switchman 3.0.14 → 3.5.20

data/{app/models → lib}/switchman/shard.rb

@@ -1,10 +1,5 @@
 # frozen_string_literal: true
 
-require 'switchman/database_server'
-require 'switchman/default_shard'
-require 'switchman/environment'
-require 'switchman/errors'
-
 module Switchman
   class Shard < UnshardedRecord
     # ten trillion possible ids per shard. yup.
@@ -18,6 +13,47 @@ module Switchman
 
     scope :primary, -> { where(name: nil).order(:database_server_id, :id).distinct_on(:database_server_id) }
 
+    scope :in_region, (lambda do |region, include_regionless: true|
+      next in_current_region if region.nil?
+
+      dbs_by_region = DatabaseServer.group_by(&:region)
+      db_count_in_this_region = dbs_by_region[region]&.length.to_i
+      db_count_in_this_region += dbs_by_region[nil]&.length.to_i if include_regionless
+      non_existent_database_servers = Shard.send(:non_existent_database_servers)
+      db_count_in_other_regions = DatabaseServer.all.length -
+                                  db_count_in_this_region +
+                                  non_existent_database_servers.length
+
+      dbs_in_this_region = dbs_by_region[region]&.map(&:id) || []
+      dbs_in_this_region += dbs_by_region[nil]&.map(&:id) || [] if include_regionless
+
+      if db_count_in_this_region <= db_count_in_other_regions
+        if dbs_in_this_region.include?(Shard.default.database_server.id)
+          where("database_server_id IN (?) OR database_server_id IS NULL", dbs_in_this_region)
+        else
+          where(database_server_id: dbs_in_this_region)
+        end
+      elsif db_count_in_other_regions.zero?
+        all
+      else
+        dbs_not_in_this_region = DatabaseServer.map(&:id) - dbs_in_this_region + non_existent_database_servers
+        if dbs_in_this_region.include?(Shard.default.database_server.id)
+          where("database_server_id NOT IN (?) OR database_server_id IS NULL", dbs_not_in_this_region)
+        else
+          where.not(database_server_id: dbs_not_in_this_region)
+        end
+      end
+    end)
+
+    scope :in_current_region, (lambda do |include_regionless: true|
+      # sharding isn't set up? maybe we're in tests, or a somehow degraded environment
+      # either way there's only one shard, and we always want to see it
+      return [default] unless default.is_a?(Switchman::Shard)
+      return all if !Switchman.region || DatabaseServer.none?(&:region)
+
+      in_region(Switchman.region, include_regionless: include_regionless)
+    end)
+
     class << self
       def sharded_models
         @sharded_models ||= [::ActiveRecord::Base, UnshardedRecord].freeze
@@ -41,16 +77,15 @@ module Switchman
 
           # Now find the actual record, if it exists
           @default = begin
-            find_cached('default_shard') { Shard.where(default: true).take } || default
-          # If we are *super* early in boot, the connection pool won't exist; we don't want to fill in the default shard yet
-          # Otherwise, rescue the fake default if the table doesn't exist
+            find_cached("default_shard") { Shard.where(default: true).take } || default
           rescue
-            sharding_initialized ? default : nil
+            default
           end
-          return default unless @default
 
           # make sure this is not erroneously cached
-          @default.database_server.remove_instance_variable(:@primary_shard) if @default.database_server.instance_variable_defined?(:@primary_shard)
+          if @default.database_server.instance_variable_defined?(:@primary_shard)
+            @default.database_server.remove_instance_variable(:@primary_shard)
+          end
 
           # and finally, check for cached references to the default shard on the existing connection
           sharded_models.each do |klass|
@@ -83,28 +118,30 @@ module Switchman
                       klass.current_switchman_shard != shard
 
