html2rss 0.32.2 → 0.32.4

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: e095abe52bc258d262e3a24909e0feb480b2eb1d537e57580fc271203b0c6dbc
-  data.tar.gz: 8fdabe7ed5afee08c46ab94c2b6f80c1f8c66df9ffec9439e40e80fa897bfa96
+  metadata.gz: a53d8374b6b4ce7551c9f1bffd0e14960995f7df7380554f1292e0a70317cf58
+  data.tar.gz: 90e91d4275c03c120924cca66230a7aecb3606f2f50c6224d019c138cae244e2
 SHA512:
-  metadata.gz: fdcce48a2a4fb3a8c3ea02fd8f4bb799bd8ed7acb4e66e74da3cde693f6a51cc3aa72d363670b13f077d1521d97321e0676d14b59c6b61f1c685fdc4db64cb67
-  data.tar.gz: ce1c1cac4c38f4660e61cb7c9506991d8c0d847c8bdfbb583c678680854b7327a16b9aa637af641aaae42bed9f15b9330cc344ee922252da16863a517569fad4
+  metadata.gz: d2543d4431b2790c89c19ab207d0cdbbc7416493ea1184347746bc26e21068415d68f025e0cc253082224ca5191a3984163e3e48ef3e727cb3b78c19cddd3d64
+  data.tar.gz: ac588fece229b58ab2ea4de91bb467fdbfaf199f43febfd8be6b39d817fc9b9b2867c21cb472949e886ad4dd50b436d9c5c149c0976a87b8cd67d09a6a62f9f6

data/lib/html2rss/error.rb

@@ -3,11 +3,60 @@
 module Html2rss
   # The Html2rss::Error base class.
   class Error < StandardError
+    # Upstream transport_meta keys promoted into error diagnostics.
+    TRANSPORT_META_KEYS = %i[request_id strategy_used render_ms error_category timeout_phase].freeze
+
+    class << self
+      ##
+      # Builds a sparse diagnostics hash from attempt hashes and optional error metadata.
+      #
+      # @param attempts [Array<Hash>, nil] tier attempt diagnostics
+      # @param transport_meta [Hash, nil] direct error transport metadata
+      # @param timeout_phase [String, nil] timeout stage when known
+      # @param surface_category [Symbol, nil] page surface classification when known
+      # @return [Hash{Symbol => Object}]
+      def diagnostics_from_attempts(attempts, transport_meta: nil, timeout_phase: nil, surface_category: nil)
+        list = Array(attempts)
+        meta = list.filter_map { |item| item[:transport_meta] || item['transport_meta'] }.last || transport_meta
+        fields = extract_transport_fields(meta)
+        fields[:timeout_phase] ||= timeout_phase
+        fields[:surface_category] = surface_category
+        { strategy_attempts: (list if list.any?), **fields }.compact
+      end
+
+      private
+
+      def extract_transport_fields(meta)
+        return {} unless meta
+
+        TRANSPORT_META_KEYS.to_h { |key| [key, meta[key] || meta[key.to_s]] }
+      end
+    end
+
     # @return [Array<Hash{Symbol => Object}>, nil] tier attempt diagnostics when set by AutoFallback
     attr_accessor :attempts
 
     # @return [Hash{String => Object}, nil] allowlisted upstream transport telemetry when attached
     attr_accessor :transport_meta
+
+    # @return [String, nil] scrape-api timeout stage when known (overridden by RequestTimedOut)
+    def timeout_phase = nil
+
+    # @return [Symbol, nil] surface classification when known (overridden by NoFeedItemsExtracted)
+    def surface_category = nil
+
+    ##
+    # Sparse telemetry hash for observability consumers.
+    #
+    # @return [Hash{Symbol => Object}]
+    def diagnostics
+      self.class.diagnostics_from_attempts(
+        attempts,
+        transport_meta:,
+        timeout_phase:,
+        surface_category:
+      )
+    end
   end
 
   # Raised when auto fallback exhausts all concrete tiers and extractors find no feed items.

data/lib/html2rss/feed_pipeline/README.md

@@ -35,6 +35,10 @@ These abort immediately (`AutoFallback::NON_FALLBACK_ERRORS`): unknown strategy,
 
 Retries share the feed's request/session policy. Pin a concrete strategy when you need a single-hop budget profile.
 
