Do not report progress on scripted s3 uploads

[chef.git] / cookbooks / planet / files / default / replication-bin / replicate-changesets

diff --git a/cookbooks/planet/files/default/replication-bin/replicate-changesets b/cookbooks/planet/files/default/replication-bin/replicate-changesets
old mode 100644 (file)
new mode 100755 (executable)
index e2d4cc1..5d47266
--- a/cookbooks/planet/files/default/replication-bin/replicate-changesets
+++ b/cookbooks/planet/files/default/replication-bin/replicate-changesets
@@ -1,18 +1,25 @@
 #!/usr/bin/ruby
 
 #!/usr/bin/ruby
 

-require 'rubygems'
-require 'pg'
-require 'yaml'
-require 'time'
-require 'fileutils'
-require 'xml/libxml'
-require 'zlib'
+require "rubygems"
+require "pg"
+require "yaml"
+require "time"
+require "fileutils"
+require "xml/libxml"
+require "zlib"
+require "set"

 
 # after this many changes, a changeset will be closed
 
 # after this many changes, a changeset will be closed

-CHANGES_LIMIT=50000
+CHANGES_LIMIT = 50000

 
 # this is the scale factor for lat/lon values stored as integers in the database
 
 # this is the scale factor for lat/lon values stored as integers in the database

-GEO_SCALE=10000000
+GEO_SCALE = 10000000
+
+##
+# replace characters which cannot be represented in XML 1.0.
+def xml_sanitize(str)
+  str.gsub(/[\x00-\x08\x0b\x0c\x0e-\x1f]/, "?")
+end

 
 ##
 # changeset class keeps some information about changesets downloaded from the
 
 ##
 # changeset class keeps some information about changesets downloaded from the


@@ -22,10 +29,10 @@ class Changeset
   attr_reader :id, :created_at, :closed_at, :num_changes
 
   def initialize(row)
   attr_reader :id, :created_at, :closed_at, :num_changes
 
   def initialize(row)

-    @id = row['id'].to_i
-    @created_at = Time.parse(row['created_at'])
-    @closed_at = Time.parse(row['closed_at'])
-    @num_changes = row['num_changes'].to_i
+    @id = row["id"].to_i
+    @created_at = Time.parse(row["created_at"])
+    @closed_at = Time.parse(row["closed_at"])
+    @num_changes = row["num_changes"].to_i

   end
 
   def closed?(t)
   end
 
   def closed?(t)


@@ -33,7 +40,7 @@ class Changeset
   end
 
   def open?(t)
   end
 
   def open?(t)

-    not closed?(t)
+    !closed?(t)

   end
 
   def activity_between?(t1, t2)
   end
 
   def activity_between?(t1, t2)


