fix(github): catch network ReadTimeout

Gemfile.lock

@@ -1,7 +1,7 @@
 PATH
   remote: .
   specs:
-    entitlements-github-plugin (1.2.5)
+    entitlements-github-plugin (1.2.6)
       contracts (~> 0.17.0)
       faraday (~> 2.0)
       faraday-net_http_persistent (~> 2.3)

lib/entitlements/service/github.rb

@@ -392,23 +392,29 @@ def graphql_http_post_real(query)
             return { code: response.code.to_i, data: { "body" => response.body } }
           end
 
-          begin
-            data = JSON.parse(response.body)
-            if data.key?("errors")
-              # Synthesized 500 below triggers a retry, so log at WARN. Note: some GraphQL
-              # `errors` are permanent (bad query, auth, schema). The retry wrapper's final
-              # "Giving up" ERROR will surface persistent cases.
-              Entitlements.logger.warn "Errors reported: #{data['errors'].inspect}"
-              return { code: 500, data: }
-            end
-            { code: response.code.to_i, data: }
-          rescue JSON::ParserError => e
-            # JSON parse errors mean the API returned something we can't interpret. The
-            # synthesized 500 below triggers a retry, but the cause is more likely a real
-            # protocol/server problem than a transient network blip, so log at ERROR.
-            Entitlements.logger.error "#{e.class} #{e.message}: #{response.body.inspect}"
-            { code: 500, data: { "body" => response.body } }
+          data = JSON.parse(response.body)
+          if data.key?("errors")
+            # Synthesized 500 below triggers a retry, so log at WARN. Note: some GraphQL
+            # `errors` are permanent (bad query, auth, schema). The retry wrapper's final
+            # "Giving up" ERROR will surface persistent cases.
+            Entitlements.logger.warn "Errors reported: #{data['errors'].inspect}"
+            return { code: 500, data: }
           end
+
+          { code: response.code.to_i, data: }
+
+        rescue Net::ReadTimeout => e
+          # Defend against network blips, those can be safely retried via the
+          # synthesized 500 below, so log at WARN.
+          reset_graphql_http
+          Entitlements.logger.warn "#{e.class} #{e.message}"
+          { code: 500, data: nil }
+        rescue JSON::ParserError => e
+          # JSON parse errors mean the API returned something we can't interpret. The
+          # synthesized 500 below triggers a retry, but the cause is more likely a real
+          # protocol/server problem than a transient network blip, so log at ERROR.
+          Entitlements.logger.error "#{e.class} #{e.message}: #{response.body.inspect}"
+          { code: 500, data: { "body" => response.body } }
         rescue => e
           # Catch-all for any unexpected exception (network blip OR local code bug).
           # We retry below via the synthesized 500, but log at ERROR because this

lib/version.rb

@@ -2,6 +2,6 @@
 
 module Entitlements
   module Version
-    VERSION = "1.2.5"
+    VERSION = "1.2.6"
   end
 end

spec/unit/entitlements/service/github_spec.rb

@@ -340,9 +340,9 @@
       expect(replacement_http).to receive(:start)
       expect(replacement_http).to receive(:request).and_return(response)
       expect_any_instance_of(Object).to receive(:sleep).with(1).once
-      expect(logger).to receive(:error).
-        with("Caught Net::ReadTimeout POSTing to https://github.fake/api/v3/graphql: Net::ReadTimeout with \"execution expired\"")
-      expect(logger).to receive(:warn).
+      expect(logger).to receive(:warn).ordered.
+        with("Net::ReadTimeout Net::ReadTimeout with \"execution expired\"")
+      expect(logger).to receive(:warn).ordered.
         with("GraphQL failed on try 1 of 3. Will retry.")
 
       expect(subject.send(:graphql_http_post, query)).to eq(code: 200, data: answer)
@@ -450,6 +450,14 @@
       expect(response).to eq(code: 200, data: answer)
     end
 
+    it "logs at warn and returns a retryable response for a read timeout" do
+      timeout = Net::ReadTimeout.new
+      stub_request(:post, "https://github.fake/api/v3/graphql").to_raise(timeout)
+      expect(logger).to receive(:warn).with("#{timeout.class} #{timeout.message}")
+      response = subject.send(:graphql_http_post_real, "nonsense")
+      expect(response).to eq(code: 500, data: nil)
+    end
+
     it "logs at error and returns code=500 and exception message for an unhandled exception" do
       exc = StandardError.new("Oh no you don't")
       stub_request(:post, "https://github.fake/api/v3/graphql").to_raise(exc)