+### Terminal fallback outcome
+
+A hop that returns zero items does **not** erase an earlier terminal transport failure (`RequestTimedOut`, `BlockedSurfaceDetected`, or `ConnectionFailed`). `AutoFallback` remembers that error across later hops and, when no later hop yields items, re-raises it with `attempts` attached. A later hop that extracts items still succeeds. Two zero-item hops with no terminal transport error still raise `NoFeedItemsExtracted`. True empty stays the web policy already in force: listing `EXTRACTION_EMPTY` is cacheable; `app_shell` and `high_entropy_surface` are not. `html2rss-web` `ErrorClassifier` maps the raised error to HTTP/cache policy; it does not invent a second cache rule for empty extraction.
+
 ## Success signals
 
 - `Html2rss::RequestService::Response` includes transport metadata for the strategy that produced it.

data/lib/html2rss/feed_pipeline/auto_fallback.rb

@@ -27,16 +27,27 @@ module Html2rss
         RequestService::RedirectLimitReached
       ].freeze
 
+      # Transport failures preserved across a later hop that extracts zero items.
+      # Re-raised from finalize_failure so web maps them to uncacheable gateway/blocked
+      # decisions instead of EXTRACTION_EMPTY.
+      TERMINAL_TRANSPORT_ERRORS = [
+        RequestService::RequestTimedOut,
+        RequestService::BlockedSurfaceDetected,
+        RequestService::ConnectionFailed
+      ].freeze
+
       ##
       # Mutable run state for one auto-fallback chain: attempts plus selected result.
       class AttemptState
-        attr_reader :attempts, :result, :last_response, :last_error, :entry_resolution
+        attr_reader :attempts, :result, :last_response, :last_error, :entry_resolution,
+                    :terminal_transport_error
 
         def initialize
           @attempts = []
           @result = nil
           @last_response = nil
           @last_error = nil
+          @terminal_transport_error = nil
           @entry_resolution = nil
           @resolution_tried = false
           @aborted = false
@@ -69,24 +80,25 @@ module Html2rss
 
         # @param strategy [Symbol] strategy that raised
         # @param error [Exception] caught error
+        # @param elapsed_ms [Integer, nil] monotonic hop duration in milliseconds
         # @return [void]
-        def record_error(strategy:, error:)
-          attempt = { strategy:, items_count: nil, error_class: error.class.name }
+        def record_error(strategy:, error:, elapsed_ms: nil)
           meta = error.is_a?(Html2rss::Error) ? error.transport_meta : nil
-          attempt[:transport_meta] = meta if meta&.any?
-          @attempts << attempt
+          append_attempt(strategy:, error_class: error.class.name, transport_meta: meta, elapsed_ms:)
           @last_error = error
+          return unless TERMINAL_TRANSPORT_ERRORS.any? { |klass| error.is_a?(klass) }
+
+          @terminal_transport_error = error
         end
 
         # @param strategy [Symbol] strategy that returned a response
         # @param items_count [Integer] extracted article count
         # @param transport_meta [Hash, nil] optional allowlisted upstream telemetry
+        # @param elapsed_ms [Integer, nil] monotonic hop duration in milliseconds
         # @return [void]
-        def record_items(strategy:, items_count:, transport_meta: nil)
+        def record_items(strategy:, items_count:, transport_meta: nil, elapsed_ms: nil)
           @last_error = nil
-          attempt = { strategy:, items_count:, error_class: nil }
-          attempt[:transport_meta] = transport_meta if transport_meta && !transport_meta.empty?
-          @attempts << attempt
+          append_attempt(strategy:, items_count:, transport_meta:, elapsed_ms:)
         end
 