@@ -41,106 +48,252 @@ class Changeset
   end
 end
 
   end
 end
 

+##
+# builds an XML representation of a changeset from the database
+class ChangesetBuilder
+  def initialize(now, conn)
+    @now = now
+    @conn = conn
+  end
+
+  def changeset_xml(cs)
+    xml = XML::Node.new("changeset")
+    xml["id"] = cs.id.to_s
+    xml["created_at"] = cs.created_at.getutc.xmlschema
+    xml["closed_at"] = cs.closed_at.getutc.xmlschema if cs.closed?(@now)
+    xml["open"] = cs.open?(@now).to_s
+    xml["num_changes"] = cs.num_changes.to_s
+
+    res = @conn.exec("select u.id, u.display_name, c.min_lat, c.max_lat, c.min_lon, c.max_lon from users u join changesets c on u.id=c.user_id where c.id=#{cs.id}")
+    xml["user"] = xml_sanitize(res[0]["display_name"])
+    xml["uid"] = res[0]["id"]
+
+    unless res[0]["min_lat"].nil? ||
+           res[0]["max_lat"].nil? ||
+           res[0]["min_lon"].nil? ||
+           res[0]["max_lon"].nil?
+      xml["min_lat"] = (res[0]["min_lat"].to_f / GEO_SCALE).to_s
+      xml["max_lat"] = (res[0]["max_lat"].to_f / GEO_SCALE).to_s
+      xml["min_lon"] = (res[0]["min_lon"].to_f / GEO_SCALE).to_s
+      xml["max_lon"] = (res[0]["max_lon"].to_f / GEO_SCALE).to_s
+    end
+
+    add_tags(xml, cs)
+    add_comments(xml, cs)
+
+    xml
+  end
+
+  def add_tags(xml, cs)
+    res = @conn.exec("select k, v from changeset_tags where changeset_id=#{cs.id}")
+    res.each do |row|
+      tag = XML::Node.new("tag")
+      tag["k"] = xml_sanitize(row["k"])
+      tag["v"] = xml_sanitize(row["v"])
+      xml << tag
+    end
+  end
+
+  def add_comments(xml, cs)
+    # grab the visible changeset comments as well
+    res = @conn.exec("select cc.author_id, u.display_name as author, cc.body, cc.created_at from changeset_comments cc join users u on cc.author_id=u.id where cc.changeset_id=#{cs.id} and cc.visible order by cc.created_at asc")
+    xml["comments_count"] = res.num_tuples.to_s
+
+    # early return if there aren't any comments
+    return unless res.num_tuples.positive?
+
+    discussion = XML::Node.new("discussion")
+    res.each do |row|
+      comment = XML::Node.new("comment")
+      comment["uid"] = row["author_id"]
+      comment["user"] = xml_sanitize(row["author"])
+      comment["date"] = Time.parse(row["created_at"]).getutc.xmlschema
+      text = XML::Node.new("text")
+      text.content = xml_sanitize(row["body"])
+      comment << text
+      discussion << comment
+    end
+    xml << discussion
+  end
+end
+
+##
+# sync a file to guarantee it's on disk
+def fsync(f)
+  File.open(f, &:fsync)
+end
+
+##
+# sync a directory to guarantee it's on disk. have to recurse to the root
+# to guarantee sync for newly created directories.
+def fdirsync(d)
+  while d != "/"
+    Dir.open(d) do |dh|
+      io = IO.for_fd(dh.fileno)
+      io.fsync
+    end
+    d = File.dirname(d)
+  end
+end
+

 ##
 # state and connections associated with getting changeset data
 # replicated to a file.
 class Replicator
   def initialize(config)
 ##
 # state and connections associated with getting changeset data
 # replicated to a file.
 class Replicator
   def initialize(config)

-    @config = YAML.load(File.read(config))
-    @state = YAML.load(File.read(@config['state_file']))
-    @conn = PGconn.connect(@config['db'])
-    @now = Time.now.getutc
+    @config = YAML.safe_load(File.read(config))
+    @state = YAML.safe_load(File.read(@config["state_file"]), [Time])
+    @conn = PG::Connection.connect(@config["db"])
+    # get current time from the database rather than the current system
+    @now = @conn.exec("select now() as now").map { |row| Time.parse(row["now"]) }[0]

   end
 
   def open_changesets
   end
 
   def open_changesets

-    last_run = @state['last_run']
+    last_run = @state["last_run"]

     last_run = (@now - 60) if last_run.nil?
     last_run = (@now - 60) if last_run.nil?

-    @state['last_run'] = @now

     # pretty much all operations on a changeset will modify its closed_at
     # pretty much all operations on a changeset will modify its closed_at

-    # time (see rails_port's changeset model). so it is probably enough 
+    # time (see rails_port's changeset model). so it is probably enough

     # for us to look at anything that was closed recently, and filter from
     # there.
     # for us to look at anything that was closed recently, and filter from
     # there.

-    @conn.
-      exec("select id, created_at, closed_at, num_changes from changesets where closed_at > ((now() at time zone 'utc') - '1 hour'::interval)").
-      map {|row| Changeset.new(row) }.
-      select {|cs| cs.activity_between?(last_run, @now) }
+    changesets = @conn
+                 .exec("select id, created_at, closed_at, num_changes from changesets where closed_at > ((now() at time zone 'utc') - '1 hour'::interval)")
+                 .map { |row| Changeset.new(row) }
+                 .select { |cs| cs.activity_between?(last_run, @now) }
+
+    # set for faster presence lookups by ID
+    cs_ids = Set.new(changesets.map(&:id))
+
+    # but also add any changesets which have new comments
+    new_ids = @conn
+              .exec("select distinct changeset_id from changeset_comments where created_at >= '#{last_run}' and created_at < '#{@now}' and visible")
+              .map { |row| row["changeset_id"].to_i }
+              .reject { |c_id| cs_ids.include?(c_id) }
+
+    new_ids.each do |id|
+      @conn
+        .exec("select id, created_at, closed_at, num_changes from changesets where id=#{id}")
+        .map { |row| Changeset.new(row) }
+        .each { |cs| changesets << cs }
+    end
+
+    changesets.sort_by(&:id)

   end
 
   end
 

-  # creates an XML file containing the changeset information from the 
+  # creates an XML file containing the changeset information from the

