openbolt 5.6.0 → 9.0.0

data/lib/bolt/config/transport/choria.rb

@@ -12,9 +12,11 @@ module Bolt
           collective
           command-timeout
           config-file
+          file-transfer-timeout
           host
           interpreters
           mcollective-certname
+          message-data-size
           broker-timeout
           brokers
           puppet-environment
@@ -25,6 +27,7 @@ module Bolt
           task-agent
           task-timeout
           tmpdir
+          upload-batch-size
         ].sort.freeze
 
         DEFAULTS = {

data/lib/bolt/config/transport/options.rb

@@ -19,8 +19,8 @@ module Bolt
           "batch-mode" => {
             type: [TrueClass, FalseClass],
             description: "Whether to disable password querying. When set to `false`, SSH will fall back to " \
-                         "prompting for a password if key authentication fails. This might cause Bolt to hang. " \
-                         "To prevent Bolt from hanging, you can configure `ssh-command` to use an SSH utility " \
+                         "prompting for a password if key authentication fails. This might cause OpenBolt to hang. " \
+                         "To prevent OpenBolt from hanging, you can configure `ssh-command` to use an SSH utility " \
                          "such as sshpass that supports providing a password non-interactively. For more " \
                          "information, see [Providing a password non-interactively using " \
                          "`native-ssh`](troubleshooting.md#providing-a-password-non-interactively-using-native-ssh).",
@@ -29,7 +29,7 @@ module Bolt
             _example: false
           },
           "bundled-ruby" => {
-            description: "Whether to use the Ruby bundled with Bolt packages for local targets.",
+            description: "Whether to use the Ruby bundled with OpenBolt packages for local targets.",
             type: [TrueClass, FalseClass],
             _plugin: false,
             _example: true,
@@ -44,7 +44,7 @@ module Bolt
           "cleanup" => {
             type: [TrueClass, FalseClass],
             description: "Whether to clean up temporary files created on targets. When running commands on a target, " \
-                         "Bolt might create temporary files. After completing the command, these files are " \
+                         "OpenBolt might create temporary files. After completing the command, these files are " \
                          "automatically deleted. This value can be set to 'false' if you wish to leave these " \
                          "temporary files on the target.",
             _plugin: true,
@@ -101,10 +101,33 @@ module Bolt
             _default: 30,
             _example: 60
           },
+          "file-transfer-timeout" => {
+            type: Integer,
+            description: "How long to wait in seconds for targets to respond to a file transfer request. Default is " \
+                         "300, and the wait is never shorter than the agent's DDL timeout of 120 seconds.",
+            minimum: 1,
+            _plugin: true,
+            _example: 600
+          },
+          "message-data-size" => {
+            type: Integer,
+            description: "The file content in one Choria RPC message, in kibibytes, for uploads, downloads and " \
+                         "directory listings. Default is 512, half the broker's 1 MiB message limit.",
+            minimum: 1,
+            _plugin: true,
+            _example: 256
+          },
+          "upload-batch-size" => {
+            type: Integer,
+            description: "How many targets one chunk request is published to at a time. Default is 256.",
+            minimum: 1,
+            _plugin: true,
+            _example: 100
+          },
           "connect-timeout" => {
             type: Integer,
             description: "How long to wait in seconds when establishing connections. Set this value higher if you " \
-                         "frequently encounter connection timeout errors when running Bolt.",
+                         "frequently encounter connection timeout errors when running OpenBolt.",
             minimum: 1,
             _plugin: true,
             _default: 10,
@@ -112,7 +135,7 @@ module Bolt
           },
           "copy-command" => {
             type: [Array, String],
-            description: "The command to use when copying files using native SSH. Bolt runs `<copy-command> <src> " \
+            description: "The command to use when copying files using native SSH. OpenBolt runs `<copy-command> <src> " \
                          "<dest>`. This option is used when you need support for features or algorithms that are not " \