         # @param response [RequestService::Response] last response that reached extraction
@@ -118,6 +130,37 @@ module Html2rss
             entry_resolution:
           )
         end
+
+        # Raises the terminal transport error, last error, or NoFeedItemsExtracted with attempts attached.
+        #
+        # @return [void] always raises
+        # @raise [Html2rss::Error, StandardError]
+        def raise_failure!
+          if (error = failure_error)
+            error.attempts = attempts if error.is_a?(Html2rss::Error)
+            raise error
+          end
+
+          response = @last_response
+          surface_category = PageRecon.surface_category_for(response:, url: response.url) if response
+          raise NoFeedItemsExtracted.new(attempts:, surface_category:)
+        end
+
+        private
+
+        def append_attempt(strategy:, items_count: nil, error_class: nil, transport_meta: nil, elapsed_ms: nil)
+          attempt = { strategy:, items_count:, error_class: }
+          attempt[:elapsed_ms] = elapsed_ms unless elapsed_ms.nil?
+          attempt[:transport_meta] = transport_meta if transport_meta&.any?
+          @attempts << attempt
+        end
+
+        def failure_error
+          return @terminal_transport_error if @terminal_transport_error
+          return if @last_error.nil? || @last_error.is_a?(RequestService::BotasaurusConfigurationError)
+
+          @last_error
+        end
       end
 
       ##
@@ -142,7 +185,7 @@ module Html2rss
         state = run_attempts
         return state.result if state.succeeded?
 
-        finalize_failure(state)
+        state.raise_failure!
       end
 
       private
@@ -158,35 +201,33 @@ module Html2rss
         end
       end
 
+      # rubocop:disable-next Metrics/MethodLength, Metrics/AbcSize -- hop timing + NON_FALLBACK rescue stay co-located
       def attempt(strategy:, next_strategy:, state:)
+        started_ms = monotonic_ms
         request_session = pipeline.request_session_for(
           config, strategy:, resources:, scrape_url: scrape_target.effective_url
         )
-        response = fetch_response(request_session:, strategy:, next_strategy:, state:)
-        return unless response
-
-        process_response(response:, strategy:, next_strategy:, request_session:, state:)
-      rescue RequestService::UnsupportedResponseContentType => error
-        state.record_error(strategy:, error:)
-        log_fallback_error(strategy:, next_strategy:, error:, request_session:) if next_strategy
-      end
-
-      def fetch_response(request_session:, strategy:, next_strategy:, state:)
-        request_session.fetch_initial_response
-      rescue *NON_FALLBACK_ERRORS
-        raise
+        response = request_session.fetch_initial_response
+        process_response(
+          response:, strategy:, next_strategy:, request_session:, state:,
+          elapsed_ms: monotonic_ms - started_ms
+        )
+      rescue *NON_FALLBACK_ERRORS => error
+        state.record_error(strategy:, error:, elapsed_ms: monotonic_ms - started_ms)
+        state.abort!
       rescue StandardError => error
-        state.record_error(strategy:, error:)
-        log_fallback_error(strategy:, next_strategy:, error:, request_session:) if next_strategy
-        nil
+        state.record_error(strategy:, error:, elapsed_ms: monotonic_ms - started_ms)
+        log_fallback_error(strategy:, next_strategy:, error:, request_session:) if next_strategy && request_session
       end
 
-      # rubocop:disable-next Metrics/MethodLength, Metrics/AbcSize -- extract + resolution delegate
-      def process_response(response:, strategy:, next_strategy:, request_session:, state:)
-        articles, dedup_dropped, admission_drops = articles_for(response:, request_session:)
+      # rubocop:disable-next Metrics/MethodLength, Metrics/AbcSize, Metrics/ParameterLists -- extract + resolution delegate
+      def process_response(response:, strategy:, next_strategy:, request_session:, state:, elapsed_ms: nil)
+        articles, dedup_dropped, admission_drops = pipeline.deduplicated_articles(
+          config:, response:, request_session:
+        )
         state.remember_response(response)
         items_count = articles.size
-        state.record_items(strategy:, items_count:, transport_meta: response.transport_meta)
+        state.record_items(strategy:, items_count:, transport_meta: response.transport_meta, elapsed_ms:)
         Log.debug("#{self.class}: strategy=#{strategy} items=#{items_count} " \
                   "host=#{response.url.host} elapsed=#{format('%.3f', budget.elapsed_seconds)}s " \
                   "budget_remaining=#{budget_remaining_label}")
@@ -204,7 +245,7 @@ module Html2rss
                                 state:)
         end
 