           (activated_classes ||= []) << klass
-          klass.connected_to_stack << { shard: shard.database_server.id.to_sym, klasses: [klass], switchman_shard: shard }
+          klass.connected_to_stack << { shard: shard.database_server.id.to_sym,
+                                        klasses: [klass],
+                                        switchman_shard: shard }
         end
         activated_classes
       end
 
       def active_shards
-        sharded_models.map do |klass|
+        sharded_models.filter_map do |klass|
           [klass, current(klass)]
-        end.compact.to_h
+        end.to_h
       end
 
       def lookup(id)
         id_i = id.to_i
-        return current if id_i == current.id || id == 'self'
-        return default if id_i == default.id || id.nil? || id == 'default'
+        return current if id_i == current.id || id == "self"
+        return default if id_i == default.id || id.nil? || id == "default"
 
         id = id_i
         raise ArgumentError if id.zero?
 
         unless cached_shards.key?(id)
           cached_shards[id] = Shard.default.activate do
-            find_cached(['shard', id]) { find_by(id: id) }
+            find_cached(["shard", id]) { find_by(id: id) }
           end
         end
         cached_shards[id]
@@ -125,11 +162,11 @@ module Switchman
       # * +classes+ - an array of classes to activate
       #    parallel: - true/false to execute in parallel, or an integer of how many
       #                sub-processes. Note that parallel invocation currently uses
-      #                forking, so should be used sparingly because you cannot get
-      #                results back
+      #                forking.
       #    exception: - :ignore, :raise, :defer (wait until the end and raise the first
       #                error), or a proc
-      def with_each_shard(*args, parallel: false, exception: :raise, &block)
+      #    output: - :simple, :decorated (with database_server_id:shard_name)
+      def with_each_shard(*args, parallel: false, exception: :raise, output: :simple)
         raise ArgumentError, "wrong number of arguments (#{args.length} for 0...2)" if args.length > 2
 
         return Array.wrap(yield) unless default.is_a?(Shard)
@@ -148,160 +185,84 @@ module Switchman
         parallel = 0 if parallel == false || parallel.nil?
 
         scope ||= Shard.all
-        scope = scope.order(::Arel.sql('database_server_id IS NOT NULL, database_server_id, id')) if ::ActiveRecord::Relation === scope && scope.order_values.empty?
+        if ::ActiveRecord::Relation === scope && scope.order_values.empty?
+          scope = scope.order(::Arel.sql("database_server_id IS NOT NULL, database_server_id, id"))
+        end
 
         if parallel > 1
           if ::ActiveRecord::Relation === scope
             # still need a post-uniq, cause the default database server could be NULL or Rails.env in the db
-            database_servers = scope.reorder('database_server_id').select(:database_server_id).distinct.
-                               map(&:database_server).compact.uniq
+            database_servers = scope.reorder("database_server_id").select(:database_server_id).distinct
+                                    .filter_map(&:database_server).uniq
             # nothing to do
             return if database_servers.count.zero?
 
             scopes = database_servers.to_h do |server|
-              [server, server.shards.merge(scope)]
+              [server, scope.merge(server.shards)]
             end
           else
             scopes = scope.group_by(&:database_server)
           end
 
-          exception_pipes = []
-          pids = []
-          out_fds = []
-          err_fds = []
-          pid_to_name_map = {}
-          fd_to_name_map = {}
-          errors = []
-
-          wait_for_output = lambda do
-            ready, = IO.select(out_fds + err_fds)
-            ready.each do |fd|
-              if fd.eof?
-                fd.close
-                out_fds.delete(fd)
-                err_fds.delete(fd)
-                next
-              end
-              line = fd.readline
-              puts "#{fd_to_name_map[fd]}: #{line}"
-            end
-          end
-
-          # only one process; don't bother forking
-          return with_each_shard(scopes.first.last, classes, exception: exception, &block) if scopes.length == 1
-
           # clear connections prior to forking (no more queries will be executed in the parent,
           # and we want them gone so that we don't accidentally use them post-fork doing something
           # silly like dealloc'ing prepared statements)
-          ::ActiveRecord::Base.clear_all_connections!
+          if ::Rails.version < "7.1"
+            ::ActiveRecord::Base.clear_all_connections!(nil)
+          else
+            ::ActiveRecord::Base.connection_handler.clear_all_connections!(:all)
+          end
 