                          "supported by the net-ssh Ruby library. **This option is experimental.** You can read more " \
                          "about this option in [Native SSH transport](experimental_features.md#native-ssh-transport).",
@@ -203,7 +226,7 @@ module Bolt
                          "override the shebang defined in a task executable. The extension can optionally be " \
                          "specified with the `.` character (`.py` and `py` both map to a task executable " \
                          "`task.py`) and the extension is case sensitive. When a target's name is `localhost`, " \
-                         "Ruby tasks run with the Bolt Ruby interpreter by default.",
+                         "Ruby tasks run with the OpenBolt Ruby interpreter by default.",
             additionalProperties: {
               type: [String, Array],
               _plugin: false
@@ -215,20 +238,6 @@ module Bolt
             _example: { "rb" => ["/usr/bin/ruby", "-r", "puppet"],
                         ".py" => "/usr/bin/python3" }
           },
-          "job-poll-interval" => {
-            type: Integer,
-            description: "The interval, in seconds, to poll orchestrator for job status.",
-            minimum: 1,
-            _plugin: true,
-            _example: 2
-          },
-          "job-poll-timeout" => {
-            type: Integer,
-            description: "The time, in seconds, to wait for orchestrator job status.",
-            minimum: 1,
-            _plugin: true,
-            _example: 2000
-          },
           "kex-algorithms" => {
             type: Array,
             description: "A list of key exchange algorithms to use when establishing a connection " \
@@ -253,7 +262,7 @@ module Bolt
           },
           "login-shell" => {
             type: String,
-            description: "Which login shell Bolt should expect on the target. Supported shells are " \
+            description: "Which login shell OpenBolt should expect on the target. Supported shells are " \
                          "#{LOGIN_SHELLS.join(', ')}. **This option is experimental.**",
             enum: LOGIN_SHELLS,
             _plugin: true,
@@ -346,13 +355,6 @@ module Bolt
             _default: "production",
             _example: "staging"
           },
-          "read-timeout" => {
-            type: Integer,
-            description: "How long to wait in seconds when making requests to the Orchestrator.",
-            minimum: 1,
-            _plugin: true,
-            _example: 15
-          },
           "realm" => {
             type: String,
             description: "The Kerberos realm (Active Directory domain) to authenticate against.",
@@ -375,9 +377,9 @@ module Bolt
           },
           "run-as-command" => {
             type: Array,
-            description: "The command to elevate permissions. Bolt appends the user and command strings to the " \
-                         "configured `run-as-command` before running it on the target. This command must not require  " \
-                         "aninteractive password prompt, and the `sudo-password` option is ignored when " \
+            description: "The command to elevate permissions. OpenBolt appends the user and command strings to the " \
+                         "configured `run-as-command` before running it on the target. This command must not require " \
+                         "an interactive password prompt, and the `sudo-password` option is ignored when " \
                          "`run-as-command` is specified. The `run-as-command` must be specified as an array.",
             items: {
               type: String
@@ -486,13 +488,6 @@ module Bolt
             _plugin: true,
             _example: "p@$$w0rd!"
           },
-          "task-environment" => {
-            type: String,
-            description: "The environment the orchestrator loads task code from.",
-            _plugin: true,
-            _default: "production",
-            _example: "development"
-          },
           "task-timeout" => {
             type: Integer,
             description: "How long to wait in seconds for tasks to complete when using the Choria transport.",
@@ -507,12 +502,6 @@ module Bolt
             _plugin: true,
             _example: "/tmp/bolt"
           },
-          "token-file" => {
-            type: String,
-            description: "The path to the token file.",
-            _plugin: true,
-            _example: "~/.puppetlabs/puppet/token.pem"
-          },
           "tty" => {
             type: [TrueClass, FalseClass],
             description: "Whether to enable tty on exec commands.",

data/lib/bolt/executor.rb

@@ -401,6 +401,10 @@ module Bolt
       @fiber_executor.get_current_plan_id(fiber: fiber)
     end
 
