kitsune-kit 0.4.0 → 0.5.0

data/lib/kitsune/kit/operations/ensure_dns_records.rb

@@ -0,0 +1,165 @@
+# frozen_string_literal: true
+
+require "public_suffix"
+require_relative "../adapters/provider"
+require_relative "../errors"
+require_relative "../plan"
+
+module Kitsune
+  module Kit
+    module Operations
+      class EnsureDnsRecords
+        def initialize(config:, provider:, state_store:)
+          @config = config
+          @provider = provider
+          @state_store = state_store
+        end
+
+        def plan
+          server = find_server
+          desired_ip = server&.public_ip
+          entries = domains.map { |domain| plan_domain(domain, desired_ip) }
+          action = entries.all? { |entry| entry[:action] == "no_change" } ? "no_change" : "update"
+          Change.new(
+            resource: "dns",
+            action: action,
+            summary: summary_for(action, entries),
+            details: { records: entries, desired_ip: desired_ip, depends_on: desired_ip ? nil : "server" }
+          )
+        end
+
+        def apply(change)
+          return [] unless change.changed?
+
+          server = find_server
+          unless server&.public_ip
+            raise Errors::VerificationError.new(
+              "server has no public IP for DNS records",
+              hint: "Wait for the server to become ready, then resume apply."
+            )
+          end
+
+          domains.map do |domain|
+            apply_domain(domain, server.public_ip).tap { |entry| save_record(entry) }
+          end
+        end
+
+        def rollback
+          managed = state_store.read(config.environment).dig("resources", "dns")
+          return false unless managed
+
+          managed.each_value do |entry|
+            record = record_from_state(entry.fetch("record"))
+            previous = entry["previous"] && record_from_state(entry["previous"])
+            if previous
+              provider.upsert_dns_record(record: previous)
+            else
+              provider.delete_dns_record(id: record.id, zone: record.zone)
+            end
+          end
+          state_store.update(config.environment) do |state|
+            state["resources"].delete("dns")
+            state["operations"] << { "resource" => resource, "action" => "rollback", "status" => "applied" }
+            state
+          end
+          true
+        end
+
+        def resource = "dns"
+
+        private
+
+        attr_reader :config, :provider, :state_store
+
+        def domains = config.dns.domains
+
+        def find_server
+          provider.find_server(name: config.server.name, tags: config.server.tags)
+        end
+
+        def plan_domain(domain, desired_ip)
+          zone, name = zone_and_name(domain)
+          existing = provider.find_dns_record(zone: zone, name: name, type: "A")
+          action = if existing && desired_ip && existing.data == desired_ip && existing.ttl == config.dns.ttl
+                     "no_change"
+                   elsif existing
+                     "update"
+                   else
+                     "create"
+                   end
+          {
+            domain: domain,
+            zone: zone,
+            name: name,
+            action: action,
+            current: existing&.to_h,
+            desired: { data: desired_ip || "pending-server-ip", ttl: config.dns.ttl }
+          }
+        end
+
+        def apply_domain(domain, ip)
+          zone, name = zone_and_name(domain)
+          existing = provider.find_dns_record(zone: zone, name: name, type: "A")
+          desired = Adapters::DnsRecord.new(
+            id: existing&.id,
+            zone: zone,
+            name: name,
+            type: "A",
+            data: ip,
+            ttl: config.dns.ttl
+          )
+          saved = if existing&.data == ip && existing.ttl == config.dns.ttl
+                    existing
+                  else
+                    provider.upsert_dns_record(record: desired)
+                  end
+          previous = prior_previous(domain, existing)
+          { domain: domain, record: saved, previous: previous }
+        end
+
+        def zone_and_name(domain)
+          parsed = PublicSuffix.parse(domain)
+          zone = parsed.domain
+          prefix = domain.delete_suffix(zone).delete_suffix(".")
+          [zone, prefix.empty? ? "@" : prefix]
+        rescue PublicSuffix::Error
+          raise Errors::ConfigurationError.new(
+            "cannot determine the DNS zone for #{domain}",
+            hint: "Use a fully qualified public domain name."