-          scopes.each do |server, subscope|
+          parent_process_name = sanitized_process_title
+          ret = ::Parallel.map(scopes, in_processes: (scopes.length > 1) ? parallel : 0) do |server, subscope|
             name = server.id
+            last_description = name
 
-            exception_pipe = IO.pipe
-            exception_pipes << exception_pipe
-            pid, io_in, io_out, io_err = Open4.pfork4(lambda do
-              Switchman.config[:on_fork_proc]&.call
-
-              # set a pretty name for the process title, up to 128 characters
-              # (we don't actually know the limit, depending on how the process
-              # was started)
-              # first, simplify the binary name by stripping directories,
-              # then truncate arguments as necessary
-              bin = File.basename($0) # Process.argv0 doesn't work on Ruby 2.5 (https://bugs.ruby-lang.org/issues/15887)
-              max_length = 128 - bin.length - name.length - 3
-              args = ARGV.join(' ')
-              args = args[0..max_length] if max_length >= 0
-              new_title = [bin, args, name].join(' ')
+            begin
+              max_length = 128 - name.length - 3
+              short_parent_name = parent_process_name[0..max_length] if max_length >= 0
+              new_title = [short_parent_name, name].join(" ")
               Process.setproctitle(new_title)
-
-              with_each_shard(subscope, classes, exception: exception, &block)
-              exception_pipe.last.close
-            rescue Exception => e # rubocop:disable Lint/RescueException
-              begin
-                dumped = Marshal.dump(e)
-                dumped = nil if dumped.length > 64 * 1024
-              rescue
-                dumped = nil
-              end
-
-              if dumped.nil?
-                # couldn't dump the exception; create a copy with just
-                # the message and the backtrace
-                e2 = e.class.new(e.message)
-                backtrace = e.backtrace
-                # truncate excessively long backtraces
-                backtrace = backtrace[0...25] + ['...'] + backtrace[-25..] if backtrace.length > 50
-                e2.set_backtrace(backtrace)
-                e2.instance_variable_set(:@active_shards, e.instance_variable_get(:@active_shards))
-                dumped = Marshal.dump(e2)
-              end
-              exception_pipe.last.set_encoding(dumped.encoding)
-              exception_pipe.last.write(dumped)
-              exception_pipe.last.flush
-              exception_pipe.last.close
-              exit! 1
-            end)
-            exception_pipe.last.close
-            pids << pid
-            io_in.close # don't care about writing to stdin
-            out_fds << io_out
-            err_fds << io_err
-            pid_to_name_map[pid] = name
-            fd_to_name_map[io_out] = name
-            fd_to_name_map[io_err] = name
-
-            while pids.count >= parallel
-              while out_fds.count >= parallel
-                # wait for output if we've hit the parallel limit
-                wait_for_output.call
+              Switchman.config[:on_fork_proc]&.call
+              with_each_shard(subscope, classes, exception: exception, output: :decorated) do
+                last_description = Shard.current.description
+                Parallel::ResultWrapper.new(yield)
               end
-              # we've gotten all the output from one fd so wait for its child process to exit
-              found_pid, status = Process.wait2
-              pids.delete(found_pid)
-              errors << pid_to_name_map[found_pid] if status.exitstatus != 0
+            rescue => e
+              logger.error e.full_message
+              Parallel::QuietExceptionWrapper.new(last_description, ::Parallel::ExceptionWrapper.new(e))
             end
-            # we've gotten all the output from one fd so wait for its child process to exit
-            found_pid, status = Process.wait2
-            pids.delete(found_pid)
-            errors << pid_to_name_map[found_pid] if status.exitstatus != 0
-          end
-
-          wait_for_output.call while out_fds.any? || err_fds.any?
-          pids.each do |pid|
-            _, status = Process.waitpid2(pid)
-            errors << pid_to_name_map[pid] if status.exitstatus != 0
-          end
-
-          # check for an exception; we only re-raise the first one
-          exception_pipes.each do |exception_pipe|
-            serialized_exception = exception_pipe.first.read
-            next if serialized_exception.empty?
-
-            ex = Marshal.load(serialized_exception) # rubocop:disable Security/MarshalLoad
-            raise ex
-          ensure
-            exception_pipe.first.close
-          end
+          end.flatten
 
