# frozen_string_literal: true # Used for $ERROR_INFO. This *must* be capitalized! require 'English' require 'json' require 'logging' require 'set' require 'bolt/analytics' require 'bolt/result' require 'bolt/config' require 'bolt/notifier' require 'bolt/result_set' require 'bolt/puppetdb' module Bolt class Executor attr_reader :noop, :transports attr_accessor :run_as # FIXME: There must be a better way # https://makandracards.com/makandra/36011-ruby-do-not-mix-optional-and-keyword-arguments def initialize(concurrency = 1, analytics = Bolt::Analytics::NoopClient.new, noop = false) # lazy-load expensive gem code require 'concurrent' @analytics = analytics @logger = Logging.logger[self] @transports = Bolt::TRANSPORTS.each_with_object({}) do |(key, val), coll| coll[key.to_s] = if key == :remote Concurrent::Delay.new do val.new(self) end else Concurrent::Delay.new do val.new end end end @reported_transports = Set.new @noop = noop @run_as = nil @pool = if concurrency > 0 Concurrent::ThreadPoolExecutor.new(max_threads: concurrency) else Concurrent.global_immediate_executor end @logger.debug { "Started with #{concurrency} max thread(s)" } @notifier = Bolt::Notifier.new end def transport(transport) impl = @transports[transport || 'ssh'] raise(Bolt::UnknownTransportError, transport) unless impl # If there was an error creating the transport, ensure it gets thrown impl.no_error! impl.value end # Starts executing the given block on a list of nodes in parallel, one thread per "batch". # # This is the main driver of execution on a list of targets. It first # groups targets by transport, then divides each group into batches as # defined by the transport. Yields each batch, along with the corresponding # transport, to the block in turn and returns an array of result promises. def queue_execute(targets) targets.group_by(&:transport).flat_map do |protocol, protocol_targets| transport = transport(protocol) report_transport(transport, protocol_targets.count) transport.batches(protocol_targets).flat_map do |batch| batch_promises = Array(batch).each_with_object({}) do |target, h| h[target] = Concurrent::Promise.new(executor: :immediate) end # Pass this argument through to avoid retaining a reference to a # local variable that will change on the next iteration of the loop. @pool.post(batch_promises) do |result_promises| begin results = yield transport, batch Array(results).each do |result| result_promises[result.target].set(result) end # NotImplementedError can be thrown if the transport is not implemented improperly rescue StandardError, NotImplementedError => e result_promises.each do |target, promise| # If an exception happens while running, the result won't be logged # by the CLI. Log a warning, as this is probably a problem with the transport. # If batch_* commands are used from the Base transport, then exceptions # normally shouldn't reach here. @logger.warn(e) promise.set(Bolt::Result.from_exception(target, e)) end ensure # Make absolutely sure every promise gets a result to avoid a # deadlock. Use whatever exception is causing this block to # execute, or generate one if we somehow got here without an # exception and some promise is still missing a result. result_promises.each do |target, promise| next if promise.fulfilled? error = $ERROR_INFO || Bolt::Error.new("No result was returned for #{target.uri}", "puppetlabs.bolt/missing-result-error") promise.set(Bolt::Result.from_exception(target, error)) end end end batch_promises.values end end end # Create a ResultSet from the results of all promises. def await_results(promises) ResultSet.new(promises.map(&:value)) end # Execute the given block on a list of nodes in parallel, one thread per "batch". # # This is the main driver of execution on a list of targets. It first # groups targets by transport, then divides each group into batches as # defined by the transport. Each batch, along with the corresponding # transport, is yielded to the block in turn and the results all collected # into a single ResultSet. def batch_execute(targets, &block) promises = queue_execute(targets, &block) await_results(promises) end def log_action(description, targets) # When running a plan, info messages like starting a task are promoted to notice. log_method = @plan_logging ? :notice : :info target_str = if targets.length > 5 "#{targets.count} targets" else targets.map(&:uri).join(', ') end @logger.send(log_method, "Starting: #{description} on #{target_str}") start_time = Time.now results = yield duration = Time.now - start_time failures = results.error_set.length plural = failures == 1 ? '' : 's' @logger.send(log_method, "Finished: #{description} with #{failures} failure#{plural} in #{duration.round(2)} sec") results end def log_plan(plan_name) log_method = @plan_logging ? :notice : :info @logger.send(log_method, "Starting: plan #{plan_name}") start_time = Time.now results = nil begin results = yield ensure duration = Time.now - start_time @logger.send(log_method, "Finished: plan #{plan_name} in #{duration.round(2)} sec") end results end def report_transport(transport, count) name = transport.class.name.split('::').last.downcase unless @reported_transports.include?