flare 0.4.0 → 0.4.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: cee6e0910ebcd04f073efc41638823f7ed4d7abf511ab71b68caead994997af1
-  data.tar.gz: f28d669e444b2abb8b6043990b8f1cfc5cbd39abc14d25942cf5f41e08a80cb2
+  metadata.gz: c94277634cce4170817b717f3ff6b3cb99ac980908f51ed6aaa6ba07a8c59de6
+  data.tar.gz: 4d882034612ac608246ea1a3f232a5f491c283d97b6082a97115f60f0b80fdf5
 SHA512:
-  metadata.gz: 06647d35a64ae38dafb6e946b181fd87a8b8dc8e7065c5b0d4af3cb47419bd5be5dc9e930f3aa322740e824279169d9b2ae867c2b3c14dfda22c1d01be25507d
-  data.tar.gz: 837f70a2ef44f4d0f8c415cc15c6455ab5250b63e7dd06580ad640492385d8d1e6842cda3f56b036d918bb38c7b55438a2d60546a6c41fd7ce77a429b2d1793a
+  metadata.gz: f8cfaf0827068925802b90582a3c12a2513649db82b23e51509ed689c111cfcbcd7397c2c260d13934dadfe3a2899481925875faf0bc75f3262c326fa8718e63
+  data.tar.gz: 38727869b8f7ebf8283354a7c696aded643a5da8260bce61d2ddacb4e6e441868d72eed7c01766b9e8d83d2d95954b1fab2fd75e8cdd1bf7adfabef3d23e4ae5

data/lib/flare/lifecycle.rb

@@ -6,11 +6,14 @@ require_relative "deadline"
 
 module Flare
   module Lifecycle
+    FLUSH_TIMEOUT = 5 # seconds, total across traces and metrics
+
     # Flush all trace processors and the separate aggregated metric pipeline
     # using one monotonic timeout budget. This is safe to call from lifecycle
-    # hooks for short-lived and fork-per-job workers.
-    def force_flush(timeout: nil)
-      deadline = Deadline.new(timeout)
+    # hooks for short-lived and fork-per-job workers. A nil timeout means the
+    # default, not unbounded: a hung endpoint must not stall the worker.
+    def force_flush(timeout: FLUSH_TIMEOUT)
+      deadline = Deadline.new(timeout || FLUSH_TIMEOUT)
       results = []
 
       results << tracer_provider_for_flush.force_flush(timeout: deadline.remaining)
@@ -26,6 +29,42 @@ module Flare
       OpenTelemetry::SDK::Trace::Export::FAILURE
     end
 
+    SHUTDOWN_TIMEOUT = 5       # seconds, total across all of Flare
+    TRACE_SHUTDOWN_TIMEOUT = 1 # seconds; sampled traces are expendable
+
+    # Stop everything Flare runs in the background under one budget, in
+    # priority order. Runs from at_exit in every forked worker, so it must
+    # finish well before the platform's kill deadline even if endpoints hang.
+    # Metrics (a minute of aggregated counts) get first claim on the time;
+    # span/trace processors get what's left, capped at TRACE_SHUTDOWN_TIMEOUT.
+    # A nil timeout means the default, not unbounded: shutdown is always bounded.
+    def shutdown(timeout: SHUTDOWN_TIMEOUT)
+      deadline = Deadline.new(timeout || SHUTDOWN_TIMEOUT)
+
+      shutdown_step("rule manager") { @rule_manager&.stop(timeout: 0) }
+      shutdown_step("metrics") { @metric_flusher&.stop(timeout: [deadline.remaining - TRACE_SHUTDOWN_TIMEOUT, 0].max) }
+
+      traces = Deadline.new([deadline.remaining, TRACE_SHUTDOWN_TIMEOUT].min)
+      if configuration.spans_enabled && @span_processor
+        shutdown_step("span processor") { span_processor.shutdown(timeout: traces.remaining) }
+      end
+      shutdown_step("trace processor") { @trace_span_processor&.shutdown(timeout: traces.remaining) }
+    end
+
+    # Each step is isolated so one failure doesn't skip the rest.
+    def shutdown_step(name)
+      yield
+    rescue => e
+      warn "[Flare] #{name} shutdown error: #{e.message}"
+    end
+
+    def install_shutdown_hook
+      return if @shutdown_hook_installed
+
+      @shutdown_hook_installed = true
+      at_exit { shutdown }
+    end
+
     def tracer_provider_for_flush
       OpenTelemetry.tracer_provider
     end

data/lib/flare/metric_flusher.rb

@@ -116,6 +116,8 @@ module Flare
     # on first span in the new process, or manually from Puma/Unicorn
     # after_fork hooks.
     def after_fork
+      return if @pid == $$
+
       @pid = $$
       @storage.after_fork if @storage.respond_to?(:after_fork)
       initialize_synchronization

data/lib/flare/rule_manager.rb

@@ -60,10 +60,10 @@ module Flare
       self
     end
 
-    def stop
+    def stop(timeout: 1)
       if @timer
         @timer.shutdown
-        @timer.wait_for_termination(1)
+        @timer.wait_for_termination(timeout)
         @timer.kill unless @timer.shutdown?
         @timer = nil
       end

data/lib/flare/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Flare
-  VERSION = "0.4.0"
+  VERSION = "0.4.2"
 end

data/lib/flare.rb

@@ -230,20 +230,7 @@ module Flare
       subscribe_to_notifications
     end
 
-    at_exit do
-      log "Shutting down..."
-      if configuration.spans_enabled && @span_processor
-        span_processor.force_flush
-        span_processor.shutdown
-        log "Span processor flushed and stopped"
-      end
-      if @trace_span_processor
-        @trace_span_processor.force_flush
-        @trace_span_processor.shutdown
-        log "Trace span processor flushed and stopped"
-      end
-      log "Shutdown complete"
-    end
+    install_shutdown_hook
 
     @otel_configured = true
   end
@@ -272,7 +259,7 @@ module Flare
     @rule_manager.start
     log "Rule manager started (poll=#{configuration.tracing_poll_interval}s)"
 
-    at_exit { @rule_manager&.stop }
+    install_shutdown_hook
   end
 
   def setup_tracing_components
@@ -357,7 +344,7 @@ module Flare
       @metric_flusher.start
       log "Metrics flusher started (interval=#{configuration.metrics_flush_interval}s)"
 
-      at_exit { @metric_flusher&.stop }
+      install_shutdown_hook
     else
       log "Metrics submission not configured (missing url or key)"
     end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: flare
 version: !ruby/object:Gem::Version
-  version: 0.4.0
+  version: 0.4.2
 platform: ruby
 authors:
 - John Nunemaker