+          errors = ret.select { |val| val.is_a?(Parallel::QuietExceptionWrapper) }
           unless errors.empty?
-            raise ParallelShardExecError,
-                  "The following subprocesses did not exit cleanly: #{errors.sort.join(', ')}"
+            raise errors.first.exception if errors.length == 1
+
+            errors_desc = errors.map(&:name).sort.join(", ")
+            raise Errors::ParallelShardExecError,
+                  "The following database server(s) did not finish processing cleanly: #{errors_desc}",
+                  cause: errors.first.exception
           end
 
-          return
+          return ret.map(&:result)
         end
 
         classes ||= []
 
-        previous_shard = nil
         result = []
         ex = nil
+        old_stdout = $stdout
+        old_stderr = $stderr
         scope.each do |shard|
           # shard references a database server that isn't configured in this environment
           next unless shard.database_server
 
           shard.activate(*classes) do
+            if output == :decorated
+              $stdout = Parallel::PrefixingIO.new(shard.description, $stdout)
+              $stderr = Parallel::PrefixingIO.new(shard.description, $stderr)
+            end
+
             result.concat Array.wrap(yield)
           rescue
             case exception
@@ -315,8 +276,10 @@ module Switchman
             else
               raise
             end
+          ensure
+            $stdout = old_stdout
+            $stderr = old_stderr
           end
-          previous_shard = shard
         end
         raise ex if ex
 
@@ -340,7 +303,7 @@ module Switchman
             else
               shard = partition_object.shard
             end
-          when Integer, /^\d+$/, /^(\d+)~(\d+)$/
+          when Integer, /\A\d+\Z/, /\A(\d+)~(\d+)\Z/
             local_id, shard = Shard.local_id_for(partition_object)
             local_id ||= partition_object
             object = local_id unless partition_proc
@@ -382,14 +345,14 @@ module Switchman
         case any_id
         when ::ActiveRecord::Base
           any_id.id
-        when /^(\d+)~(-?\d+)$/
+        when /\A(\d+)~(-?\d+)\Z/
           local_id = $2.to_i
           signed_id_operation(local_id) do |id|
             return nil if id > IDS_PER_SHARD
 
             ($1.to_i * IDS_PER_SHARD) + id
           end
-        when Integer, /^-?\d+$/
+        when Integer, /\A-?\d+\Z/
           any_id.to_i
         end
       end
@@ -467,17 +430,13 @@ module Switchman
 
       private
 
-      def sharding_initialized
-        @sharding_initialized ||= false
-      end
-
       def add_sharded_model(klass)
         @sharded_models = (sharded_models + [klass]).freeze
-        initialize_sharding
+        configure_connects_to
       end
 
-      def initialize_sharding
-        full_connects_to_hash = DatabaseServer.all.to_h { |db| [db.id.to_sym, db.connects_to_hash] }
+      def configure_connects_to
+        full_connects_to_hash = DatabaseServer.to_h { |db| [db.id.to_sym, db.connects_to_hash] }
         sharded_models.each do |klass|
           connects_to_hash = full_connects_to_hash.deep_dup
           if klass == UnshardedRecord
@@ -496,10 +455,11 @@ module Switchman
             end
           end
 
+          # this resets the default shard on rails 7.1+, but we want to preserve it
+          shard_was = klass.default_shard
           klass.connects_to shards: connects_to_hash
+          klass.default_shard = shard_was
         end
-
-        @sharding_initialized = true
       end
 
