Mark workflow as failed when trigger fails (#740)

Author: nileshgr

app/components/live_release/build_component.rb

@@ -10,6 +10,7 @@ class LiveRelease::BuildComponent < BaseComponent
     unavailable: {text: "Build unavailable", status: :failure},
     started: {text: "Workflow running", status: :ongoing, kind: :spinner_pill},
     failed: {text: "Workflow failed", status: :failure},
+    trigger_failed: {text: "Failed to trigger workflow", status: :failure},
     halted: {text: "Workflow halted", status: :failure},
     finished: {text: "Workflow finished", status: :ongoing},
     cancelled: {text: "Workflow cancelled", status: :inert},

app/components/live_release/pre_prod_release/current_release_component.html.erb

@@ -62,6 +62,12 @@
           <%= render BadgeComponent.new(**status) %>
         </div>
 
+        <% if workflow_run.trigger_failed? %>
+          <div class="self-start">
+            <%= render AlertComponent.new(type: :error, title: raw(workflow_run.trigger_failed_reason), full_screen: false) %>
+          </div>
+        <% end %>
+
         <% if build.present? %>
           <%= render LiveRelease::BuildComponent.new(build) %>
         <% end %>

app/jobs/workflow_runs/trigger_job.rb

@@ -1,11 +1,22 @@
 class WorkflowRuns::TriggerJob < ApplicationJob
   queue_as :high
 
+  TRIGGER_FAILED_REASONS = [
+    :workflow_parameter_not_provided,
+    :workflow_dispatch_missing
+  ]
+
   def perform(workflow_run_id, retrigger = false)
     workflow_run = WorkflowRun.find(workflow_run_id)
     return unless workflow_run.active?
     return unless workflow_run.may_initiated?
 
     workflow_run.trigger!(retrigger:)
+  rescue Installations::Error => err
+    if err.reason.in?(TRIGGER_FAILED_REASONS)
+      workflow_run.trigger_failed!(err)
+    else
+      raise
+    end
   end
 end

app/libs/coordinators.rb

@@ -80,6 +80,10 @@ def self.beta_release_is_finished!(build)
     def self.production_release_is_complete!(release_platform_run)
       Coordinators::FinishPlatformRun.call(release_platform_run)
     end
+
+    def self.workflow_run_trigger_failed!(workflow_run)
+      workflow_run.triggering_release.fail!
+    end
   end
 
   module Actions

app/libs/installations/github/error.rb

@@ -58,6 +58,14 @@ class Github::Error < Installations::Error
       {
         message_matcher: /Merge conflict/i,
         decorated_reason: :merge_conflict
+      },
+      {
+        message_matcher: /Required input\s+.?\w+.?\s+not provided/,
+        decorated_reason: :workflow_parameter_not_provided
+      },
+      {
+        message_matcher: /Workflow does not have 'workflow_dispatch' trigger/,
+        decorated_reason: :workflow_dispatch_missing
       }
     ]
 

app/models/workflow_run.rb

@@ -41,6 +41,7 @@ class WorkflowRun < ApplicationRecord
     failed
     halted
     finished
+    trigger_failed
   ]
 
   KINDS = {
@@ -52,6 +53,7 @@ class WorkflowRun < ApplicationRecord
     created: "created",
     triggering: "triggering",
     triggered: "triggered",
+    trigger_failed: "trigger_failed",
     unavailable: "unavailable",
     started: "started",
     failed: "failed",
@@ -64,8 +66,8 @@ class WorkflowRun < ApplicationRecord
 
   NOT_STARTED = [:created]
   IN_PROGRESS = [:triggering, :triggered, :started]
-  WORKFLOW_IMMUTABLE = %w[unavailable failed halted finished cancelled cancelling cancelled_before_start]
-  FAILED_STATES = %w[failed halted unavailable cancelled cancelled_before_start cancelling]
+  WORKFLOW_IMMUTABLE = %w[unavailable failed halted finished cancelled cancelling cancelled_before_start trigger_failed]
+  FAILED_STATES = %w[failed halted unavailable cancelled cancelled_before_start cancelling trigger_failed]
 
   enum :status, STATES
   enum :kind, KINDS
@@ -111,6 +113,10 @@ class WorkflowRun < ApplicationRecord
       transitions from: NOT_STARTED, to: :cancelled_before_start
       transitions from: :cancelling, to: :cancelled
     end
+
+    event :trigger_failed, after_commit: :on_trigger_fail! do
+      transitions from: :triggering, to: :trigger_failed
+    end
   end
 
   def self.create_and_trigger!(workflow_config, triggering_release, commit, release_platform_run)
@@ -198,6 +204,11 @@ def notification_params
     )
   end
 
