Adding Node, Cluster and ExitCode to Jobs (#3442)

* Adding Node, Cluster and ExitCode to Jobs (#80) * Adding Node, Cluster and ExitCode to Jobs * Wip fixing tests * Fixing tests * Linting * Linting * Add unit tests * Update exit code * Go Linting --------- Co-authored-by: Mia Mijovic <Mia.Mijovic@gresearch.co.uk>
armadaproject · Mar 4, 2024 · d76d5ad · d76d5ad
1 parent e5a0ce5
commit d76d5ad
Show file tree

Hide file tree

Showing 8 changed files with 134 additions and 1 deletion.
diff --git a/internal/lookout/ui/src/models/lookoutV2Models.ts b/internal/lookout/ui/src/models/lookoutV2Models.ts
@@ -78,6 +78,9 @@ export type Job = {
   lastActiveRunId?: string
   lastTransitionTime: string
   cancelReason?: string
+  node?: string
+  cluster?: string
+  exitCode?: number
 }
 
 export type JobKey = keyof Job

diff --git a/internal/lookout/ui/src/utils/jobsTableColumns.tsx b/internal/lookout/ui/src/utils/jobsTableColumns.tsx
@@ -51,6 +51,9 @@ export enum StandardColumnId {
   SelectorCol = "selectorCol",
 
   Count = "jobCount",
+  Node = "node",
+  Cluster = "cluster",
+  ExitCode = "exitCode",
 }
 
 export const ANNOTATION_COLUMN_PREFIX = "annotation_"
@@ -394,6 +397,30 @@ export const JOB_COLUMNS: JobTableColumn[] = [
       enableSorting: true,
     },
   }),
+  accessorColumn({
+    id: StandardColumnId.Node,
+    accessor: "node",
+    displayName: "Node",
+    additionalOptions: {
+      size: 200,
+    },
+  }),
+  accessorColumn({
+    id: StandardColumnId.Cluster,
+    accessor: "cluster",
+    displayName: "Cluster",
+    additionalOptions: {
+      size: 200,
+    },
+  }),
+  accessorColumn({
+    id: StandardColumnId.ExitCode,
+    accessor: "exitCode",
+    displayName: "Exit Code",
+    additionalOptions: {
+      size: 100,
+    },
+  }),
 ]
 
 export const DEFAULT_COLUMNS_TO_DISPLAY: Set<ColumnId> = new Set([

diff --git a/internal/lookoutv2/conversions/convert.go b/internal/lookoutv2/conversions/convert.go
@@ -37,6 +37,9 @@ func ToSwaggerJob(job *model.Job) *models.Job {
 		State:              job.State,
 		Submitted:          strfmt.DateTime(job.Submitted),
 		CancelReason:       job.CancelReason,
+		Node:               job.Node,
+		Cluster:            job.Cluster,
+		ExitCode:           job.ExitCode,
 	}
 }
 

diff --git a/internal/lookoutv2/gen/models/job.go b/internal/lookoutv2/gen/models/job.go
diff --git a/internal/lookoutv2/model/model.go b/internal/lookoutv2/model/model.go
@@ -40,6 +40,9 @@ type Job struct {
 	State              string
 	Submitted          time.Time
 	CancelReason       *string
+	Node               *string
+	Cluster            string
+	ExitCode           *int32
 }
 
 // PostgreSQLTime is a wrapper around time.Time that converts to UTC when

diff --git a/internal/lookoutv2/repository/getjobs.go b/internal/lookoutv2/repository/getjobs.go
@@ -125,6 +125,7 @@ func (r *SqlGetJobsRepository) getJobs(ctx *armadacontext.Context, filters []*mo
 			log.WithError(err).Error("failed getting run rows")
 			return err
 		}
+
 		annotationRows, err = makeAnnotationRows(ctx, tx, tempTableName)
 		if err != nil {
 			log.WithError(err).Error("failed getting annotation rows")
@@ -135,7 +136,6 @@ func (r *SqlGetJobsRepository) getJobs(ctx *armadacontext.Context, filters []*mo
 	if err != nil {
 		return nil, err
 	}
-
 	jobs, err := rowsToJobs(jobRows, runRows, annotationRows)
 	if err != nil {
 		return nil, err
@@ -201,6 +201,13 @@ func (r *SqlGetJobsRepository) getJobsJsonb(ctx *armadacontext.Context, filters
 					return err
 				}
 			}
+			if len(job.Runs) > 0 {
+				lastRun := job.Runs[len(job.Runs)-1] // Get the last run
+				job.Node = lastRun.Node
+				job.Cluster = lastRun.Cluster
+				job.ExitCode = lastRun.ExitCode
+
+			}
 			jobs = append(jobs, job)
 		}
 		return nil
@@ -251,6 +258,13 @@ func rowsToJobs(jobRows []*jobRow, runRows []*runRow, annotationRows []*annotati
 	for i, jobId := range orderedJobIds {
 		job := jobMap[jobId]
 		sortRuns(job.Runs)
+		if len(job.Runs) > 0 {
+			lastRun := job.Runs[len(job.Runs)-1] // Get the last run
+			job.Node = lastRun.Node
+			job.Cluster = lastRun.Cluster
+			job.ExitCode = lastRun.ExitCode
+
+		}
 		jobs[i] = job
 	}
 