   # list of changesets output by open_changesets.
   def changeset_dump(changesets)
     doc = XML::Document.new
     doc.root = XML::Node.new("osm")
   # list of changesets output by open_changesets.
   def changeset_dump(changesets)
     doc = XML::Document.new
     doc.root = XML::Node.new("osm")

-    { 'version' => '0.6',
-      'generator' => 'replicate_changesets.rb',
-      'copyright' => "OpenStreetMap and contributors",
-      'attribution' => "http://www.openstreetmap.org/copyright",
-      'license' => "http://opendatacommons.org/licenses/odbl/1-0/" }.
-      each { |k,v| doc.root[k] = v }
+    { "version" => "0.6",
+      "generator" => "replicate_changesets.rb",
+      "copyright" => "OpenStreetMap and contributors",
+      "attribution" => "https://www.openstreetmap.org/copyright",
+      "license" => "https://opendatacommons.org/licenses/odbl/1-0/" }
+      .each { |k, v| doc.root[k] = v }

 
 

+    builder = ChangesetBuilder.new(@now, @conn)

     changesets.each do |cs|
     changesets.each do |cs|

-      xml = XML::Node.new("changeset")
-      xml['id'] = cs.id.to_s
-      xml['created_at'] = cs.created_at.getutc.xmlschema
-      xml['closed_at'] = cs.closed_at.getutc.xmlschema if cs.closed?(@now)
-      xml['open'] = cs.open?(@now).to_s
-      xml['num_changes'] = cs.num_changes.to_s
-
-      res = @conn.exec("select u.id, u.display_name, c.min_lat, c.max_lat, c.min_lon, c.max_lon from users u join changesets c on u.id=c.user_id where c.id=#{cs.id}")
-      xml['user'] = res[0]['display_name']
-      xml['uid'] = res[0]['id']
-
-      unless (res[0]['min_lat'].nil? ||
-              res[0]['max_lat'].nil? ||
-              res[0]['min_lon'].nil? ||
-              res[0]['max_lon'].nil?)
-        xml['min_lat'] = (res[0]['min_lat'].to_f / GEO_SCALE).to_s
-        xml['max_lat'] = (res[0]['max_lat'].to_f / GEO_SCALE).to_s
-        xml['min_lon'] = (res[0]['min_lon'].to_f / GEO_SCALE).to_s
-        xml['max_lon'] = (res[0]['max_lon'].to_f / GEO_SCALE).to_s
-      end
+      doc.root << builder.changeset_xml(cs)
+    end

 
 

-      res = @conn.exec("select k, v from changeset_tags where changeset_id=#{cs.id}")
-      res.each do |row|
-        tag = XML::Node.new("tag")
-        tag['k'] = row['k']
-        tag['v'] = row['v']
-        xml << tag
-      end
+    doc.to_s
+  end
+
+  def sequence
+    @state.key?("sequence") ? @state["sequence"] + 1 : 0
+  end

 
 