-        if deterministic_status?(response)
+        if DETERMINISTIC_HTTP_STATUSES.member?(response&.status)
           state.abort!
           Log.debug("#{self.class}: deterministic HTTP #{response.status}; aborting auto fallback")
           return
@@ -213,14 +254,6 @@ module Html2rss
         log_info_fallback_zero_items(strategy:, next_strategy:, response:) if next_strategy
       end
 
-      def deterministic_status?(response)
-        DETERMINISTIC_HTTP_STATUSES.member?(response&.status)
-      end
-
-      def articles_for(response:, request_session:)
-        pipeline.deduplicated_articles(config:, response:, request_session:)
-      end
-
       # rubocop:disable-next Metrics/ParameterLists, Metrics/MethodLength -- success kwargs + structured selected log
       def record_success(response:, strategy:, articles:, dedup_dropped:, admission_drops:, state:)
         attempt_count = state.attempts.size
@@ -242,58 +275,24 @@ module Html2rss
         }.to_json)
       end
 
-      def finalize_failure(state)
-        if raise_last_error?(state)
-          error = state.last_error
-          error.attempts = state.attempts if error.is_a?(Html2rss::Error)
-          raise error
-        end
-
-        raise NoFeedItemsExtracted.new(
-          attempts: state.attempts,
-          surface_category: surface_category_for(state.last_response)
-        )
-      end
-
-      def raise_last_error?(state)
-        error = state.last_error
-        return false if error.nil? || error.is_a?(RequestService::BotasaurusConfigurationError)
+      def monotonic_ms = Process.clock_gettime(Process::CLOCK_MONOTONIC, :millisecond)
 
-        true
-      end
-
-      def surface_category_for(response)
-        return unless response
-
-        PageRecon.surface_category_for(response:, url: response.url)
-      end
-
-      # rubocop:disable-next Metrics/AbcSize, Metrics/MethodLength
+      # rubocop:disable-next Metrics/AbcSize, Metrics/MethodLength -- shared payload then info/warn dispatch
       def log_fallback_error(strategy:, next_strategy:, error:, request_session:)
         host = request_session.url.host
         detail = "host=#{host} elapsed=#{format('%.3f', budget.elapsed_seconds)}s " \
                  "budget_remaining=#{budget_remaining_label}"
-        if error.is_a?(RequestService::RequestTimedOut)
-          Log.info({
-            event_name: 'auto_fallback.timeout',
-            host:,
-            strategy: strategy.to_s,
-            next_strategy: next_strategy.to_s,
-            error_class: error.class.name,
-            message: "#{self.class}: auto fallback #{strategy} -> #{next_strategy} " \
-                     "after timeout=#{error.class} #{detail}"
-          }.to_json)
-        else
-          Log.warn({
-            event_name: 'auto_fallback.error',
-            host:,
-            strategy: strategy.to_s,
-            next_strategy: next_strategy.to_s,
-            error_class: error.class.name,
-            message: "#{self.class}: auto fallback #{strategy} -> #{next_strategy} " \
-                     "after error=#{error.class} #{detail}"
-          }.to_json)
-        end
+        timed_out = error.is_a?(RequestService::RequestTimedOut)
+        payload = {
+          event_name: timed_out ? 'auto_fallback.timeout' : 'auto_fallback.error',
+          host:,
+          strategy: strategy.to_s,
+          next_strategy: next_strategy.to_s,
+          error_class: error.class.name,
+          message: "#{self.class}: auto fallback #{strategy} -> #{next_strategy} " \
+                   "after #{timed_out ? "timeout=#{error.class}" : "error=#{error.class}"} #{detail}"
+        }.to_json
+        timed_out ? Log.info(payload) : Log.warn(payload)
         Log.debug("#{self.class}: strategy=#{strategy} error=#{error.class}: #{error.message} #{detail}")
       end
 

