feat: add network monitors (ICMP/TCP/HTTP/DNS) (#2266)

Co-authored-by: xiaomiku01 <xiaomiku01@outlook.com>
Co-authored-by: henrygd <hank@henrygd.me>

internal/records/monitor_averaging_test.go

@@ -0,0 +1,226 @@
+//go:build testing
+
+package records_test
+
+import (
+	"testing"
+	"time"
+
+	monitorEntity "github.com/henrygd/beszel/internal/entities/monitor"
+	"github.com/henrygd/beszel/internal/records"
+	"github.com/henrygd/beszel/internal/tests"
+
+	"github.com/pocketbase/dbx"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestAverageMonitorStats(t *testing.T) {
+	hub, err := tests.NewTestHub(t.TempDir())
+	require.NoError(t, err)
+	defer hub.Cleanup()
+
+	collection, err := hub.FindCachedCollectionByNameOrId("network_monitor_stats")
+	require.NoError(t, err)
+	assert.Nil(t, collection.Fields.GetByName("res_avg"))
+	assert.Nil(t, collection.Fields.GetByName("loss"))
+
+	rm := records.NewRecordManager(hub)
+	user, err := tests.CreateUser(hub, "monitor-avg@example.com", "testtesttest")
+	require.NoError(t, err)
+	sys, err := tests.CreateRecord(hub, "systems", map[string]any{
+		"name":   "monitor-avg-system",
+		"host":   "localhost",
+		"port":   "45876",
+		"status": "up",
+		"users":  []string{user.Id},
+	})
+	require.NoError(t, err)
+	monitor, err := tests.CreateRecord(hub, "network_monitors", map[string]any{
+		"system":   sys.Id,
+		"name":     "cloudflare",
+		"target":   "1.1.1.1",
+		"protocol": "icmp",
+		"interval": 30,
+		"enabled":  true,
+	})
+	require.NoError(t, err)
+
+	created := time.Now().UnixMilli()
+	// Unequal probe counts must weight both latency and loss.
+	recordA, err := tests.CreateRecord(hub, "network_monitor_stats", map[string]any{
+		"system":      sys.Id,
+		"monitor":     monitor.Id,
+		"type":        "1m",
+		"created":     created,
+		"res_min":     5,
+		"res_max":     20,
+		"total_count": 6, "success_count": 6, "res_sum": 60,
+	})
+	require.NoError(t, err)
+	recordB, err := tests.CreateRecord(hub, "network_monitor_stats", map[string]any{
+		"system":      sys.Id,
+		"monitor":     monitor.Id,
+		"type":        "1m",
+		"created":     created,
+		"res_min":     10,
+		"res_max":     60,
+		"total_count": 1, "success_count": 1, "res_sum": 22,
+	})
+	require.NoError(t, err)
+
+	result, count, err := rm.AverageMonitorStats(hub.DB(), monitor.Id, "1m", created-1)
+	require.NoError(t, err)
+	assert.Equal(t, 2, count)
+	assert.Equal(t, monitorEntity.Stats{ResAvg: 11.71, ResMin: 5, ResMax: 60, TotalCount: 7, SuccessCount: 7, ResponseSum: 82}, result)
+
+	for _, tc := range []struct {
+		name, monitor, recordType string
+		after                     int64
+	}{
+		{"other monitor", "missing", "1m", created - 1},
+		{"other type", monitor.Id, "10m", created - 1},
+		{"exclusive cutoff", monitor.Id, "1m", created},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			stats, count, err := rm.AverageMonitorStats(hub.DB(), tc.monitor, tc.recordType, tc.after)
+			require.NoError(t, err)
+			assert.Zero(t, count)
+			assert.Equal(t, monitorEntity.Stats{}, stats)
+		})
+	}
+
+	// A failure-only bucket counts toward loss but must not lower latency.
+	recordB.Set("res_min", 0)
+	recordB.Set("res_max", 0)
+	recordB.Set("success_count", 0)
+	recordB.Set("res_sum", 0)
+	require.NoError(t, hub.Save(recordB))
+	result, count, err = rm.AverageMonitorStats(hub.DB(), monitor.Id, "1m", created-1)
+	require.NoError(t, err)
+	assert.Equal(t, 2, count)
+	assert.Equal(t, monitorEntity.Stats{ResAvg: 10, ResMin: 5, ResMax: 20, Loss: 14.29, TotalCount: 7, SuccessCount: 6, ResponseSum: 60}, result)
+	// Sparse monitor records must propagate through every rollup level.
+	rm.CreateLongerRecords()
+	for _, recordType := range []string{"10m", "20m", "120m", "480m"} {
+		rollups, err := hub.FindAllRecords("network_monitor_stats", dbx.HashExp{"monitor": monitor.Id, "type": recordType})
+		require.NoError(t, err)
+		require.Len(t, rollups, 1, recordType)
+		assert.Equal(t, 5.0, rollups[0].GetFloat("res_min"))
+		assert.Equal(t, 20.0, rollups[0].GetFloat("res_max"))
+		assert.Equal(t, 7, rollups[0].GetInt("total_count"))
+		assert.Equal(t, 6, rollups[0].GetInt("success_count"))
+		assert.Equal(t, 60, rollups[0].GetInt("res_sum"))
+		// A sibling with a different number of probes must retain its actual
+		// weight when the next tier combines their underlying counts.
+		_, err = tests.CreateRecord(hub, "network_monitor_stats", map[string]any{