+    def run_in_new_fiber(&block)
+      @fiber_executor.run_in_new_fiber(&block)
+    end
+
     def plan_complete?
       @fiber_executor.plan_complete?
     end

data/lib/bolt/fiber_executor.rb

@@ -5,6 +5,11 @@ require_relative '../bolt/plan_future'
 
 module Bolt
   class FiberExecutor
+    # Limits how many plans one future can nest, so a plan that runs itself
+    # with no condition that stops it fails instead of starting Fibers until
+    # memory runs out.
+    MAX_NESTED_PLANS = 1000
+
     attr_reader :active_futures, :finished_futures
 
     def initialize
@@ -99,6 +104,7 @@ module Bolt
       # `wait` call, Bolt should pause for long enough that something can
       # execute before checking again. This mitigates CPU
       # thrashing.
+      return if active_futures.empty?
       return unless active_futures.all? { |f| %i[returned_immediately unfinished].include?(f.value) }
 
       @logger.trace("Nothing can be resumed. Rechecking in 0.5 seconds.")
@@ -121,7 +127,45 @@ module Bolt
     # Get the PlanFuture object that is currently executing
     #
     def get_current_future(fiber:)
-      all_futures.select { |f| f.fiber == fiber }.first
+      all_futures.find { |f| f.fiber == fiber || f.nested_fibers.include?(fiber) }
+    end
+
+    # Run the block on a new Fiber, so it starts with an empty stack instead
+    # of adding to the current one. Every plan a plan runs adds dozens of
+    # frames, and a Fiber's stack is small and fixed. Values the block yields
+    # while it waits are yielded on to the scheduler, and an error raised into
+    # the current Fiber, like a wait() timeout, is raised inside the block.
+    #
+    def run_in_new_fiber(&block)
+      future = get_current_future(fiber: Fiber.current)
+      return yield unless future
+
+      if future.nested_fibers.length >= MAX_NESTED_PLANS
+        msg = "Plans cannot be nested more than #{MAX_NESTED_PLANS} levels deep. Check for a " \
+              "plan that runs itself without a condition that stops it."
+        # A plain error, not a Bolt::Error, so plans can't catch it with
+        # _catch_errors or catch_errors() and keep a runaway plan going.
+        raise msg
+      end
+
+      fiber = Fiber.new(&block)
+      future.nested_fibers.push(fiber)
+      @logger.trace("Running a nested plan on a new Fiber in future '#{future.name}'")
+      begin
+        value = fiber.resume
+        while fiber.alive?
+          begin
+            resumed_with = Fiber.yield(value)
+          rescue StandardError => e
+            value = fiber.raise(e)
+            next
+          end
+          value = fiber.resume(resumed_with)
+        end
+        value
+      ensure
+        future.nested_fibers.delete(fiber)
+      end
     end
 
     # Get the plan invocation ID for the PlanFuture that is currently executing
@@ -146,7 +190,7 @@ module Bolt
         # finished. This ensures that we include futures created inside of
         # futures being waited on.
         until (futures = get_futures_for_plan(plan_id: plan_id)).map(&:alive?).none?
-          if futures.map(&:fiber).include?(Fiber.current)
+          if futures.include?(get_current_future(fiber: Fiber.current))
             msg = "The wait() function cannot be called with no arguments inside a " \
                   "background block in the same plan."
             raise Bolt::Error.new(msg, 'bolt/infinite-wait')

data/lib/bolt/inventory/options.rb

@@ -84,7 +84,7 @@ module Bolt
         "plugin_hooks" => {
           description: "Configuration for the Puppet library plugin used to install the " \
                        "Puppet agent on the target. For more information, see " \
-                       "https://pup.pt/bolt-plugin-hooks",
+                       "[plugin hooks](writing_plugins.md#plugin-hooks).",
           type: Hash,
           properties: {
             "puppet_library" => {

data/lib/bolt/inventory/target.rb

@@ -179,7 +179,7 @@ module Bolt
             'dotted_fact_name',
             "Target '#{safe_name}' includes dotted fact names: '#{dotted.join("', '")}'. Dotted fact " \
             "names are deprecated and Bolt does not automatically convert facts with dotted names to " \
-            "structured facts. For more information, see https://pup.pt/bolt-dotted-facts"
+            "structured facts. For more information, see https://docs.openvoxproject.org/openbolt/latest/developer_updates.html#deprecating-dotted-fact-names"
           )
         end
       end