data/lib/html2rss/request_service/httpx_strategy.rb

@@ -21,9 +21,9 @@ module Html2rss
       # combined with ping! on a reused closed selector that never decrements retries,
       # that overflows the stack and kills Falcon workers.
       RETRY_AFTER_SECONDS = 0.25
-      # Bounds passed to the retries plugin and every +session.with+.
-      # HTTPX retries stay inside one {RequestService::Budget} slot and the shared
-      # wall-clock deadline; do not consume a slot per attempt.
+      # Base-session retries plugin defaults (stack-overflow guard for ping reconnects).
+      # Per-request +session.with+ sets +max_retries: 0+; Ruby owns TimeoutError retries
+      # against the monotonic deadline so HTTPX cannot stack read timeouts past the cap.
       # Do not reload +:fiber_concurrency+: a reused closed selector made +ping!+
       # skip decrementing retries and recurse until +SystemStackError+.
       #
@@ -54,19 +54,19 @@ module Html2rss
       def perform_execute
         deadline = request_deadline
         reset_redirect_tracking!
-        raw_response = request_with_terminal_redirect_retry(deadline:)
+        raw_response = request_with_deadline_retries(deadline:)
         build_response(raw_response)
       end
 
       private
 
-      # Caps one HTTPX execute (including internal MAX_RETRIES) so a hanging
-      # :default tier cannot consume the full shared auto-fallback budget.
+      # Caps one strategy execute (Ruby retries included) so a hanging :default tier
+      # cannot consume the full shared auto-fallback budget.
       #
       # @return [Float] monotonic deadline
       def request_deadline
         budget_timeout = ctx.budget.effective_timeout_seconds(fallback: ctx.policy.total_timeout_seconds)
-        monotonic_now + [budget_timeout, attempt_timeout_ceiling].min
+        Process.clock_gettime(Process::CLOCK_MONOTONIC) + [budget_timeout, attempt_timeout_ceiling].min
       end
 
       # @return [Float] connect + read ceiling for a single attempt
@@ -81,8 +81,30 @@ module Html2rss
         @current_url = request_url
       end
 
-      def request_with_terminal_redirect_retry(deadline:)
-        raw_response = execute_http_request(deadline:)
+      # Retries HTTPX timeouts in Ruby while the shared monotonic deadline has time left.
+      # First try consumes the budget slot; retries do not.
+      def request_with_deadline_retries(deadline:)
+        state = { retries_remaining: MAX_RETRIES, consume_budget: true }
+
+        begin
+          request_with_terminal_redirect_retry(deadline:, consume_budget: state[:consume_budget])
+        rescue HTTPX::TimeoutError => error
+          prepare_deadline_retry!(deadline:, state:, error:)
+          retry
+        end
+      end
+
+      def prepare_deadline_retry!(deadline:, state:, error:)
+        raise error if state[:retries_remaining].zero?
+
+        remaining_timeout_seconds(deadline)
+        state[:retries_remaining] -= 1
+        state[:consume_budget] = false
+        reset_redirect_tracking!
+      end
+
+      def request_with_terminal_redirect_retry(deadline:, consume_budget: true)
+        raw_response = execute_http_request(deadline:, consume_budget:)
         return raw_response unless redirect_limit_reached?(raw_response)
 
         unless terminal_redirect_retryable?(raw_response)
@@ -164,6 +186,7 @@ module Html2rss
         session.with(session_options(deadline))
       end
 
+      # Plugin retries stay off per request; Ruby owns deadline-bounded TimeoutError retries.
       def session_options(deadline)
         remaining = remaining_timeout_seconds(deadline)
         {
@@ -173,7 +196,7 @@ module Html2rss
           max_response_body_size: ctx.policy.max_response_bytes,
           resolver_class: :system,
           redirect_on: redirect_callback,
-          **RETRY_OPTIONS
+          **RETRY_OPTIONS.merge(max_retries: 0)
         }
       end
 
@@ -219,16 +242,12 @@ module Html2rss
       end
 
       def remaining_timeout_seconds(deadline)
-        remaining = deadline - monotonic_now
+        remaining = deadline - Process.clock_gettime(Process::CLOCK_MONOTONIC)
         raise RequestTimedOut, 'Request timed out' if remaining <= 0
 
         remaining
       end
 
-      def monotonic_now
-        Process.clock_gettime(Process::CLOCK_MONOTONIC)
-      end
-
       def timeout_error?(error)
         error.is_a?(HTTPX::TimeoutError) || super
       end

data/lib/html2rss/request_service/local_file_strategy.rb

@@ -15,6 +15,9 @@ module Html2rss
 
       def skip_postflight? = true
 
+      # Local files never open a socket; keep filesystem errors as wire classes.
+      def connection_error?(_error) = false
+
       ##
       # @return [Response] the mock response wrapped around the file contents
       # @raise [ArgumentError] if the local file path is missing

data/lib/html2rss/request_service/strategy.rb

@@ -133,9 +133,9 @@ module Html2rss
 
       # @param error [StandardError]
       # @return [void]
-      # @raise [StandardError]
+      # @raise [ConnectionFailed]
       def translate_connection_error(error)
-        raise error
+        raise ConnectionFailed, error.message
       end
 