+			"system": sys.Id, "monitor": monitor.Id, "type": recordType, "created": created,
+			"res_min": 100, "res_max": 100,
+			"total_count": 3, "success_count": 1, "res_sum": 100,
+		})
+		require.NoError(t, err)
+		merged, count, err := rm.AverageMonitorStats(hub.DB(), monitor.Id, recordType, created-1)
+		require.NoError(t, err)
+		assert.Equal(t, 2, count)
+		assert.Equal(t, monitorEntity.Stats{
+			ResAvg: 22.86, ResMin: 5, ResMax: 100, Loss: 30,
+			TotalCount: 10, SuccessCount: 7, ResponseSum: 160,
+		}, merged)
+	}
+	// All failures produce zero latency, while a genuine zero-microsecond
+	// success remains a valid minimum (it must not be filtered out as a sentinel).
+	recordA.Set("success_count", 0)
+	recordA.Set("res_sum", 0)
+	require.NoError(t, hub.Save(recordA))
+	result, _, err = rm.AverageMonitorStats(hub.DB(), monitor.Id, "1m", created-1)
+	require.NoError(t, err)
+	assert.Equal(t, monitorEntity.Stats{TotalCount: 7, Loss: 100}, result)
+	recordA.Set("success_count", 1)
+	recordA.Set("res_min", 0)
+	recordA.Set("res_max", 0)
+	require.NoError(t, hub.Save(recordA))
+	result, _, err = rm.AverageMonitorStats(hub.DB(), monitor.Id, "1m", created-1)
+	require.NoError(t, err)
+	assert.Equal(t, monitorEntity.Stats{TotalCount: 7, SuccessCount: 1, Loss: 85.71}, result)
+}
+
+func TestSparseMonitorRollups(t *testing.T) {
+	for _, tc := range []struct {
+		name     string
+		interval int
+		samples  int
+		disabled bool
+	}{
+		{"five minute interval", 300, 2, false},
+		{"ten minute interval", 600, 1, false},
+		{"fifteen minute interval", 900, 1, false},
+		{"empty window", 900, 0, false},
+		{"disabled with pending history", 300, 2, true},
+		{"disabled empty window", 900, 0, true},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			hub, err := tests.NewTestHub(t.TempDir())
+			require.NoError(t, err)
+			defer hub.Cleanup()
+			user, err := tests.CreateUser(hub, "sparse-monitor@example.com", "testtesttest")
+			require.NoError(t, err)
+			sys, err := tests.CreateRecord(hub, "systems", map[string]any{
+				"name": "sparse-monitor-system", "host": "localhost", "port": "45876", "status": "up",
+				"users": []string{user.Id},
+			})
+			require.NoError(t, err)
+			monitor, err := tests.CreateRecord(hub, "network_monitors", map[string]any{
+				"system": sys.Id, "target": "1.1.1.1", "protocol": "icmp",
+				"interval": tc.interval, "enabled": true,
+			})
+			require.NoError(t, err)
+			now := time.Now()
+			for i := range tc.samples {
+				_, err := tests.CreateRecord(hub, "network_monitor_stats", map[string]any{
+					"system": sys.Id, "monitor": monitor.Id, "type": "1m",
+					"created": now.Add(-time.Minute - time.Duration(i*tc.interval)*time.Second).UnixMilli(),
+					"res_min": 8, "res_max": 20,
+					"total_count": 4, "success_count": 3, "res_sum": 36,
+				})
+				require.NoError(t, err)
+			}
+			// Other collections must still reject fewer than nine minute records.
+			for _, collection := range []string{"system_stats", "container_stats"} {
+				stats := `{"cpu":10}`
+				if collection == "container_stats" {
+					stats = `[{"name":"test","cpu":10}]`
+				}
+				for range 8 {
+					_, err := tests.CreateRecord(hub, collection, map[string]any{
+						"system": sys.Id, "type": "1m", "stats": stats,
+					})
+					require.NoError(t, err)
+				}
+			}
+			if tc.disabled {
+				monitor.Set("enabled", false)
+				require.NoError(t, hub.Save(monitor))
+			}
+			records.NewRecordManager(hub).CreateLongerRecords()
+			for _, recordType := range []string{"10m", "20m", "120m", "480m"} {
+				rollups, err := hub.FindAllRecords("network_monitor_stats", dbx.HashExp{"monitor": monitor.Id, "type": recordType})
+				require.NoError(t, err)
+				if tc.samples == 0 {
+					assert.Empty(t, rollups, recordType)
+				} else {
+					require.Len(t, rollups, 1, recordType)
+					assert.Equal(t, 8.0, rollups[0].GetFloat("res_min"))
+					assert.Equal(t, 20.0, rollups[0].GetFloat("res_max"))
+					assert.Equal(t, 4*tc.samples, rollups[0].GetInt("total_count"))
+					assert.Equal(t, 3*tc.samples, rollups[0].GetInt("success_count"))
+					assert.Equal(t, 36*tc.samples, rollups[0].GetInt("res_sum"))
+				}
+				for _, collection := range []string{"system_stats", "container_stats"} {
+					count, err := hub.CountRecords(collection, dbx.HashExp{"system": sys.Id, "type": recordType})
+					require.NoError(t, err)
+					assert.Zero(t, count, "%s %s", collection, recordType)
+				}
+			}
+		})
+	}
+}

