Fix race condition in ProcessesSync causing SSH key mismatch

lib/cloud_controller/diego/processes_sync.rb

@@ -21,15 +21,15 @@ def sync
         @bump_freshness = true
         diego_lrps = bbs_apps_client.fetch_scheduling_infos.index_by { |d| d.desired_lrp_key.process_guid }
         logger.info('fetched-scheduling-infos')
-        to_desire = []
+        to_desire = {}
         to_update = {}
         batched_processes_for_sync do |processes|
           processes.each do |process|
             process_guid = ProcessGuid.from_process(process)
             diego_lrp = diego_lrps.delete(process_guid)
 
             if diego_lrp.nil?
-              to_desire.append(process.id)
+              to_desire[process.id] = process_guid
             elsif process.updated_at.to_f.to_s != diego_lrp.annotation
               to_update[process.id] = diego_lrp
             end
@@ -87,7 +87,10 @@ def process_workpool_exceptions(exceptions)
       def update_lrps(to_update)
         batched_processes(to_update.keys) do |processes|
           processes.each do |process|
-            workpool.submit(process, to_update[process.id]) do |p, l|
+            existing_lrp = to_update[process.id]
+            next if ProcessGuid.from_process(process) != existing_lrp.desired_lrp_key.process_guid
+
+            workpool.submit(process, existing_lrp) do |p, l|
               logger.info('updating-lrp', process_guid: p.guid, app_guid: p.app_guid)
               bbs_apps_client.update_app(p, l)
               logger.info('update-lrp', process_guid: p.guid)
@@ -97,8 +100,11 @@ def update_lrps(to_update)
       end
 
       def desire_lrps(to_desire)
-        batched_processes(to_desire) do |processes|
+        batched_processes(to_desire.keys) do |processes|
           processes.each do |process|
+            desired_process_guid = to_desire[process.id]
+            next if ProcessGuid.from_process(process) != desired_process_guid
+
             workpool.submit(process) do |p|
               logger.info('desiring-lrp', process_guid: p.guid, app_guid: p.app_guid)
               bbs_apps_client.desire_app(p)

spec/isolated_specs/processes_sync_spec.rb

@@ -121,6 +121,22 @@ module Diego
             expect(bbs_apps_client).to have_received(:bump_freshness).once
           end
 
+          context 'when the stale process gets a new version between fetching for sync (only guid + version + created_at) and fetching for actual update (full model)' do
+            before do
+              allow(subject).to receive(:update_lrps).and_wrap_original do |method, *args|
+                stale_process.set_new_version
+                stale_process.save
+                method.call(*args)
+              end
+            end
+
+            it 'skips the (obsolete) update' do
+              allow(bbs_apps_client).to receive(:update_app)
+              subject.sync
+              expect(bbs_apps_client).not_to have_received(:update_app)
+            end
+          end
+
           context 'when the process is a cnb app' do
             let(:app) { AppModel.make(:cnb, droplet: DropletModel.make(:cnb)) }
             let!(:stale_process) { ProcessModel.make(:cnb, state: 'STARTED', app: app) }
@@ -253,6 +269,22 @@ module Diego
             expect(bbs_apps_client).to have_received(:bump_freshness).once
           end
 
+          context 'when the missing process gets a new version between fetching for sync (only guid + version + created_at) and fetching for actual creation (full model)' do
+            before do
+              allow(subject).to receive(:desire_lrps).and_wrap_original do |method, *args|
+                missing_process.set_new_version
+                missing_process.save
+                method.call(*args)
+              end
+            end
+
+            it 'skips the (obsolete) creation' do
+              allow(bbs_apps_client).to receive(:desire_app)
+              subject.sync
+              expect(bbs_apps_client).not_to have_received(:desire_app)
+            end
+          end
+
           context 'when desiring app fails' do
             # bbs_apps_client will raise ApiErrors as of right now, we should think about factoring that out so that
             # the background job doesn't have to deal with API concerns