(name) @analytics&.event('Transport', 'initialize', label: name, value: count) end @reported_transports.add(name) end def report_function_call(function) @analytics&.event('Plan', 'call_function', label: function) end def report_bundled_content(mode, name) if @analytics.bundled_content&.include?(name) @analytics&.event('Bundled Content', mode, label: name) end end def report_apply(statement_count, resource_counts) data = { statement_count: statement_count } unless resource_counts.empty? sum = resource_counts.inject(0) { |accum, i| accum + i } # Intentionally rounded to an integer. High precision isn't useful. data[:resource_mean] = sum / resource_counts.length end @analytics&.event('Apply', 'ast', data) end def report_yaml_plan(plan) steps = plan.steps.count return_type = case plan.return when Bolt::PAL::YamlPlan::EvaluableString 'expression' when nil nil else 'value' end @analytics&.event('Plan', 'yaml', plan_steps: steps, return_type: return_type) rescue StandardError => e @logger.debug { "Failed to submit analytics event: #{e.message}" } end def with_node_logging(description, batch) @logger.info("#{description} on #{batch.map(&:uri)}") result = yield @logger.info(result.to_json) result end def run_command(targets, command, options = {}, &callback) description = options.fetch('_description', "command '#{command}'") log_action(description, targets) do notify = proc { |event| @notifier.notify(callback, event) if callback } options = { '_run_as' => run_as }.merge(options) if run_as results = batch_execute(targets) do |transport, batch| with_node_logging("Running command '#{command}'", batch) do transport.batch_command(batch, command, options, ¬ify) end end @notifier.shutdown results end end def run_script(targets, script, arguments, options = {}, &callback) description = options.fetch('_description', "script #{script}") log_action(description, targets) do notify = proc { |event| @notifier.notify(callback, event) if callback } options = { '_run_as' => run_as }.merge(options) if run_as results = batch_execute(targets) do |transport, batch| with_node_logging("Running script #{script} with '#{arguments}'", batch) do transport.batch_script(batch, script, arguments, options, ¬ify) end end @notifier.shutdown results end end def run_task(targets, task, arguments, options = {}, &callback) description = options.fetch('_description', "task #{task.name}") log_action(description, targets) do notify = proc { |event| @notifier.notify(callback, event) if callback } options = { '_run_as' => run_as }.merge(options) if run_as arguments['_task'] = task.name results = batch_execute(targets) do |transport, batch| with_node_logging("Running task #{task.name} with '#{arguments}'", batch) do transport.batch_task(batch, task, arguments, options, ¬ify) end end @notifier.shutdown results end end def upload_file(targets, source, destination, options = {}, &callback) description = options.fetch('_description', "file upload from #{source} to #{destination}") log_action(description, targets) do notify = proc { |event| @notifier.notify(callback, event) if callback } options = { '_run_as' => run_as }.merge(options) if run_as results = batch_execute(targets) do |transport, batch| with_node_logging("Uploading file #{source} to #{destination}", batch) do transport.batch_upload(batch, source, destination, options, ¬ify) end end @notifier.shutdown results end end class TimeoutError < RuntimeError; end def wait_until_available(targets, description: 'wait until available', wait_time: 120, retry_interval: 1) log_action(description, targets) do batch_execute(targets) do |transport, batch| with_node_logging('Waiting until available', batch) do begin wait_until(wait_time, retry_interval) { transport.batch_connected?(batch) } batch.map { |target| Result.new(target) } rescue TimeoutError => e batch.map { |target| Result.from_exception(target, e) } end end end end end # Used to simplify unit testing, to avoid having to mock other calls to Time.now. private def wait_now Time.now end def wait_until(timeout, retry_interval) start = wait_now until yield raise(TimeoutError, 'Timed out waiting for target') if (wait_now - start).to_i >= timeout sleep(retry_interval) end end # Plan context doesn't make sense for most transports but it is tightly # coupled with the orchestrator transport since the transport behaves # differently when a plan is running. In order to limit how much this # pollutes the transport API we only handle the orchestrator transport here. # Since we callt this function without resolving targets this will result # in the orchestrator transport always being initialized during plan runs. # For now that's ok. # # In the future if other transports need this or if we want a plan stack # we'll need to refactor. def start_plan(plan_context) transport('pcp').plan_context = plan_context @plan_logging = true end def finish_plan(plan_result) transport('pcp').finish_plan(plan_result) end def without_default_logging old_log = @plan_logging @plan_logging = false yield ensure @plan_logging = old_log end end end