+          )
+        end
+
+        def save_record(entry)
+          state_store.update(config.environment) do |state|
+            state["resources"]["dns"] ||= {}
+            state["resources"]["dns"][entry[:domain]] = {
+              "record" => stringify(entry[:record].to_h),
+              "previous" => entry[:previous] && stringify(entry[:previous].to_h)
+            }
+            state["operations"] << { "resource" => resource, "action" => "update", "status" => "applied" }
+            state
+          end
+        end
+
+        def prior_previous(domain, existing)
+          prior = state_store.read(config.environment).dig("resources", "dns", domain)
+          return existing unless prior
+
+          prior["previous"] && record_from_state(prior["previous"])
+        end
+
+        def stringify(hash) = hash.transform_keys(&:to_s)
+
+        def record_from_state(value)
+          Adapters::DnsRecord.new(**value.transform_keys(&:to_sym))
+        end
+
+        def summary_for(action, entries)
+          return "DNS records are current" if action == "no_change"
+
+          "Apply #{entries.count { |entry| entry[:action] != 'no_change' }} DNS record change(s)"
+        end
+      end
+    end
+  end
+end

data/lib/kitsune/kit/operations/ensure_server.rb

@@ -0,0 +1,135 @@
+# frozen_string_literal: true
+
+require_relative "../errors"
+require_relative "../plan"
+
+module Kitsune
+  module Kit
+    module Operations
+      class EnsureServer
+        attr_reader :config
+
+        def initialize(config:, provider:, state_store:, maximum_timeout: nil)
+          if maximum_timeout && maximum_timeout <= 0
+            raise Errors::ConfigurationError, "timeout must be greater than zero"
+          end
+
+          @config = config
+          @provider = provider
+          @state_store = state_store
+          @ready_timeout = maximum_timeout || 180
+        end
+
+        def plan
+          @provider.validate_server_spec!(spec: desired_spec)
+          existing = @provider.find_server(name: config.server.name, tags: config.server.tags)
+          return create_change unless existing
+
+          differences = differences_for(existing)
+          return unchanged_change(existing) if differences.empty?
+
+          Change.new(
+            resource: "server.#{config.server.name}",
+            action: "update",
+            summary: "Server exists with immutable configuration differences",
+            details: { id: existing.id, differences: differences, replacement_required: true },
+            destructive: true
+          )
+        end
+
+        def apply(change)
+          return current_server(change) unless change.changed?
+          if change.action == "update"
+            raise Errors::UnsafeOperationError.new(
+              "server replacement is required for #{config.server.name}",
+              hint: "Destroy the managed server explicitly, then run apply again.",
+              context: change.details
+            )
+          end
+
+          created = @provider.find_server(name: config.server.name, tags: config.server.tags) ||
+                    @provider.create_server(spec: desired_spec)
+          record_created(created)
+          ready = @provider.wait_until_ready(id: created.id, timeout: @ready_timeout)
+          record_ready(ready)
+          ready
+        rescue Errors::Error
+          raise
+        rescue StandardError => e
+          raise Errors::ProviderError.new(
+            "server operation failed for #{config.server.name}",
+            context: { cause: e.class.name },
+            retryable: true
+          )
+        end
+
+        def resource = "server.#{config.server.name}"
+        def state_key = "server"
+
+        private
+
+        def desired_spec
+          {
+            name: config.server.name,
+            region: config.server.region,
+            size: config.server.size,
+            image: config.server.image,
+            ssh_key_id: config.server.ssh_key_id,
+            tags: config.server.tags
+          }
+        end
+
+        def create_change
+          Change.new(
+            resource: resource,
+            action: "create",
+            summary: "Create server #{config.server.name}",
+            details: desired_spec.except(:ssh_key_id)
+          )
+        end
+
+        def unchanged_change(existing)
+          Change.new(
+            resource: resource,
+            action: "no_change",
+            summary: "Server #{config.server.name} is ready",
+            details: { id: existing.id, public_ip: existing.public_ip, status: existing.status }
+          )
+        end
+
+        def differences_for(existing)
+          {
+            region: [existing.region, config.server.region],
+            size: [existing.size, config.server.size],
+            image: [existing.image, config.server.image]
+          }.reject { |_key, (actual, desired)| actual.nil? || actual == desired }
+        end
+
+        def current_server(_change)
+          @provider.find_server(name: config.server.name, tags: config.server.tags) ||
+            raise(Errors::VerificationError, "server disappeared after planning")
+        end
+
+        def record_created(server)
+          @state_store.update(config.environment) do |state|
+            state["resources"]["server"] = state_record(server)
+            state["operations"] << { "resource" => resource, "action" => "create", "status" => "started" }
+            state
+          end
+        end
+
+        def record_ready(server)
+          @state_store.update(config.environment) do |state|
+            state["resources"]["server"] = state_record(server)
+            state["operations"] << { "resource" => resource, "action" => "create", "status" => "applied" }
+            state
+          end
+        end
+
+        def state_record(server)