data/lib/bolt/module_installer/puppetfile.rb

@@ -72,7 +72,7 @@ module Bolt
         File.open(path, 'w') do |file|
           if moduledir
             file.puts "# This Puppetfile is managed by Bolt. Do not edit."
-            file.puts "# For more information, see https://pup.pt/bolt-modules"
+            file.puts "# For more information, see https://docs.openvoxproject.org/openbolt/latest/modules.html"
             file.puts
             file.puts "# The following directive installs modules to the managed moduledir."
             file.puts "moduledir '#{moduledir.basename}'"

data/lib/bolt/module_installer/specs.rb

@@ -57,7 +57,7 @@ module Bolt
           Invalid module specification:
           #{hash.to_yaml.lines.drop(1).join.chomp}
 
-          To read more about specifying modules, see https://pup.pt/bolt-module-specs
+          To read more about specifying modules, see https://docs.openvoxproject.org/openbolt/latest/bolt_installing_modules.html#manually-specify-modules-in-a-bolt-project
         MESSAGE
       end
 

data/lib/bolt/outputter/human.rb

@@ -51,7 +51,6 @@ module Bolt
 
         @spinning = false
         @spin_thread.terminate
-        @stream.print("\b")
       end
 
       def remove_trail(string)
@@ -645,7 +644,7 @@ module Bolt
         info << indent(2, "#{modulepath.join(File::PATH_SEPARATOR)}\n\n")
 
         info << colorize(:cyan, "Additional information\n")
-        info << indent(2, "For more information about using plugins see https://pup.pt/bolt-plugins")
+        info << indent(2, "For more information about using plugins see https://docs.openvoxproject.org/openbolt/latest/using_plugins.html")
 
         @stream.puts info.chomp
       end

data/lib/bolt/plan_creator.rb

@@ -113,7 +113,7 @@ module Bolt
     private_class_method def self.yaml_plan(plan_name)
       <<~YAML
         # This is the structure of a simple plan. To learn more about writing
-        # YAML plans, see the documentation: http://pup.pt/bolt-yaml-plans
+        # YAML plans, see the documentation: https://docs.openvoxproject.org/openbolt/latest/writing_yaml_plans.html
 
         # The description sets the description of the plan that will appear
         # in 'bolt plan show' output.
@@ -146,7 +146,7 @@ module Bolt
     private_class_method def self.yaml_script_plan(script)
       <<~YAML
         # This is the structure of a simple plan. To learn more about writing
-        # YAML plans, see the documentation: http://pup.pt/bolt-yaml-plans
+        # YAML plans, see the documentation: https://docs.openvoxproject.org/openbolt/latest/writing_yaml_plans.html
 
         # The description sets the description of the plan that will appear
         # in 'bolt plan show' output.
@@ -177,7 +177,7 @@ module Bolt
     private_class_method def self.puppet_plan(plan_name)
       <<~PUPPET
         # This is the structure of a simple plan. To learn more about writing
-        # Puppet plans, see the documentation: http://pup.pt/bolt-puppet-plans
+        # Puppet plans, see the documentation: https://docs.openvoxproject.org/openbolt/latest/writing_plans.html
 
         # The summary sets the description of the plan that will appear
         # in 'bolt plan show' output. Bolt uses puppet-strings to parse the
@@ -202,7 +202,7 @@ module Bolt
     private_class_method def self.puppet_script_plan(plan_name, script)
       <<~PUPPET
         # This is the structure of a simple plan. To learn more about writing
-        # Puppet plans, see the documentation: http://pup.pt/bolt-puppet-plans
+        # Puppet plans, see the documentation: https://docs.openvoxproject.org/openbolt/latest/writing_plans.html
 