internal/records/records.go

@@ -3,15 +3,16 @@ package records
 
 import (
 	"encoding/json"
-	"log/slog"
 	"math"
 	"time"
 
 	"github.com/henrygd/beszel/internal/entities/container"
+	"github.com/henrygd/beszel/internal/entities/monitor"
 	"github.com/henrygd/beszel/internal/entities/system"
 
 	"github.com/pocketbase/dbx"
 	"github.com/pocketbase/pocketbase/core"
+	"github.com/pocketbase/pocketbase/tools/types"
 )
 
 type RecordManager struct {
@@ -39,7 +40,7 @@ type StatsRecord struct {
 
 // Create longer records by averaging shorter records
 func (rm *RecordManager) CreateLongerRecords() {
-	// start := time.Now()
+	now := time.Now().UTC()
 	longerRecordData := []LongerRecordData{
 		{
 			shorterType: "1m",
@@ -69,23 +70,28 @@ func (rm *RecordManager) CreateLongerRecords() {
 	}
 	// wrap the operations in a transaction
 	// Pocketbase cron does not handle errors, log them here.
-	rm.app.RunInTransaction(func(txApp core.App) error {
+	err := rm.app.RunInTransaction(func(txApp core.App) error {
 		var err error
+
 		collections := [2]*core.Collection{}
 		collections[0], err = txApp.FindCachedCollectionByNameOrId("system_stats")
 		if err != nil {
-			slog.Error("Error finding cached collection using system stats:", "err", err)
 			return err
 		}
 		collections[1], err = txApp.FindCachedCollectionByNameOrId("container_stats")
 		if err != nil {
-			slog.Error("Error finding cached collection using container stats:", "err", err)
+			return err
+		}
+		monitorStatsColl, err := txApp.FindCachedCollectionByNameOrId("network_monitor_stats")
+		if err != nil {
 			return err
 		}
 		var systems RecordIds
 		db := txApp.DB()
 
-		db.NewQuery("SELECT id FROM systems WHERE status='up'").All(&systems)
+		if err := db.NewQuery("SELECT id FROM systems WHERE status='up'").All(&systems); err != nil {
+			return err
+		}
 