+          server.to_h.transform_keys(&:to_s).merge("provider" => config.provider.name)
+        end
+      end
+    end
+  end
+end

data/lib/kitsune/kit/operations/ensure_service.rb

@@ -0,0 +1,285 @@
+# frozen_string_literal: true
+
+require "digest"
+require_relative "../errors"
+require_relative "../plan"
+require_relative "../service_compose"
+require_relative "service_files"
+require_relative "service_firewall"
+require_relative "service_backup"
+require_relative "service_state"
+
+module Kitsune
+  module Kit
+    module Operations
+      class EnsureService
+        TYPES = %w[postgres redis].freeze
+        MARKER_DIRECTORY = "/var/lib/kitsune/state"
+
+        attr_reader :resource
+
+        def initialize(config:, type:, transport_factory:, state_store:, secret_store:, clock: -> { Time.now.utc })
+          @config = config
+          @type = type.to_s
+          raise ArgumentError, "unknown service type: #{@type}" unless TYPES.include?(@type)
+
+          @service = config.services.public_send(@type)
+          @transport_factory = transport_factory
+          @state_store = state_store
+          @secret_store = secret_store
+          @resource = "service.#{@type}"
+          @managed_state = ServiceState.new(state_store: state_store, environment: config.environment,
+                                            resource: resource)
+          @compose = ServiceCompose.new(config: config, type: @type, service: @service)
+          @clock = clock
+        end
+
+        def plan
+          return pending_change unless transport_factory.server&.public_ip
+
+          actual = remote_fingerprint
+          files.ensure_managed_or_absent!(actual)
+          return unchanged_change if actual == fingerprint
+
+          Change.new(
+            resource: resource,
+            action: actual.empty? ? "create" : "update",
+            summary: "#{actual.empty? ? 'Install' : 'Update'} #{@type} service",
+            details: safe_details(actual)
+          )
+        rescue Errors::ConnectionError
+          pending_change
+        end
+
+        def apply(change)
+          return true unless change.changed?
+
+          @firewall_changed = false
+          previous_state = managed_state
+          password = validated_secret
+          files.ensure_managed_or_absent!(remote_fingerprint)
+          prepare_backup
+          transport.execute("mkdir", arguments: ["-p", service_directory], timeout: 10).then do |result|
+            raise_remote!(result, "create service directory")
+          end
+          transport.upload(content: compose_content, remote_path: compose_path, mode: "0600")
+          transport.upload(content: env_content(password), remote_path: env_path, mode: "0600")
+          validate_compose
+          start_service
+          verify_service
+          firewall.reconcile(managed_state)
+          @firewall_changed = true
+          write_marker
+          record_state(change)
+          true
+        rescue StandardError => e
+          recover_failed_apply(previous_state, e)
+        end
+
+        def remove
+          return false unless managed?
+
+          compose("down", "--remove-orphans")
+          firewall.remove
+          remove_marker
+          @managed_state.record_remove
+          true
+        end
+
+        def backup
+          backup_operation.call(managed_state)
+        end
+
+        def rollback
+          previous = managed_state
+          return false unless previous
+
+          remove
+          if files.restore_previous(previous, marker_path: marker_path)
+            firewall.restore(previous["previous_state"] || {})
+            start_service if previous.dig("previous_state", "installed") != false
+          end
+          @managed_state.record_rollback(previous)
+          true
+        end
+
+        def destroy_data
+          previous = managed_state
+          return false unless previous
+
+          compose("down", "--volumes", "--remove-orphans")
+          files.destroy(previous, marker_path: marker_path)
+          @managed_state.delete("destroy_data")
+          true
+        end
+
+        private
+
+        attr_reader :config, :service, :transport_factory, :state_store, :secret_store
+
+        def transport = @transport ||= transport_factory.deploy
+        def project_name = "kitsune-#{config.environment}-#{@type}"
+        def service_directory = files.directory
+        def compose_path = files.compose_path
+        def env_path = files.env_path
+        def marker_path = "#{MARKER_DIRECTORY}/service-#{@type}.sha256"
+
+        def fingerprint
+          @fingerprint ||= Digest::SHA256.hexdigest(
+            [compose_content, service.password_env, secret_digest, service.publish, service.bind,
+             *service.allowed_cidrs].join("\0")
+          )
+        end
+
+        def secret_digest = Digest::SHA256.hexdigest(validated_secret)
+
+        def remote_fingerprint
+          result = transport.execute("sudo", arguments: ["cat", marker_path], timeout: 10)
+          result.success? ? result.stdout.strip : ""
+        end
+
+        def validated_secret
+          value = secret_store.fetch(service.password_env)
+          if value.match?(/[\n\r\0]/)
+            raise Errors::ConfigurationError, "#{service.password_env} contains unsupported control characters"
+          end
+
+          value
+        end
+
+        def compose_content = @compose.content
+        def env_content(password) = @compose.env_content(password)
+
+        def validate_compose = compose("config", "--quiet")
+        def start_service = compose("up", "--detach", "--wait", "--wait-timeout", "120")
+
+        def verify_service
+          result = compose("ps", "--status", "running", "--quiet")
+          raise Errors::VerificationError, "#{@type} did not reach running state" if result.stdout.strip.empty?