       # in-process caching
@@ -532,8 +492,50 @@ module Switchman
         shard = nil unless shard.database_server
         shard
       end
+
+      # Determines the name of the current process, including arguments, but stripping
+      # any shebang from the invoked script, and any additional path info from the
+      # executable.
+      #
+      # @return [String]
+      def sanitized_process_title
+        # get the effective process name from `ps`; this will include any changes
+        # from Process.setproctitle _or_ assigning to $0.
+        parent_process_name = `ps -ocommand= -p#{Process.pid}`.strip
+        # Effective process titles may be shorter than the actual
+        # command; truncate our ARGV[0] so that they are comparable
+        # for the next step
+        argv0 = if parent_process_name.length < Process.argv0.length
+                  Process.argv0[0..parent_process_name.length]
+                else
+                  Process.argv0
+                end
+
+        # when running via a shebang, the `ps` output will include the shebang
+        # (i.e. it will be "ruby bin/rails c"); attempt to strip it off.
+        # Note that argv0 in this case will _only_ be `bin/rails` (no shebang,
+        # no arguments). We want to preserve the arguments we got from `ps`
+        if (index = parent_process_name.index(argv0))
+          parent_process_name.slice!(0...index)
+        end
+
+        # remove directories from the main executable to make more room
+        # for additional info
+        argv = parent_process_name.shellsplit
+        argv[0] = File.basename(argv[0])
+        argv.shelljoin
+      end
+
+      # @return [Array<String>] the list of database servers that are in the
+      #   config, but don't have any shards on them
+      def non_existent_database_servers
+        @non_existent_database_servers ||=
+          Shard.distinct.pluck(:database_server_id).compact - DatabaseServer.all.map(&:id)
+      end
     end
 
+    delegate :region, :in_region?, :in_current_region?, to: :database_server
+
     def name
       unless instance_variable_defined?(:@name)
         # protect against re-entrancy
@@ -562,7 +564,7 @@ module Switchman
     end
 
     def description
-      [database_server.id, name].compact.join(':')
+      [database_server.id, name].compact.join(":")
     end
 
     # Shards are always on the default shard
@@ -574,6 +576,10 @@ module Switchman
       id
     end
 
+    def original_id_value
+      id
+    end
+
     def activate(*classes, &block)
       shards = hashify_classes(classes)
       Shard.activate(shards, &block)
@@ -596,48 +602,39 @@ module Switchman
     end
 
     def drop_database
-      raise('Cannot drop the database of the default shard') if default?
+      raise "Cannot drop the database of the default shard" if default?
       return unless read_attribute(:name)
 
       begin
-        adapter = database_server.config[:adapter]
-        sharding_config = Switchman.config || {}
-        drop_statement = sharding_config[adapter]&.[](:drop_statement)
-        drop_statement ||= sharding_config[:drop_statement]
-        if drop_statement
-          drop_statement = Array(drop_statement).dup.
-                           map { |statement| statement.gsub('%{name}', name) }
+        activate do
+          self.class.drop_database(name)
         end
+      rescue ::ActiveRecord::StatementInvalid => e
+        logger.error "Drop failed: #{e}"
+      end
+    end
 
-        case adapter
-        when 'mysql', 'mysql2'
-          activate do
-            ::GuardRail.activate(:deploy) do
-              drop_statement ||= "DROP DATABASE #{name}"
-              Array(drop_statement).each do |stmt|
-                ::ActiveRecord::Base.connection.execute(stmt)
-              end
-            end
-          end
-        when 'postgresql'
-          activate do
-            ::GuardRail.activate(:deploy) do
-              # Shut up, Postgres!
-              conn = ::ActiveRecord::Base.connection
-              old_proc = conn.raw_connection.set_notice_processor {}
-              begin
-                drop_statement ||= "DROP SCHEMA #{name} CASCADE"
-                Array(drop_statement).each do |stmt|
-                  ::ActiveRecord::Base.connection.execute(stmt)
-                end
-              ensure
-                conn.raw_connection.set_notice_processor(&old_proc) if old_proc
-              end
-            end
+    #
+    # Drops a specific database/schema from the currently active connection
+    #
+    def self.drop_database(name)
+      sharding_config = Switchman.config || {}
+      drop_statement = sharding_config["postgresql"]&.[](:drop_statement)
+      drop_statement ||= sharding_config[:drop_statement]
+      drop_statement = Array(drop_statement).map { |statement| statement.gsub("%{name}", name) } if drop_statement
+
+      ::GuardRail.activate(:deploy) do
+        # Shut up, Postgres!
+        conn = ::ActiveRecord::Base.connection
+        old_proc = conn.raw_connection.set_notice_processor {}
+        begin
+          drop_statement ||= "DROP SCHEMA #{name} CASCADE"
+          Array(drop_statement).each do |stmt|
+            ::ActiveRecord::Base.connection.execute(stmt)
           end
+        ensure
+          conn.raw_connection.set_notice_processor(&old_proc) if old_proc
         end
-      rescue
-        logger.info "Drop failed: #{$!}"
       end
     end
 