         # The summary sets the description of the plan that will appear
         # in 'bolt plan show' output. Bolt uses puppet-strings to parse the

data/lib/bolt/plan_future.rb

@@ -4,7 +4,7 @@ require 'fiber'
 
 module Bolt
   class PlanFuture
-    attr_reader :fiber, :id, :scope
+    attr_reader :fiber, :id, :scope, :nested_fibers
     attr_accessor :value, :plan_stack
 
     def initialize(fiber, id, plan_id:, name: nil, scope: nil)
@@ -24,6 +24,10 @@ module Bolt
       # since now we're in the calling plan. These IDs are unique to each plan
       # invocation, not just plan names.
       @plan_stack = [plan_id]
+
+      # Fibers this Future started with FiberExecutor#run_in_new_fiber, which
+      # still belong to this Future
+      @nested_fibers = []
     end
 
     def original_plan

data/lib/bolt/project_manager/inventory_migrator.rb

@@ -35,7 +35,7 @@ module Bolt
         rescue StandardError => e
           raise Bolt::FileError.new(
             "Unable to write to #{inventory_file}: #{e.message}. See " \
-            "http://pup.pt/bolt-inventory to manually update.",
+            "https://docs.openvoxproject.org/openbolt/latest/inventory_files.html to manually update.",
             inventory_file
           )
         end

data/lib/bolt/project_manager/module_migrator.rb

@@ -23,7 +23,7 @@ module Bolt
           @outputter.print_action_step(
             "Project has a non-default configured modulepath, unable to automatically " \
             "migrate project modules. To migrate project modules manually, see " \
-            "http://pup.pt/bolt-modules"
+            "https://docs.openvoxproject.org/openbolt/latest/modules.html"
           )
           true
         # Migrate modules from Puppetfile

data/lib/bolt/project_manager.rb

@@ -8,7 +8,7 @@ module Bolt
   class ProjectManager
     INVENTORY_TEMPLATE = <<~INVENTORY
       # This is an example inventory.yaml
-      # To read more about inventory files, see https://pup.pt/bolt-inventory
+      # To read more about inventory files, see https://docs.openvoxproject.org/openbolt/latest/inventory_files.html
       #
       # groups:
       #   - name: linux

data/lib/bolt/shell/bash.rb

@@ -230,15 +230,17 @@ module Bolt
         ''
       end
 
+      # GNU sudo and sudo-rs word these differently
       def handle_sudo_errors(err)
         case err
-        when /^#{conn.user} is not in the sudoers file\./
+        when /^#{conn.user} is not in the sudoers file\./,
+             /Sorry, user #{conn.user} (may not run|is not allowed to execute)/
           @logger.trace { err }
           raise Bolt::Node::EscalateError.new(
             "User #{conn.user} does not have sudo permission on #{target}",
             'SUDO_DENIED'
           )
-        when /^Sorry, try again\./
+        when /^Sorry, try again\./, /Authentication failed, try again/
           @logger.trace { err }
           raise Bolt::Node::EscalateError.new(
             "Sudo password for user #{conn.user} not recognized on #{target}",
@@ -299,7 +301,7 @@ module Bolt
 
         result = execute(command)
         if result.exit_code != 0
-          raise Bolt::Node::FileError.new("Could not make tmpdir: #{result.stderr.string}", 'TMPDIR_ERROR')
+          raise Bolt::Node::FileError.new("Could not make tmpdir #{tmppath}: #{result.stderr.string}", 'TMPDIR_ERROR')
         end
 
         path = tmppath || result.stdout.string.chomp
@@ -484,7 +486,8 @@ module Bolt
           @logger.trace { "Command `#{command_str}` returned successfully" }
         when 126
           msg = "\n\nThis might be caused by the default tmpdir being mounted " \
-            "using 'noexec'. See http://pup.pt/task-failure for details and workarounds."
+            "using 'noexec'. See https://docs.openvoxproject.org/openbolt/latest/troubleshooting.html" \
+            "#my-task-fails-with-a-permission-denied-error-noexec-issue for details and workarounds."
           result_output.stderr        << msg
           result_output.merged_output << msg
           @logger.trace { "Command #{command_str} failed with exit code #{result_output.exit_code}" }

