From 3f07a6b38b3e53c7b562c66ee2a6d2beb344ac1a Mon Sep 17 00:00:00 2001
From: Sergio Cazzolato <sergio.cazzolato@canonical.com>
Date: Thu, 22 Feb 2024 17:57:16 -0300
Subject: [PATCH] send halt-timeout as a tag to allow per backend garbage
 collection

---
 spread/testflinger.go                | 36 ++++++++++++++++++---
 tests/tesflinger/fake_testflinger.py | 48 ++++++++++++++++++++++++++--
 tests/tesflinger/task.yaml           |  4 +++
 3 files changed, 81 insertions(+), 7 deletions(-)

diff --git a/spread/testflinger.go b/spread/testflinger.go
index d226ed89..b074c281 100644
--- a/spread/testflinger.go
+++ b/spread/testflinger.go
@@ -11,6 +11,7 @@ import (
 	"net/url"
 	"os"
 	"regexp"
+	"strings"
 	"time"
 
 	"golang.org/x/net/context"
@@ -65,6 +66,11 @@ type TestFlingerJobResponse struct {
 	JobId string `json:"job_id"`
 }
 
+type TestFlingerJobInfoResponse struct {
+	JobId string  `json:"job_id"`
+	Tags []string `json:"tags"`
+}
+
 type TestFlingerDeviceInfo struct {
 	DeviceIP string `json:"device_ip"`
 }
@@ -142,14 +148,33 @@ func (p *TestFlingerProvider) GarbageCollect() error {
 	// Iterate over all the running instances
 	for _, s := range jobs {
 		jobTimeout := haltTimeout
-		if jobTimeout == 0 {
-			continue
-		}
 		if s.d.JobState == CANCELLED || s.d.JobState == COMPLETE || s.d.JobState == COMPLETED {
 			continue
 		}
 		printf("Checking %s...", s.d.JobId)
 
+		var result TestFlingerJobInfoResponse
+		err := p.do("GET", "/job/" + s.d.JobId , nil, &result)
+		if err != nil {
+			return fmt.Errorf("cannot get instance info: %v", err)
+		}
+
+		// Use specific job timeout if a tag is set with halt-timeout=TIMEOUT
+		for _, tag := range result.Tags {
+			if strings.HasPrefix(tag, "halt-timeout=") {
+				value := strings.SplitAfter(tag, "=")[1]
+				d, err := time.ParseDuration(strings.TrimSpace(value))
+				if err != nil {
+					printf("WARNING: Ignoring bad TestFlinger job %s halt-timeout tag: %q", s.d.JobId, value)
+				} else {
+					jobTimeout = d
+				}
+			}
+		}
+		if jobTimeout == 0 {
+			continue
+		}
+
 		runningTime := now.Sub(s.d.CreatedAt)
 		if runningTime > jobTimeout {
 			printf("Job %s exceeds halt-timeout. Shutting it down...", s.d.JobId)
@@ -222,7 +247,10 @@ func (p *TestFlingerProvider) requestDevice(ctx context.Context, system *System)
 		AllocateData: TestFlingerAllocateData{
 			Allocate: true,
 		},
-		Tags: []string{"spread"},
+		// Tags used are:
+		// 1. spread which is used to find the spread active jobs
+		// 2. halt-timeout=DURATION which is used to determine when a running job has to be cancelled
+		Tags: []string{"spread", "halt-timeout=" + p.backend.HaltTimeout.Duration.String()},
 	}
 
 	var jobRes TestFlingerJobResponse
diff --git a/tests/tesflinger/fake_testflinger.py b/tests/tesflinger/fake_testflinger.py
index 0085db02..3de2e846 100644
--- a/tests/tesflinger/fake_testflinger.py
+++ b/tests/tesflinger/fake_testflinger.py
@@ -112,19 +112,61 @@ def mock_action_delete():
 
     return jsonify({})
 
+@app.route('/v1/job/00000000-0000-0000-0000-000000000004', methods=['GET'])
+def mock_job_tags_4():
+    return jsonify({
+      "allocate_data": {},
+      "allocation_timeout": 0,
+      "firmware_update_data": {},
+      "global_timeout": 0,
+      "job_id": "00000000-0000-0000-0000-000000000004",
+      "job_queue": "queue",
+      "name": "name",
+      "output_timeout": 0,
+      "parent_job_id": "job",
+      "provision_data": {},
+      "reserve_data": {},
+      "tags": ["spread", "halt-timeout=4h"],
+      "test_data": {}
+      })
+
 
 @app.route('/v1/job/00000000-0000-0000-0000-000000000005/action', methods=['POST'])
-def mock_action_error_1():
+def mock_action_error_5():
     abort(500)
 
+@app.route('/v1/job/00000000-0000-0000-0000-000000000005', methods=['GET'])
+def mock_job_tags_5():
+    return jsonify({
+      "allocate_data": {},
+      "allocation_timeout": 0,
+      "firmware_update_data": {},
+      "global_timeout": 0,
+      "job_id": "00000000-0000-0000-0000-000000000005",
+      "job_queue": "queue",
+      "name": "name",
+      "output_timeout": 0,
+      "parent_job_id": "job",
+      "provision_data": {},
+      "reserve_data": {},
+      "tags": ["spread", "halt-timeout=4h"],
+      "test_data": {}
+    })
 
 @app.route('/v1/job/00000000-0000-0000-0000-000000000006/action', methods=['POST'])
-def mock_action_error_2():
+def mock_action_error_6():
     abort(500)
 
+@app.route('/v1/job/00000000-0000-0000-0000-000000000006', methods=['GET'])
+def mock_job_error_6():
+    abort(500)
 
 @app.route('/v1/job/00000000-0000-0000-0000-000000000007/action', methods=['POST'])
-def mock_action_error_3():
+def mock_action_error_7():
+    abort(500)
+
+@app.route('/v1/job/00000000-0000-0000-0000-000000000007', methods=['GET'])
+def mock_job_error_7():
     abort(500)
 
 
diff --git a/tests/tesflinger/task.yaml b/tests/tesflinger/task.yaml
index 886ccd86..68f9e17a 100644
--- a/tests/tesflinger/task.yaml
+++ b/tests/tesflinger/task.yaml
@@ -75,6 +75,10 @@ execute: |
     stop_mock
 
     # Garbage collection scenario
+    # Job 4: has to be deleted (it is active and halt-timeout exceeded)
+    # Job 5: has not to be deleted (it is active but halt-timeout not exceeded)
+    # Job 6: has not to be deleted (it is not active)
+    # Job 7: has not to be deleted (it is not active)
     start_mock 00000000-0000-0000-0000-000000000004
     spread -gc &> /tmp/task.out
     cat /tmp/task.out | grep 'Checking 00000000-0000-0000-0000-000000000004'