@@ -656,7 +653,7 @@ module Switchman
     end
 
     def destroy
-      raise('Cannot destroy the default shard') if default?
+      raise("Cannot destroy the default shard") if default?
 
       super
     end
@@ -665,8 +662,8 @@ module Switchman
 
     def clear_cache
       Shard.default.activate do
-        Switchman.cache.delete(['shard', id].join('/'))
-        Switchman.cache.delete('default_shard') if default?
+        Switchman.cache.delete(["shard", id].join("/"))
+        Switchman.cache.delete("default_shard") if default?
       end
       self.class.clear_cache
     end

data/lib/switchman/sharded_instrumenter.rb

@@ -16,7 +16,11 @@ module Switchman
         payload[:shard] = {
           database_server_id: shard.database_server.id,
           id: shard.id,
-          env: shard.database_server.guard_rail_environment
+          env: if ::Rails.version < "7.0"
+                 @shard_host.pool.connection_klass&.current_role
+               else
+                 @shard_host.pool.connection_class&.current_role
+               end
         }
       end
       super name, payload

data/lib/switchman/shared_schema_cache.rb

@@ -0,0 +1,11 @@
+# frozen_string_literal: true
+
+module Switchman
+  class SharedSchemaCache
+    def self.get_schema_cache(connection)
+      @schema_cache ||= ::ActiveRecord::ConnectionAdapters::SchemaCache.new(connection)
+      @schema_cache.connection = connection
+      @schema_cache
+    end
+  end
+end

data/lib/switchman/standard_error.rb

@@ -6,16 +6,17 @@ module Switchman
       super
       # These seem to get themselves into a bad state if we try to lookup shards while processing
       return if is_a?(IO::EAGAINWaitReadable)
+
       return if Thread.current[:switchman_error_handler]
 
       begin
         Thread.current[:switchman_error_handler] = true
 
-        begin
-          @active_shards = Shard.active_shards if defined?(Shard)
-        rescue
-          # If we hit an error really early in boot, activerecord may not be initialized yet
-        end
+        @active_shards ||= Shard.active_shards
+      rescue
+        # intentionally empty - don't allow calculating the active_shards to prevent
+        # creating the StandardError for any reason. this prevents various random issues
+        # when a StandardError is created within a finalizer
       ensure
         Thread.current[:switchman_error_handler] = nil
       end

data/lib/switchman/test_helper.rb

@@ -19,7 +19,7 @@ module Switchman
         end
 
         server1 = Shard.default.database_server
-        server2 = DatabaseServer.create(Shard.default.database_server.config.merge(server2: true))
+        server2 = DatabaseServer.create(Shard.default.database_server.config.merge(server2: true, schema_dump: false))
 
         if server1 == Shard.default.database_server && server1.config[:shard1] && server1.config[:shard2]
           # look for the shards in the db already
@@ -65,7 +65,7 @@ module Switchman
         if server == Shard.default.database_server
           server.shards.where(name: name).first
         else
-          shard = Shard.where('database_server_id IS NOT NULL AND name=?', name).first
+          shard = Shard.where("database_server_id IS NOT NULL AND name=?", name).first
           # if somehow databases got created in a different order, change the shard to match
           shard.database_server = server if shard
           shard

data/{app/models → lib}/switchman/unsharded_record.rb

@@ -2,6 +2,6 @@
 
 module Switchman
   class UnshardedRecord < ::ActiveRecord::Base
-    sharded_model
+    self.abstract_class = true
   end
 end

data/lib/switchman/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Switchman
-  VERSION = '3.0.14'
+  VERSION = "3.5.20"
 end