diff --git a/internal/lookoutv2/repository/getjobs_test.go b/internal/lookoutv2/repository/getjobs_test.go
@@ -1997,3 +1997,69 @@ func TestGetJobsActiveJobSet(t *testing.T) {
 	})
 	require.NoError(t, err)
 }
+
+func TestGetJobsWithLatestRunDetails(t *testing.T) {
+	err := withGetJobsSetup(func(converter *instructions.InstructionConverter, store *lookoutdb.LookoutDb, repo *SqlGetJobsRepository) error {
+		runIdLatest := uuid.NewString()
+		// Simulate job submission and multiple runs, with the latest run being successful
+		NewJobSimulator(converter, store).
+			Submit(queue, jobSet, owner, namespace, baseTime, basicJobOpts).
+			Pending(uuid.NewString(), "first-cluster", baseTime).
+			Running(uuid.NewString(), "first-node", baseTime.Add(time.Minute)).
+			Pending(runIdLatest, "latest-cluster", baseTime.Add(2*time.Minute)).
+			Running(runIdLatest, "latest-node", baseTime.Add(3*time.Minute)).
+			RunSucceeded(runIdLatest, baseTime.Add(4*time.Minute)).
+			Build().
+			Job()
+
+		result, err := repo.GetJobs(armadacontext.TODO(), []*model.Filter{}, false, &model.Order{}, 0, 10)
+		require.NoError(t, err)
+		require.Len(t, result.Jobs, 1)
+
+		// Adjusting assertions to dereference pointer fields
+		if assert.NotNil(t, result.Jobs[0].Node) {
+			assert.Equal(t, "latest-node", *result.Jobs[0].Node)
+		}
+		if assert.NotNil(t, result.Jobs[0].ExitCode) {
+			assert.Equal(t, int32(0), *result.Jobs[0].ExitCode)
+		}
+		if assert.NotNil(t, result.Jobs[0].Cluster) {
+			assert.Equal(t, "latest-cluster", result.Jobs[0].Cluster)
+		}
+
+		return nil
+	})
+	require.NoError(t, err)
+}
+
+func TestGetJobsWithSpecificRunDetails(t *testing.T) {
+	err := withGetJobsSetup(func(converter *instructions.InstructionConverter, store *lookoutdb.LookoutDb, repo *SqlGetJobsRepository) error {
+		runIdSpecific := uuid.NewString()
+		// Simulate job submission and a specific failed run
+		NewJobSimulator(converter, store).
+			Submit(queue, jobSet, owner, namespace, baseTime, basicJobOpts).
+			Pending(runIdSpecific, "specific-cluster", baseTime).
+			Running(runIdSpecific, "specific-node", baseTime.Add(time.Minute)).
+			RunFailed(runIdSpecific, "specific-node", 2, "Specific failure message", baseTime.Add(2*time.Minute)).
+			Build().
+			Job()
+
+		result, err := repo.GetJobs(armadacontext.TODO(), []*model.Filter{}, false, &model.Order{}, 0, 10)
+		require.NoError(t, err)
+		require.Len(t, result.Jobs, 1)
+
+		// Adjusting assertions to dereference pointer fields
+		if assert.NotNil(t, result.Jobs[0].Node) {
+			assert.Equal(t, "specific-node", *result.Jobs[0].Node)
+		}
+		if assert.NotNil(t, result.Jobs[0].ExitCode) {
+			assert.Equal(t, int32(2), *result.Jobs[0].ExitCode)
+		}
+		if assert.NotNil(t, result.Jobs[0].Cluster) {
+			assert.Equal(t, "specific-cluster", result.Jobs[0].Cluster)
+		}
+
+		return nil
+	})
+	require.NoError(t, err)
+}
diff --git a/internal/lookoutv2/repository/util.go b/internal/lookoutv2/repository/util.go
@@ -225,6 +225,7 @@ func (js *JobSimulator) Pending(runId string, cluster string, timestamp time.Tim
 	js.job.LastActiveRunId = &runId
 	js.job.LastTransitionTime = ts
 	js.job.State = string(lookout.JobPending)
+	js.job.Cluster = cluster
 	rp := &runPatch{
 		runId:       runId,
 		cluster:     &cluster,
@@ -264,6 +265,7 @@ func (js *JobSimulator) Running(runId string, node string, timestamp time.Time)
 	js.job.LastActiveRunId = &runId
 	js.job.LastTransitionTime = ts
 	js.job.State = string(lookout.JobRunning)
+	js.job.Node = &node
 	updateRun(js.job, &runPatch{
 		runId:       runId,
 		jobRunState: lookout.JobRunRunning,
@@ -613,6 +615,9 @@ func timestampOrNow(timestamp time.Time) time.Time {
 }
 
 func updateRun(job *model.Job, patch *runPatch) {
+	if patch.exitCode != nil {
+		job.ExitCode = patch.exitCode
+	}
 	for _, run := range job.Runs {
 		if run.RunId == patch.runId {
 			patchRun(run, patch)