 		// loop through all active systems, time periods, and collections
 		for _, system := range systems {
@@ -94,44 +100,52 @@ func (rm *RecordManager) CreateLongerRecords() {
 				recordData := longerRecordData[i]
 				// log.Println("processing longer record type", recordData.longerType)
 				// add one minute padding for longer records because they are created slightly later than the job start time
-				longerRecordPeriod := time.Now().UTC().Add(recordData.longerTimeDuration + time.Minute)
+				longerRecordPeriod := now.Add(recordData.longerTimeDuration + time.Minute)
 				// shorter records are created independently of longer records, so we shouldn't need to add padding
-				shorterRecordPeriod := time.Now().UTC().Add(recordData.longerTimeDuration)
-				// loop through both collections
+				shorterRecordPeriod := now.Add(recordData.longerTimeDuration)
 				for _, collection := range collections {
 					// check creation time of last longer record if not 10m, since 10m is created every run
 					if recordData.longerType != "10m" {
-						count, err := txApp.CountRecords(
-							collection.Id,
-							dbx.NewExp(
-								"system = {:system} AND type = {:type} AND created > {:created}",
-								dbx.Params{"type": recordData.longerType, "system": system.Id, "created": longerRecordPeriod},
-							),
-						)
+						count, err := txApp.CountRecords(collection.Id, dbx.NewExp(
+							"system = {:system} AND type = {:type} AND created > {:created}",
+							dbx.Params{
+								"type":    recordData.longerType,
+								"system":  system.Id,
+								"created": longerRecordPeriod.Format(types.DefaultDateLayout),
+							},
+						))
+						if err != nil {
+							return err
+						}
 						// continue if longer record exists
-						if err != nil || count > 0 {
+						if count > 0 {
 							continue
 						}
 					}
 					// get shorter records from the past x minutes
 					var recordIds RecordIds
 
-					err := txApp.DB().
+					params := dbx.Params{
+						"type":    recordData.shorterType,
+						"system":  system.Id,
+						"created": shorterRecordPeriod.Format(types.DefaultDateLayout),
+					}
+
+					err := db.
 						Select("id").
 						From(collection.Name).
-						AndWhere(dbx.NewExp(
+						Where(dbx.NewExp(
 							"system={:system} AND type={:type} AND created > {:created}",
-							dbx.Params{
-								"type":    recordData.shorterType,
-								"system":  system.Id,
-								"created": shorterRecordPeriod,
-							},
+							params,
 						)).
 						OrderBy("created").
 						All(&recordIds)
+					if err != nil {
+						return err
+					}
 
 					// continue if not enough shorter records
-					if err != nil || len(recordIds) < recordData.minShorterRecords {
+					if len(recordIds) < recordData.minShorterRecords {
 						continue
 					}
 					// average the shorter records and create longer record
@@ -142,20 +156,88 @@ func (rm *RecordManager) CreateLongerRecords() {
 					case "system_stats":
 						longerRecord.Set("stats", rm.AverageSystemStats(db, recordIds))
 					case "container_stats":
-
 						longerRecord.Set("stats", rm.AverageContainerStats(db, recordIds))
 					}
 					if err := txApp.SaveNoValidate(longerRecord); err != nil {
-						slog.Error("failed to save longer record", "err", err)
+						txApp.Logger().Error("failed to save longer record", "err", err)
 					}
 				}
 			}
 		}
 
+		// network_monitor_stats is aggregated per monitor (not per system)
+		var monitors []struct {
+			Id     string `db:"id"`
+			System string `db:"system"`
+		}
+		// Disabled monitors still have history that must advance through retention tiers.
+		if err := db.NewQuery("SELECT id, system FROM network_monitors").All(&monitors); err != nil {
+			return err
+		}
+
+		for _, monitorRec := range monitors {
+			for i := range longerRecordData {
+				recordData := longerRecordData[i]
+				longerRecordPeriod := now.Add(recordData.longerTimeDuration + time.Minute)
+				shorterRecordPeriod := now.Add(recordData.longerTimeDuration)
+
+				if recordData.longerType != "10m" {
+					count, err := txApp.CountRecords(monitorStatsColl.Id, dbx.NewExp(
+						"monitor={:monitor} AND type={:type} AND created>{:created}",
+						dbx.Params{
+							"monitor": monitorRec.Id,
+							"type":    recordData.longerType,
+							"created": longerRecordPeriod.UnixMilli(),
+						},
+					))
+					if err != nil {
+						return err
+					}
+					if count > 0 {
+						continue
+					}
+				}
+
+				stats, count, err := rm.AverageMonitorStats(db, monitorRec.Id, recordData.shorterType, shorterRecordPeriod.UnixMilli())
+				if err != nil {
+					txApp.Logger().Error("failed to average monitor stats", "monitor", monitorRec.Id, "err", err)
+					continue
+				}
+				// Monitor intervals can exceed the aggregation window, so average
+				// any available records at every level and skip only empty windows.
+				if count == 0 {
+					continue
+				}
+
+				longerRecord := core.NewRecord(monitorStatsColl)
+				longerRecord.Set("system", monitorRec.System)
+				longerRecord.Set("monitor", monitorRec.Id)
+				longerRecord.Set("type", recordData.longerType)
+				longerRecord.Set("created", now.UnixMilli())
+				longerRecord.Set("res_min", stats.ResMin)
+				longerRecord.Set("res_max", stats.ResMax)
+				longerRecord.Set("total_count", stats.TotalCount)
+				longerRecord.Set("success_count", stats.SuccessCount)
+				longerRecord.Set("res_sum", stats.ResponseSum)
+				if err := txApp.SaveNoValidate(longerRecord); err != nil {
+					txApp.Logger().Error("failed to save monitor longer record", "err", err)
+				}
+			}
+		}
+
 		return nil
 	})
+	if err != nil {
+		rm.app.Logger().Error("failed to create longer records", "err", err)
+	}
+}
 