data/lib/bolt/shell/powershell/snippets.rb

@@ -1,10 +1,23 @@
 # frozen_string_literal: true
 
+require 'base64'
+
 module Bolt
   class Shell
     class Powershell < Shell
       module Snippets
         class << self
+          # An expression that evaluates to the value as a string. The value
+          # is sent as base64, so no character in it can end a quoted string
+          # or here-string, whatever encoding PowerShell reads the script in.
+          # A binary string, such as a command-line argument under the C
+          # locale, goes as its bytes, which PowerShell reads as UTF-8.
+          def base64_string(value)
+            value = value.to_s
+            value = value.encode('UTF-8') unless value.encoding == Encoding::BINARY
+            "[System.Text.Encoding]::UTF8.GetString([System.Convert]::FromBase64String('#{Base64.strict_encode64(value)}'))"
+          end
+
           def execute_process(command)
             <<~PS
             if ([Console]::InputEncoding -eq [System.Text.Encoding]::UTF8) {
@@ -46,7 +59,7 @@ module Bolt
 
           def run_script(arguments, script_path)
             build_arg_list = arguments.map do |a|
-              "$invokeArgs.ArgumentList += @'\n#{a}\n'@"
+              "$invokeArgs.ArgumentList += #{base64_string(a)}"
             end.join("\n")
             <<~PS
             $invokeArgs = @{
@@ -138,6 +151,12 @@ module Bolt
             "${boltBaseDir}\\hiera\\lib;" +
             $ENV:RUBYLIB
 
+            #{json_functions.chomp}
+            PS
+          end
+
+          def json_functions
+            <<~PS
             function ConvertFrom-PSCustomObject
             {
             PARAM([Parameter(ValueFromPipeline = $true)] $InputObject)

data/lib/bolt/shell/powershell.rb

@@ -88,12 +88,12 @@ module Bolt
 
       def env_declarations(env_vars)
         env_vars.map do |var, val|
-          "[Environment]::SetEnvironmentVariable('#{var}', @'\n#{val}\n'@)"
+          "[Environment]::SetEnvironmentVariable(#{quote_string(var.to_s)}, #{Snippets.base64_string(val)})"
         end
       end
 
       def quote_string(string)
-        "'" + string.gsub("'", "''") + "'"
+        "'" + Bolt::Util.escape_powershell_single_quotes(string) + "'"
       end
 
       def write_executable(dir, file, filename = nil)

data/lib/bolt/transport/base.rb

@@ -165,18 +165,23 @@ module Bolt
       #
       # Transports may override this method to implement their own batch processing.
       def batch_download(targets, source, destination, options = {}, position = [], &callback)
-        require 'erb'
-
         assert_batch_size_one("batch_download()", targets)
         target = targets.first
         with_events(target, callback, 'download', position) do
-          escaped_name = ERB::Util.url_encode(target.safe_name)
-          target_destination = File.expand_path(escaped_name, destination)
+          target_destination = download_dir(destination, target)
           @logger.debug { "Downloading: '#{source}' on #{target.safe_name} to #{target_destination}" }
           download(target, source, target_destination, options)
         end
       end
 
+      # The directory under the destination that receives one target's
+      # download, named after the target so that a batch keeps them apart.
+      def download_dir(destination, target)
+        require 'erb'
+
+        File.expand_path(ERB::Util.url_encode(target.safe_name), destination)
+      end
+
       def batch_connected?(targets)
         assert_batch_size_one("connected?()", targets)
         connected?(targets.first)

data/lib/bolt/transport/choria/agent_discovery.rb

@@ -3,10 +3,9 @@
 module Bolt
   module Transport
     class Choria
-      SHELL_MIN_VERSION = '1.2.1'
-
       AGENT_MIN_VERSIONS = {
-        'shell' => SHELL_MIN_VERSION
+        'shell' => '1.2.1',
+        'file_transfer' => '1.0.0'
       }.freeze
 
       # Discover agents and detect OS on targets via two batched RPC calls
@@ -14,18 +13,20 @@ module Bolt
       # Populates @agent_cache with { agents: [...], os: 'redhat' | 'windows' | ... }.
       #
       # @param targets [Array<Bolt::Target>] Targets to discover agents on
+      # @return [Hash{Bolt::Target => Hash}] What discover_agent_list returns for the uncached
+      #   targets, or an empty hash when every target was already discovered.
       def discover_agents(targets)
-        uncached = targets.reject { |target| @agent_cache.key?(choria_identity(target)) }
-        return if uncached.empty?
+        uncached = targets.reject { |target| discovered?(target) }
+        return {} if uncached.empty?
 