-      doc.root << xml
+  def data_stem
+    @config["data_dir"] + format("/%03d/%03d/%03d", sequence / 1000000, (sequence / 1000) % 1000, (sequence % 1000))
+  end
+
+  def s3_stem
+    @config["s3_dir"] + format("/%03d/%03d/%03d", sequence / 1000000, (sequence / 1000) % 1000, (sequence % 1000))
+  end
+
+  def write_tmp_files!(changesets)
+    data_file = data_stem + ".osm.gz"
+    tmp_state = @config["state_file"] + ".tmp"
+    tmp_data = data_file + ".tmp"
+
+    FileUtils.mkdir_p(File.dirname(data_file))
+    Zlib::GzipWriter.open(tmp_data) do |fh|
+      fh.write(changeset_dump(changesets))
+    end
+    File.write(tmp_state, YAML.dump(@state))
+
+    # fsync the files in their old locations.
+    fsync(tmp_data)
+    fsync(tmp_state)
+
+    # sync the directory as well, to ensure that the file is reachable
+    # from the dirent and has been updated to account for any allocations.
+    fdirsync(File.dirname(tmp_data))
+    fdirsync(File.dirname(tmp_state))
+
+    # sanity check: the files we're moving into place
+    # should be non-empty.
+    raise "Temporary gzip file should exist, but doesn't." unless File.exist?(tmp_data)
+    raise "Temporary state file should exist, but doesn't." unless File.exist?(tmp_state)
+    raise "Temporary gzip file should be non-empty, but isn't." if File.zero?(tmp_data)
+    raise "Temporary state file should be non-empty, but isn't." if File.zero?(tmp_state)
+  end
+
+  def move_tmp_files_into_place!
+    data_file = data_stem + ".osm.gz"
+    data_state_file = data_stem + ".state.txt"
+    tmp_state = @config["state_file"] + ".tmp"
+    tmp_data = data_file + ".tmp"
+
+    FileUtils.mv(tmp_data, data_file)
+    FileUtils.cp(tmp_state, @config["state_file"])
+    FileUtils.mv(tmp_state, data_state_file)
+
+    # fsync the files in their new locations, in case the inodes have
+    # changed in the move / copy.
+    fsync(data_file)
+    fsync(@config["state_file"])
+    fsync(data_state_file)
+
+    # sync the directory as well, to ensure that the file is reachable
+    # from the dirent and has been updated to account for any allocations.
+    fdirsync(File.dirname(data_file))
+    fdirsync(File.dirname(@config["state_file"]))
+
+    if @config["s3_dir"]
+      s3_file = s3_stem + ".osm.gz"
+      s3_state_file = s3_stem + ".state.txt"
+
+      system("/opt/awscli/v2/current/bin/aws", "--profile=osm-pds-upload", "s3", "cp", "--storage-class=INTELLIGENT_TIERING", "--no-progress", data_file, s3_file)
+      system("/opt/awscli/v2/current/bin/aws", "--profile=osm-pds-upload", "s3", "cp", "--storage-class=INTELLIGENT_TIERING", "--no-progress", data_state_file, s3_state_file)

     end
     end

-    
-    doc.to_s

   end
 
   # saves new state (including the changeset dump xml)
   def save!
   end
 
   # saves new state (including the changeset dump xml)
   def save!

-    File.open(@config['state_file'], "r") do |fl|
-      fl.flock(File::LOCK_EX)
+    File.open(@config["lock_file"], File::RDWR | File::CREAT, 0o600) do |fl|
+      # take the lock in non-blocking mode. if this process doesn't get the lock
+      # then another will be run from cron shortly. this prevents a whole bunch
+      # of processes queueing on the lock and causing weirdness if/when they
+      # get woken up in a random order.
+      got_lock = fl.flock(File::LOCK_EX | File::LOCK_NB)
+      break unless got_lock

 
 

-      sequence = (@state.has_key?('sequence') ? @state['sequence'] + 1 : 0)
-      data_file = @config['data_dir'] + sprintf("/%03d/%03d/%03d.osm.gz", sequence / 1000000, (sequence / 1000) % 1000, (sequence % 1000));
-      tmp_state = @config['state_file'] + ".tmp"
-      tmp_data = "/tmp/changeset_data.osm.tmp"

       # try and write the files to tmp locations and then
       # move them into place later, to avoid in-progress
       # clashes, or people seeing incomplete files.
       begin
       # try and write the files to tmp locations and then
       # move them into place later, to avoid in-progress
       # clashes, or people seeing incomplete files.
       begin

-        FileUtils.mkdir_p(File.dirname(data_file))
-        Zlib::GzipWriter.open(tmp_data) do |fh|
-          fh.write(changeset_dump(open_changesets))
-        end
-        @state['sequence'] = sequence
-        File.open(tmp_state, "w") do |fh|
-          fh.write(YAML.dump(@state))
-        end
-        FileUtils.mv(tmp_data, data_file)
-        FileUtils.mv(tmp_state, @config['state_file'])
-        fl.flock(File::LOCK_UN)
+        changesets = open_changesets
+        @state["sequence"] = sequence
+        @state["last_run"] = @now
+
+        write_tmp_files!(changesets)
+
+        move_tmp_files_into_place!

 
 

-      rescue
-        STDERR.puts("Error! Couldn't update state.")
+        fl.flock(File::LOCK_UN)
+      rescue StandardError
+        warn "Error! Couldn't update state."

         fl.flock(File::LOCK_UN)
         raise
       end
         fl.flock(File::LOCK_UN)
         raise
       end


@@ -148,6 +301,13 @@ class Replicator
   end
 end
 
   end
 end
 

-rep = Replicator.new(ARGV[0])
-rep.save!
-
+begin
+  rep = Replicator.new(ARGV[0])
+  rep.save!
+rescue StandardError => e
+  warn "ERROR: #{e.message}"
+  e.backtrace.each do |frame|
+    warn "ERROR: #{frame}"
+  end
+  exit 1
+end