-	// log.Println("finished creating longer records", "time (ms)", time.Since(start).Milliseconds())
+func getCreatedTimeField(collectionName string, period time.Time) any {
+	// network_monitor_stats stores created as unix timestamp in ms, not as a date string
+	if collectionName == "network_monitor_stats" {
+		return period.UnixMilli()
+	}
+	return period.Format(types.DefaultDateLayout)
 }
 
 // Calculate the average stats of a list of system_stats records without reflect
@@ -596,6 +678,36 @@ func AverageContainerStatsSlice(records [][]container.Stats) []container.Stats {
 	return result
 }
 
+// AverageMonitorStats merges probe counts and response sums, preserving their
+// weights through every retention tier. Failed probes do not contribute latency.
+func (rm *RecordManager) AverageMonitorStats(db dbx.Builder, monitorID, recordType string, createdAfter int64) (monitor.Stats, int, error) {
+	var result struct {
+		monitor.Stats
+		Count int `db:"count"`
+	}
+	err := db.Select(
+		"COUNT(*) AS count",
+		"COALESCE(SUM(total_count), 0) AS total_count",
+		"COALESCE(SUM(success_count), 0) AS success_count",
+		"COALESCE(SUM(res_sum), 0) AS res_sum",
+		"COALESCE(MIN(CASE WHEN success_count > 0 THEN res_min END), 0) AS res_min",
+		"COALESCE(MAX(CASE WHEN success_count > 0 THEN res_max END), 0) AS res_max",
+	).From("network_monitor_stats").Where(dbx.NewExp(
+		"monitor={:monitor} AND type={:type} AND created>{:created}",
+		dbx.Params{"monitor": monitorID, "type": recordType, "created": createdAfter},
+	)).One(&result)
+	if err != nil {
+		return monitor.Stats{}, 0, err
+	}
+	if result.SuccessCount > 0 {
+		result.ResAvg = twoDecimals(float64(result.ResponseSum) / float64(result.SuccessCount))
+	}
+	if result.TotalCount > 0 {
+		result.Loss = twoDecimals(float64(result.TotalCount-result.SuccessCount) * 100 / float64(result.TotalCount))
+	}
+	return result.Stats, result.Count, nil
+}
+
 /* Round float to two decimals */
 func twoDecimals(value float64) float64 {
 	return math.Round(value*100) / 100

internal/records/records_deletion.go

@@ -3,7 +3,6 @@ package records
 import (
 	"fmt"
 	"log/slog"
-	"strings"
 	"time"
 
 	"github.com/pocketbase/dbx"
@@ -60,7 +59,7 @@ func deleteOldAlertsHistory(app core.App, countToKeep, countBeforeDeletion int)
 // Deletes system_stats records older than what is displayed in the UI
 func deleteOldSystemStats(app core.App) error {
 	// Collections to process
-	collections := [2]string{"system_stats", "container_stats"}
+	collections := [3]string{"system_stats", "container_stats", "network_monitor_stats"}
 