         logger.debug { "Discovering agents on #{target_count(uncached)}" }
-        discover_agent_list(uncached)
+        failures = discover_agent_list(uncached)
         discover_os_family(uncached)
+        failures
+      end
 
-        uncached.each do |target|
-          identity = choria_identity(target)
-          logger.warn { "No response from #{identity} during agent discovery" } unless @agent_cache.key?(identity)
-        end
+      def discovered?(target)
+        @agent_cache.key?(choria_identity(target))
       end
 
       def has_agent?(target, agent_name)
@@ -40,26 +41,37 @@ module Bolt
       # and populate @agent_cache with agent lists.
       #
       # @param targets [Array<Bolt::Target>] Targets to query for agent inventory
+      # @return [Hash{Bolt::Target => Hash}] The error output of each target that did not answer,
+      #   answered with an RPC error or answered without an agent list, or of every target when
+      #   the call failed as a whole, such as when the client could not connect. An empty hash
+      #   when discovery itself raises, which leaves those targets undiscovered.
       def discover_agent_list(targets)
         response = rpc_request('rpcutil', targets, 'rpcutil.agent_inventory') do |client|
           client.agent_inventory
         end
-        response[:errors].each { |target, output| logger.debug { "agent_inventory failed for #{target.safe_name}: #{output[:error]}" } }
+        response[:errors].each { |target, output| logger.warn { "agent_inventory failed for #{target.safe_name}: #{output[:error]}" } }
 
+        malformed = {}
         response[:responded].each do |target, data|
           sender = choria_identity(target)
-          agents = filter_agents(sender, data[:agents])
+          # A node can send data that isn't a hash, which would raise and stop
+          # discovery for every target after it
+          agents = filter_agents(sender, data[:agents]) if data.is_a?(Hash)
           unless agents
-            logger.warn { "Unexpected agent_inventory response from #{sender}. This target will be treated as unreachable." }
+            logger.warn { "Unexpected agent_inventory response from #{sender}" }
+            malformed[target] = error_output("rpcutil.agent_inventory on #{target.safe_name} returned no agent list",
+                                             'bolt/choria-missing-data')
             next
           end
           @agent_cache[sender] = { agents: agents }
           logger.debug { "Discovered agents on #{sender}: #{agents.join(', ')}" }
         end
+        response[:errors].merge(malformed)
       rescue StandardError => e
         raise if e.is_a?(Bolt::Error)
 
         logger.warn { "Agent discovery failed: #{e.class}: #{e.message}" }
+        {}
       end
 
       # Detect the OS family on targets via rpcutil.get_fact and update
@@ -68,7 +80,7 @@ module Bolt
       # @param targets [Array<Bolt::Target>] Targets to detect OS on
       def discover_os_family(targets)
         # Only fetch OS for targets that responded to agent_inventory
-        responded = targets.select { |target| @agent_cache.key?(choria_identity(target)) }
+        responded = targets.select { |target| discovered?(target) }
         return if responded.empty?
 
         response = rpc_request('rpcutil', responded, 'rpcutil.get_fact') do |client|
@@ -82,6 +94,10 @@ module Bolt
 
         response[:responded].each do |target, data|
           sender = choria_identity(target)
+          unless data.is_a?(Hash)
+            logger.warn { "Unexpected get_fact response from #{sender}. Defaulting to POSIX command syntax." }
+            next
+          end
           os_family = data[:value].to_s.downcase
           if os_family.empty?
             logger.warn { "os.family fact is empty on #{sender}. Defaulting to POSIX command syntax." }
@@ -107,6 +123,8 @@ module Bolt
         return nil unless agent_list.is_a?(Array)
 
         agent_list.filter_map do |entry|
+          next unless entry.is_a?(Hash)
+
           name = entry['agent']
           next unless name