Add process name and ID tags to GVL metrics

When reporting the GVL metrics, which measure the time spent by the Ruby VM in its global VM lock, and as such are per-process metrics, add the process name and process ID as tags, alongside the existing hostname tag. As the process name can change during the lifetime of the process, store it when initialising the GVL probe, to prevent potentially reporting the same metric with different tag combinations. To attempt to account for custom process names, keep only the first word of the process name, after removing path prefixes from it. Fixes #1106.
appsignal · Jun 27, 2024 · e9aa060 · e9aa060
1 parent d299589
commit e9aa060
Show file tree

Hide file tree

Showing 3 changed files with 110 additions and 5 deletions.
diff --git a/.changesets/report-gvl-metrics-per-process.md b/.changesets/report-gvl-metrics-per-process.md
@@ -0,0 +1,6 @@
+---
+bump: patch
+type: change
+---
+
+Report Global VM Lock metrics per process. In addition to the existing `hostname` tag, add `process_name` and `process_id` tags to the `gvl_global_timer` and `gvl_waiting_threads` metrics emitted by the [GVL probe](https://docs.appsignal.com/ruby/integrations/global-vm-lock.html), allowing these metrics to be tracked in a per-process basis.
diff --git a/lib/appsignal/probes/gvl.rb b/lib/appsignal/probes/gvl.rb
@@ -25,6 +25,11 @@ def initialize(appsignal: Appsignal, gvl_tools: ::GVLTools)
         Appsignal.internal_logger.debug("Initializing GVL probe")
         @appsignal = appsignal
         @gvl_tools = gvl_tools
+
+        # Store the process name and ID at initialization time
+        # to avoid picking up changes to the process name at runtime
+        @process_name = File.basename($PROGRAM_NAME).split.first || "[unknown process]"
+        @process_id = Process.pid
       end
 
       def call
@@ -39,13 +44,30 @@ def probe_global_timer
         gauge_delta :gvl_global_timer, monotonic_time_ns do |time_delta_ns|
           if time_delta_ns > 0
             time_delta_ms = time_delta_ns / 1_000_000
-            set_gauge_with_hostname("gvl_global_timer", time_delta_ms)
+            set_gauges_with_hostname_and_process(
+              "gvl_global_timer",
+              time_delta_ms
+            )
           end
         end
       end
 
       def probe_waiting_threads
-        set_gauge_with_hostname("gvl_waiting_threads", @gvl_tools::WaitingThreads.count)
+        set_gauges_with_hostname_and_process(
+          "gvl_waiting_threads",
+          @gvl_tools::WaitingThreads.count
+        )
+      end
+
+      def set_gauges_with_hostname_and_process(name, value)
+        set_gauge_with_hostname(name, value, {
+          :process_name => @process_name,
+          :process_id => @process_id
+        })
+
+        # Also set the gauge without the process name and ID for
+        # compatibility with existing automated dashboards
+        set_gauge_with_hostname(name, value)
       end
     end
   end

diff --git a/spec/lib/appsignal/probes/gvl_spec.rb b/spec/lib/appsignal/probes/gvl_spec.rb
@@ -4,6 +4,13 @@
 
   let(:hostname) { "some-host" }
 
+  around do |example|
+    real_program_name = $PROGRAM_NAME
+    example.run
+  ensure
+    $PROGRAM_NAME = real_program_name
+  end
+
   def gauges_for(metric)
     gauges = appsignal_mock.gauges.select do |gauge|
       gauge[0] == metric
@@ -14,7 +21,7 @@ def gauges_for(metric)
     end
   end
 
-  after(:each) { FakeGVLTools.reset }
+  after { FakeGVLTools.reset }
 
   it "gauges the global timer delta" do
     FakeGVLTools::GlobalTimer.monotonic_time = 100_000_000
@@ -26,6 +33,11 @@ def gauges_for(metric)
     probe.call
 
     expect(gauges_for("gvl_global_timer")).to eq [
+      [200, {
+        :hostname => hostname,
+        :process_name => "rspec",
+        :process_id => Process.pid
+      }],
       [200, { :hostname => hostname }]
     ]
   end
@@ -58,7 +70,7 @@ def gauges_for(metric)
   end
 
   context "when the waiting threads count is enabled" do
-    before(:each) do
+    before do
       FakeGVLTools::WaitingThreads.enabled = true
     end
 
@@ -67,13 +79,18 @@ def gauges_for(metric)
       probe.call
 
       expect(gauges_for("gvl_waiting_threads")).to eq [
+        [3, {
+          :hostname => hostname,
+          :process_name => "rspec",
+          :process_id => Process.pid
+        }],
         [3, { :hostname => hostname }]
       ]
     end
   end
 
   context "when the waiting threads count is disabled" do
-    before(:each) do
+    before do
       FakeGVLTools::WaitingThreads.enabled = false
     end
 
@@ -84,4 +101,64 @@ def gauges_for(metric)
       expect(gauges_for("gvl_waiting_threads")).to be_empty
     end
   end
+
+  context "when the process name is a custom value" do
+    before do
+      FakeGVLTools::WaitingThreads.enabled = true
+    end
+
+    it "uses only the first word as the process name" do
+      $PROGRAM_NAME = "sidekiq 7.1.6 app [0 of 5 busy]"
+      probe.call
+
+      expect(gauges_for("gvl_waiting_threads")).to eq [
+        [0, {
+          :hostname => hostname,
+          :process_name => "sidekiq",
+          :process_id => Process.pid
+        }],
+        [0, { :hostname => hostname }]
+      ]
+    end
+  end
+
+  context "when the process name is a path" do
+    before do
+      FakeGVLTools::WaitingThreads.enabled = true
+    end
+
+    it "uses only the binary name as the process name" do
+      $PROGRAM_NAME = "/foo/folder with spaces/bin/rails"
+      probe.call
+
+      expect(gauges_for("gvl_waiting_threads")).to eq [
+        [0, {
+          :hostname => hostname,
+          :process_name => "rails",
+          :process_id => Process.pid
+        }],
+        [0, { :hostname => hostname }]
+      ]
+    end
+  end
+
+  context "when the process name is an empty string" do
+    before do
+      FakeGVLTools::WaitingThreads.enabled = true
+    end
+
+    it "uses [unknown process] as the process name" do
+      $PROGRAM_NAME = ""
+      probe.call
+
+      expect(gauges_for("gvl_waiting_threads")).to eq [
+        [0, {
+          :hostname => hostname,
+          :process_name => "[unknown process]",
+          :process_id => Process.pid
+        }],
+        [0, { :hostname => hostname }]
+      ]
+    end
+  end
 end