+  def trigger_failed_reason
+    last_error = passports.where(reason: :trigger_failed, kind: :error).last
+    last_error&.message
+  end
+
   private
 
   def trigger_external_run!
@@ -266,6 +277,12 @@ def on_fail!
     notify!("The workflow run has failed!", :workflow_run_failed, notification_params)
   end
 
+  def on_trigger_fail!(error)
+    event_stamp!(reason: :trigger_failed, kind: :error, data: stamp_data.merge(error_message: error.message, error_reason: error.reason))
+    notify!("Failed to trigger the workflow run!", :workflow_trigger_failed, notification_params)
+    Signal.workflow_run_trigger_failed!(self)
+  end
+
   def on_halt!
     event_stamp!(reason: :halted, kind: :error, data: stamp_data)
     notify!("The workflow run has been halted!", :workflow_run_halted, notification_params)

config/locales/passport/en.yml

@@ -39,6 +39,7 @@ en:
       halted_html: 'The %{kind} workflow run <a class="emphasize-link" href="%{url}">#%{ref}</a> was halted for <a class="emphasize-link" href="%{commit_url}">#%{commit_sha}</a> to build <span class="emphasize">%{version_name} (%{build_number})</span>'
       unavailable_html: 'The %{kind} workflow run could not be found for <a class="emphasize-link" href="%{commit_url}">#%{commit_sha}</a> to build <span class="emphasize">%{version_name} (%{build_number})</span>'
       finished_html: 'The %{kind} workflow run <a class="emphasize-link" href="%{url}">#%{ref}</a> completed for <a class="emphasize-link" href="%{commit_url}">#%{commit_sha}</a> to build <span class="emphasize">%{version_name} (%{build_number})</span>'
+      trigger_failed_html: 'The %{kind} workflow run could not be triggered for <a class="emphasize-link" href="%{commit_url}">#%{commit_sha}</a> to build <span class="emphasize">%{version_name} (%{build_number})</span>. Error: <span class="emphasize">%{error_message}.</span>'
     google_firebase_submission:
       triggered_html: 'Sending <span class="emphasize">%{version} (%{build_number})</span> to Google Firebase groups <span class="emphasize">%{channels}</span>'
       finished_html: 'Sent <span class="emphasize">%{version} (%{build_number})</span> to Google Firebase groups <span class="emphasize">%{channels}</span>'

spec/jobs/workflow_runs/trigger_job_spec.rb

@@ -0,0 +1,67 @@
+# frozen_string_literal: true
+
+require "rails_helper"
+
+RSpec.describe WorkflowRuns::TriggerJob do
+  let(:workflow_run) { create(:workflow_run, :triggering) }
+
+  before do
+    allow(WorkflowRun).to receive(:find).and_return(workflow_run)
+  end
+
+  context "when trigger does not cause error" do
+    before do
+      allow(workflow_run).to receive(:trigger!)
+    end
+
+    it "triggers successfully" do
+      described_class.new.perform(workflow_run.id)
+      expect(workflow_run).to have_received(:trigger!)
+    end
+  end
+
+  context "when trigger results in error" do
+    let(:workflow_run) { create(:workflow_run, :triggering) }
+
+    shared_examples "with known error" do |error|
+      before do
+        allow(workflow_run).to receive(:trigger!).and_raise(error)
+      end
+
+      context "when error is #{error.message}" do
+        it "changes state of workflow_run to trigger_failed" do
+          described_class.new.perform(workflow_run.id)
+          expect(workflow_run.reload.status).to eq("trigger_failed")
+        end
+      end
+    end
+
+    include_examples "with known error", Installations::Github::Error.new(
+      OpenStruct.new(
+        response_body: {message: "Workflow does not have 'workflow_dispatch' trigger"}.to_json
+      )
+    )
+    include_examples "with known error", Installations::Github::Error.new(
+      OpenStruct.new(
+        response_body: {message: "Required input 'parameter_X' not provided"}.to_json
+      )
+    )
+
+    context "when error is unknown" do
+      before do
+        err = Installations::Error.new("Some Error", reason: :unknown_failure)
+        allow(workflow_run).to receive(:trigger!).and_raise(err)
+      end
+
+      it "does not change state of workflow_run to trigger_failed" do
+        begin
+          described_class.new.perform(workflow_run.id)
+        rescue Installations::Error
+          nil
+        end
+
+        expect(workflow_run.reload.status).not_to eq("trigger_failed")
+      end
+    end
+  end
+end