 	// Record types and their retention periods
 	type RecordDeletionData struct {
@@ -76,24 +75,17 @@ func deleteOldSystemStats(app core.App) error {
 	}
 
 	now := time.Now().UTC()
+	db := app.DB()
 
 	for _, collection := range collections {
-		// Build the WHERE clause
-		var conditionParts []string
-		var params dbx.Params = make(map[string]any)
-		for i := range recordData {
-			rd := recordData[i]
-			// Create parameterized condition for this record type
-			dateParam := fmt.Sprintf("date%d", i)
-			conditionParts = append(conditionParts, fmt.Sprintf("(type = '%s' AND created < {:%s})", rd.recordType, dateParam))
-			params[dateParam] = now.Add(-rd.retention)
-		}
-		// Combine conditions with OR
-		conditionStr := strings.Join(conditionParts, " OR ")
-		// Construct and execute the full raw query
-		rawQuery := fmt.Sprintf("DELETE FROM %s WHERE %s", collection, conditionStr)
-		if _, err := app.DB().NewQuery(rawQuery).Bind(params).Execute(); err != nil {
-			return fmt.Errorf("failed to delete from %s: %v", collection, err)
+		query := db.Delete(collection, dbx.NewExp("type={:type} AND created<{:created}"))
+		for _, rd := range recordData {
+			if _, err := query.Bind(dbx.Params{
+				"type":    rd.recordType,
+				"created": getCreatedTimeField(collection, now.Add(-rd.retention)),
+			}).Execute(); err != nil {
+				return fmt.Errorf("failed to delete from %s: %v", collection, err)
+			}
 		}
 	}
 	return nil

internal/records/records_rollup_test.go

@@ -0,0 +1,86 @@
+//go:build testing
+
+package records_test
+
+import (
+	"testing"
+	"time"
+
+	"github.com/henrygd/beszel/internal/records"
+	"github.com/henrygd/beszel/internal/tests"
+	"github.com/pocketbase/dbx"
+	"github.com/pocketbase/pocketbase/tools/types"
+	"github.com/stretchr/testify/require"
+)
+
+func TestLongerRecordsPreventDuplicates(t *testing.T) {
+	for _, collection := range []string{"system_stats", "container_stats", "network_monitor_stats"} {
+		for _, tier := range []struct {
+			shorter, longer string
+			count           int
+		}{
+			{"10m", "20m", 2},
+			{"20m", "120m", 6},
+			{"120m", "480m", 4},
+		} {
+			t.Run(collection+"/"+tier.longer, func(t *testing.T) {
+				hub, err := tests.NewTestHub(t.TempDir())
+				require.NoError(t, err)
+				defer hub.Cleanup()
+
+				user, err := tests.CreateUser(hub, "rollup@example.com", "testtesttest")
+				require.NoError(t, err)
+				sys, err := tests.CreateRecord(hub, "systems", map[string]any{
+					"name": "rollup-system", "host": "localhost", "port": "45876",
+					"status": "up", "users": []string{user.Id},
+				})
+				require.NoError(t, err)
+
+				created := time.Now().UTC().Add(-time.Minute)
+				data := map[string]any{
+					"system": sys.Id, "type": tier.shorter,
+					"created": created.Format(types.DefaultDateLayout),
+				}
+				filter := dbx.HashExp{"system": sys.Id, "type": tier.longer}
+				switch collection {
+				case "system_stats":
+					data["stats"] = `{"cpu":10}`
+				case "container_stats":
+					data["stats"] = `[{"name":"test","cpu":10}]`
+				case "network_monitor_stats":
+					monitor, err := tests.CreateRecord(hub, "network_monitors", map[string]any{
+						"system": sys.Id, "target": "1.1.1.1", "protocol": "icmp",
+						"interval": 30, "enabled": true,
+					})
+					require.NoError(t, err)
+					data["monitor"] = monitor.Id
+					data["created"] = created.UnixMilli()
+					data["total_count"] = 1
+					data["success_count"] = 1
+					data["res_sum"] = 10
+					data["res_min"] = 10
+					data["res_max"] = 10
+					filter["monitor"] = monitor.Id
+				}
+				for range tier.count {
+					_, err := tests.CreateRecord(hub, collection, data)
+					require.NoError(t, err)
+				}
+
+				rm := records.NewRecordManager(hub)
+				rm.CreateLongerRecords()
+				first, err := hub.FindAllRecords(collection, filter)
+				require.NoError(t, err)
+				require.Len(t, first, 1)
+
+				// The shorter records remain eligible, but the existing longer
+				// record must prevent another rollup on a subsequent invocation.
+				rm.CreateLongerRecords()
+				second, err := hub.FindAllRecords(collection, filter)
+				require.NoError(t, err)
+				require.Len(t, second, 1)
+				require.Equal(t, first[0].Id, second[0].Id)
+			})
+		}
+	}
+}