       # @param error [StandardError]

data/lib/html2rss/request_service.rb

@@ -28,6 +28,8 @@ module Html2rss
     class RedirectLimitReached < Html2rss::Error; end
     # Raised when blocked content surfaces are detected.
     class BlockedSurfaceDetected < Html2rss::Error; end
+    # Raised when the target connection fails before a response document exists.
+    class ConnectionFailed < Html2rss::Error; end
 
     # Raised when a request times out.
     class RequestTimedOut < Html2rss::Error
@@ -76,6 +78,12 @@ module Html2rss
     # @return [Symbol]
     DEFAULT_STRATEGY_NAME = :default
 
+    # Maximum distinct [strategy, host] deprecation warnings tracked per process.
+    MAX_DEPRECATION_WARNINGS = 256
+    # Mutex guarding process-local deprecation warning deduplication.
+    DEPRECATION_MUTEX = Mutex.new
+    private_constant :DEPRECATION_MUTEX
+
     class << self
       # @return [Symbol] the default strategy name
       def default_strategy_name = DEFAULT_STRATEGY_NAME
@@ -107,6 +115,13 @@ module Html2rss
         strategy_class.new(ctx).execute
       end
 
+      ##
+      # Clears process-local deprecation warning state (used by specs).
+      # @return [void]
+      def reset_deprecation_warnings!
+        DEPRECATION_MUTEX.synchronize { @warned_deprecations = Set.new }
+      end
+
       private
 
       ##
@@ -125,14 +140,15 @@ module Html2rss
       end
 
       ##
-      # Emits a structured +strategy_deprecated+ warning once per session (+ :initial + only).
+      # Emits a structured +strategy_deprecated+ warning once per +(strategy, host)+ per process
+      # on initial session requests (+relation == :initial+). Multi-worker servers may emit once per worker.
       # @param strategy [Symbol] deprecated strategy alias
       # @param replacement [Symbol] preferred strategy name
       # @param relation [Symbol] request relation
       # @param host [String, nil] request host
       # @return [void]
       def warn_migration_strategy(strategy, replacement, relation:, host:)
-        return unless relation == :initial
+        return unless relation == :initial && claim_deprecation_warning?(strategy, host)
 
         Log.warn({
           event_name: 'strategy_deprecated',
@@ -142,6 +158,14 @@ module Html2rss
           message: "#{name}: strategy=#{strategy.inspect} is deprecated; using #{replacement.inspect} instead."
         }.to_json)
       end
+
+      def claim_deprecation_warning?(strategy, host)
+        DEPRECATION_MUTEX.synchronize do
+          @warned_deprecations ||= Set.new
+          @warned_deprecations.clear if @warned_deprecations.size >= MAX_DEPRECATION_WARNINGS
+          !@warned_deprecations.add?([strategy, host]).nil?
+        end
+      end
     end
   end
 end

data/lib/html2rss/version.rb

@@ -2,6 +2,6 @@
 
 module Html2rss
   # Current application version.
-  VERSION = '0.32.2'
+  VERSION = '0.32.4'
   public_constant :VERSION
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: html2rss
 version: !ruby/object:Gem::Version
-  version: 0.32.2
+  version: 0.32.4
 platform: ruby
 authors:
 - Gil Desmarais
@@ -478,7 +478,7 @@ licenses:
 - MIT
 metadata:
   allowed_push_host: https://rubygems.org
-  changelog_uri: https://github.com/html2rss/html2rss/releases/tag/v0.32.2
+  changelog_uri: https://github.com/html2rss/html2rss/releases/tag/v0.32.4
   rubygems_mfa_required: 'true'
 rdoc_options: []
 require_paths: