} users!"
|
|
last_id = ""
|
|
|
|
batches do |offset|
|
|
rows, last_id = @db.fetch_users(last_id)
|
|
break if rows.empty?
|
|
|
|
next if all_records_exist?(:users, rows.map { |row| row["id"] })
|
|
|
|
create_users(rows, total: total_count, offset: offset) do |row|
|
|
user = User.find_by_email(row["email"].downcase)
|
|
if user
|
|
user.custom_fields["import_id"] = row["id"]
|
|
user.custom_fields["matched_existing"] = "yes"
|
|
user.save
|
|
add_user(row["id"].to_s, user)
|
|
next
|
|
end
|
|
{
|
|
id: row["id"],
|
|
email: row["email"],
|
|
name: row["name"],
|
|
created_at: row["created_at"],
|
|
last_seen_at: row["last_seen_at"],
|
|
active: row["active"] == 1,
|
|
}
|
|
end
|
|
end
|
|
end
|
|
|
|
def import_topics
|
|
puts "", "creating topics"
|
|
staff_guardian = Guardian.new(Discourse.system_user)
|
|
|
|
total_count = @db.count_topics
|
|
last_id = ""
|
|
|
|
batches do |offset|
|
|
rows, last_id = @db.fetch_topics(last_id)
|
|
base_category = Category.find_by(name: PARENT_CATEGORY_NAME)
|
|
break if rows.empty?
|
|
|
|
next if all_records_exist?(:posts, rows.map { |row| import_topic_id(row["id"]) })
|
|
|
|
create_posts(rows, total: total_count, offset: offset) do |row|
|
|
{
|
|
id: import_topic_id(row["id"]),
|
|
title:
|
|
(
|
|
if row["title"].present?
|
|
row["title"]
|
|
else
|
|
row["raw"].split(/\W/)[0..(NUM_WORDS_IN_TITLE - 1)].join(" ")
|
|
end
|
|
),
|
|
raw: normalize_raw(row["raw"]),
|
|
category:
|
|
(
|
|
if IMPORT_GROUPS_AS_TAGS
|
|
base_category.id
|
|
else
|
|
category_id_from_imported_category_id(row["category_id"])
|
|
end
|
|
),
|
|
user_id: user_id_from_imported_user_id(row["user_id"]) || Discourse.system_user.id,
|
|
created_at: row["created_at"],
|
|
closed: row["closed"] == 1,
|
|
post_create_action:
|
|
proc do |post|
|
|
if IMPORT_GROUPS_AS_TAGS
|
|
topic = Topic.find(post.topic_id)
|
|
tag_names = [@tag_map[row["category_id"]]]
|
|
DiscourseTagging.tag_topic_by_names(topic, staff_guardian, tag_names)
|
|
end
|
|
end,
|
|
}
|
|
end
|
|
end
|
|
end
|
|
|
|
def import_pm_topics
|
|
puts "", "creating pm topics"
|
|
staff_guardian = Guardian.new(Discourse.system_user)
|
|
|
|
total_count = @db.count_pm_topics
|
|
last_id = ""
|
|
|
|
batches do |offset|
|
|
rows, last_id = @db.fetch_pm_topics(last_id)
|
|
base_category = Category.find_by(name: PARENT_CATEGORY_NAME)
|
|
break if rows.empty?
|
|
|
|
next if all_records_exist?(:posts, rows.map { |row| import_topic_id(row["id"]) })
|
|
create_posts(rows, total: total_count, offset: offset) do |row|
|
|
target_users = []
|
|
row["target_users"]
|
|
.split(",")
|
|
.each do |u|
|
|
user_id = user_id_from_imported_user_id(u)
|
|
next unless user_id
|
|
user = User.find(user_id)
|
|
target_users.append(user.username)
|
|
end
|
|
target_usernames = target_users.join(",")
|
|
{
|
|
id: import_topic_id(row["id"]),
|
|
title:
|
|
(
|
|
if row["title"].present?
|
|
row["title"]
|
|
else
|
|
row["raw"].split(/\W/)[0..(NUM_WORDS_IN_TITLE - 1)].join(" ")
|
|
end
|
|
),
|
|
raw: normalize_raw(row["raw"]),
|
|
category:
|
|
(
|
|
if IMPORT_GROUPS_AS_TAGS
|
|
base_category.id
|
|
else
|
|
category_id_from_imported_category_id(row["category_id"])
|
|
end
|
|
),
|
|
user_id: user_id_from_imported_user_id(row["user_id"]) || Discourse.system_user.id,
|
|
created_at: row["created_at"],
|
|
closed: row["closed"] == 1,
|
|
archetype: Archetype.private_message,
|
|
target_usernames: target_usernames,
|
|
post_create_action:
|
|
proc do |post|
|
|
if PM_TAG
|
|
topic = Topic.find(post.topic_id)
|
|
tag_names = [PM_TAG]
|
|
DiscourseTagging.tag_topic_by_names(topic, staff_guardian, tag_names)
|
|
end
|
|
end,
|
|
}
|
|
end
|
|
end
|
|
end
|
|
|
|
def import_topic_id(topic_id)
|
|
"T#{topic_id}"
|
|
end
|
|
|
|
def import_posts
|
|
puts "", "creating posts"
|
|
total_count = @db.count_posts
|
|
last_row_id = 0
|
|
|
|
batches do |offset|
|
|
rows, last_row_id = @db.fetch_sorted_posts(last_row_id)
|
|
break if rows.empty?
|
|
|
|
next if all_records_exist?(:posts, rows.map { |row| row["id"] })
|
|
|
|
create_posts(rows, total: total_count, offset: offset) do |row|
|
|
topic = topic_lookup_from_imported_post_id(import_topic_id(row["topic_id"]))
|
|
if topic.nil?
|
|
p "MISSING TOPIC #{row["topic_id"]}"
|
|
p row
|
|
next
|
|
end
|
|
{
|
|
id: row["id"],
|
|
raw: normalize_raw(row["raw"]),
|
|
user_id: user_id_from_imported_user_id(row["user_id"]) || Discourse.system_user.id,
|
|
topic_id: topic[:topic_id],
|
|
created_at: row["created_at"],
|
|
}
|
|
end
|
|
end
|
|
end
|
|
|
|
def import_pm_posts
|
|
puts "", "creating pm posts"
|
|
total_count = @db.count_pm_posts
|
|
last_row_id = 0
|
|
|
|
batches do |offset|
|
|
rows, last_row_id = @db.fetch_pm_posts(last_row_id)
|
|
break if rows.empty?
|
|
|
|
next if all_records_exist?(:posts, rows.map { |row| row["id"] })
|
|
|
|
create_posts(rows, total: total_count, offset: offset) do |row|
|
|
topic = topic_lookup_from_imported_post_id(import_topic_id(row["topic_id"]))
|
|
|
|
if topic.nil?
|
|
p "MISSING TOPIC #{row["topic_id"]}"
|
|
p row
|
|
next
|
|
end
|
|
|
|
{
|
|
id: row["id"],
|
|
raw: normalize_raw(row["raw"]),
|
|
user_id: user_id_from_imported_user_id(row["user_id"]) || Discourse.system_user.id,
|
|
topic_id: topic[:topic_id],
|
|
created_at: row["created_at"],
|
|
}
|
|
end
|
|
end
|
|
end
|
|
|
|
def normalize_raw(raw)
|
|
return "<missing>" if raw.blank?
|
|
|
|
raw = raw.gsub('\n', "")
|
|
raw.gsub!(/\[\[user:(\d+)\]\]/) do
|
|
u = Regexp.last_match(1)
|
|
user_id = user_id_from_imported_user_id(u) || Discourse.system_user.id
|
|
if user_id
|
|
user = User.find(user_id)
|
|
"@#{user.username}"
|
|
else
|
|
u
|
|
end
|
|
end
|
|
raw
|
|
end
|
|
|
|
def permalink_exists?(url)
|
|
Permalink.find_by(url: url)
|
|
end
|
|
|
|
def csv_parse(table_name)
|
|
CSV.foreach(
|
|
File.join(@path, "#{table_name}.csv"),
|
|
headers: true,
|
|
header_converters: :symbol,
|
|
skip_blanks: true,
|
|
encoding: "bom|utf-8",
|
|
) { |row| yield row }
|
|
end
|
|
end
|
|
|
|
unless ARGV[0] && Dir.exist?(ARGV[0])
|
|
puts "", "Usage:", "", "bundle exec ruby script/import_scripts/yammer.rb DIRNAME", ""
|
|
exit 1
|
|
end
|
|
|
|
ImportScripts::Yammer.new(ARGV[0]).perform
|