+        end
+
+        def compose(*arguments)
+          result = transport.execute(
+            "docker",
+            arguments: ["compose", "--project-directory", service_directory, "--file", compose_path, *arguments],
+            timeout: 180
+          )
+          raise_remote!(result, "docker compose #{arguments.first}")
+          result
+        end
+
+        def prepare_backup = @backed_up_files = managed? ? files.prepare_backup : []
+
+        def restore_after_failed_apply(previous_state)
+          return unless defined?(@backed_up_files)
+
+          firewall.restore_after_failure(previous_state) if @firewall_changed
+          safe_compose_down
+          files.restore_after_failure(@backed_up_files)
+          start_service unless @backed_up_files.empty?
+        end
+
+        def recover_failed_apply(previous_state, original_error)
+          restore_after_failed_apply(previous_state)
+        rescue StandardError => e
+          raise Errors::VerificationError.new(
+            "#{@type} apply failed and automatic recovery also failed",
+            hint: "Stop changes and inspect the service files, container, volume and firewall before retrying.",
+            context: { original_error: error_name(original_error), recovery_error: error_name(e) }
+          )
+        else
+          raise original_error
+        end
+
+        def safe_compose_down
+          transport.execute(
+            "docker",
+            arguments: ["compose", "--project-directory", service_directory, "--file", compose_path,
+                        "down", "--remove-orphans"],
+            timeout: 180
+          )
+        end
+
+        def write_marker
+          transport.execute("sudo", arguments: ["install", "-d", "-m", "0755", MARKER_DIRECTORY], timeout: 10)
+          result = transport.execute("sudo", arguments: ["tee", marker_path], stdin: "#{fingerprint}\n", timeout: 10)
+          raise_remote!(result, "write service marker")
+        end
+
+        def remove_marker
+          result = transport.execute("sudo", arguments: ["rm", "-f", marker_path], timeout: 10)
+          raise_remote!(result, "remove service marker")
+        end
+
+        def record_state(change)
+          @managed_state.record_apply(change.action, state_attributes(change))
+        end
+
+        def state_attributes(change)
+          {
+            "managed" => true, "installed" => true, "fingerprint" => fingerprint, "project" => project_name,
+            "directory" => service_directory, "published" => service.publish, "port" => service.port,
+            "firewall_rules_added" => firewall.owned_rules, "firewall_drop_added" => firewall.drop_owned,
+            "volume" => "#{project_name}_data",
+            "previous_fingerprint" => change.details[:actual_fingerprint] || change.details["actual_fingerprint"],
+            "backup_directory" => @backed_up_files ? files.backup_directory : nil,
+            "backed_up_files" => @backed_up_files || []
+          }
+        end
+
+        def managed? = @managed_state.managed?
+        def managed_state = @managed_state.current
+
+        def files
+          @files ||= ServiceFiles.new(
+            config: config, type: @type, transport: transport, state_store: state_store,
+            resource: resource, fingerprint: -> { fingerprint }
+          )
+        end
+
+        def firewall
+          @firewall ||= ServiceFirewall.new(
+            config: config, type: @type, service: @service, transport: transport,
+            state_store: state_store, resource: resource
+          )
+        end
+
+        def backup_operation
+          @backup_operation ||= ServiceBackup.new(
+            config: config, type: @type, transport: transport, compose: method(:compose), clock: @clock
+          )
+        end
+
+        def raise_remote!(result, action)
+          return if result.success?
+
+          raise Errors::RemoteCommandError.new(
+            "unable to #{action} for #{@type}",
+            context: { stderr: result.stderr, exit_status: result.exit_status }
+          )
+        end
+
+        def pending_change
+          Change.new(resource: resource, action: "create", summary: "Install #{@type} after Docker is ready",
+                     details: safe_details("").merge(depends_on: "docker"))
+        end
+
+        def unchanged_change
+          Change.new(resource: resource, action: "no_change", summary: "#{@type} is configured",
+                     details: safe_details(fingerprint))
+        end
+
+        def safe_details(actual)
+          {
+            image: service.image, published: service.publish,
+            bind: service.publish ? service.bind : nil, port: service.publish ? service.port : nil,
+            fingerprint: fingerprint,
+            actual_fingerprint: actual.empty? ? nil : actual
+          }
+        end
+
+        def error_name(error) = error.respond_to?(:code) ? error.code : error.class.name
+      end
+    end
+  end
+end