mirror of
https://github.com/henrygd/beszel.git
synced 2026-04-05 12:31:49 +02:00
refactor: simplify/improve status alert handling (#1519)
also adds new functionality to restore any pending down alerts that were lost by hub restart before creation
This commit is contained in:
@@ -12,7 +12,6 @@ import (
|
||||
"testing/synctest"
|
||||
"time"
|
||||
|
||||
"github.com/henrygd/beszel/internal/alerts"
|
||||
beszelTests "github.com/henrygd/beszel/internal/tests"
|
||||
|
||||
"github.com/pocketbase/dbx"
|
||||
@@ -369,87 +368,6 @@ func TestUserAlertsApi(t *testing.T) {
|
||||
}
|
||||
}
|
||||
|
||||
func TestStatusAlerts(t *testing.T) {
|
||||
synctest.Test(t, func(t *testing.T) {
|
||||
hub, user := beszelTests.GetHubWithUser(t)
|
||||
defer hub.Cleanup()
|
||||
|
||||
systems, err := beszelTests.CreateSystems(hub, 4, user.Id, "paused")
|
||||
assert.NoError(t, err)
|
||||
|
||||
var alerts []*core.Record
|
||||
for i, system := range systems {
|
||||
alert, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
|
||||
"name": "Status",
|
||||
"system": system.Id,
|
||||
"user": user.Id,
|
||||
"min": i + 1,
|
||||
})
|
||||
assert.NoError(t, err)
|
||||
alerts = append(alerts, alert)
|
||||
}
|
||||
|
||||
time.Sleep(10 * time.Millisecond)
|
||||
|
||||
for _, alert := range alerts {
|
||||
assert.False(t, alert.GetBool("triggered"), "Alert should not be triggered immediately")
|
||||
}
|
||||
if hub.TestMailer.TotalSend() != 0 {
|
||||
assert.Zero(t, hub.TestMailer.TotalSend(), "Expected 0 messages, got %d", hub.TestMailer.TotalSend())
|
||||
}
|
||||
for _, system := range systems {
|
||||
assert.EqualValues(t, "paused", system.GetString("status"), "System should be paused")
|
||||
}
|
||||
for _, system := range systems {
|
||||
system.Set("status", "up")
|
||||
err = hub.SaveNoValidate(system)
|
||||
assert.NoError(t, err)
|
||||
}
|
||||
time.Sleep(time.Second)
|
||||
assert.EqualValues(t, 0, hub.GetPendingAlertsCount(), "should have 0 alerts in the pendingAlerts map")
|
||||
for _, system := range systems {
|
||||
system.Set("status", "down")
|
||||
err = hub.SaveNoValidate(system)
|
||||
assert.NoError(t, err)
|
||||
}
|
||||
// after 30 seconds, should have 4 alerts in the pendingAlerts map, no triggered alerts
|
||||
time.Sleep(time.Second * 30)
|
||||
assert.EqualValues(t, 4, hub.GetPendingAlertsCount(), "should have 4 alerts in the pendingAlerts map")
|
||||
triggeredCount, err := hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
|
||||
assert.NoError(t, err)
|
||||
assert.EqualValues(t, 0, triggeredCount, "should have 0 alert triggered")
|
||||
assert.EqualValues(t, 0, hub.TestMailer.TotalSend(), "should have 0 messages sent")
|
||||
// after 1:30 seconds, should have 1 triggered alert and 3 pending alerts
|
||||
time.Sleep(time.Second * 60)
|
||||
assert.EqualValues(t, 3, hub.GetPendingAlertsCount(), "should have 3 alerts in the pendingAlerts map")
|
||||
triggeredCount, err = hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
|
||||
assert.NoError(t, err)
|
||||
assert.EqualValues(t, 1, triggeredCount, "should have 1 alert triggered")
|
||||
assert.EqualValues(t, 1, hub.TestMailer.TotalSend(), "should have 1 messages sent")
|
||||
// after 2:30 seconds, should have 2 triggered alerts and 2 pending alerts
|
||||
time.Sleep(time.Second * 60)
|
||||
assert.EqualValues(t, 2, hub.GetPendingAlertsCount(), "should have 2 alerts in the pendingAlerts map")
|
||||
triggeredCount, err = hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
|
||||
assert.NoError(t, err)
|
||||
assert.EqualValues(t, 2, triggeredCount, "should have 2 alert triggered")
|
||||
assert.EqualValues(t, 2, hub.TestMailer.TotalSend(), "should have 2 messages sent")
|
||||
// now we will bring the remaning systems back up
|
||||
for _, system := range systems {
|
||||
system.Set("status", "up")
|
||||
err = hub.SaveNoValidate(system)
|
||||
assert.NoError(t, err)
|
||||
}
|
||||
time.Sleep(time.Second)
|
||||
// should have 0 alerts in the pendingAlerts map and 0 alerts triggered
|
||||
assert.EqualValues(t, 0, hub.GetPendingAlertsCount(), "should have 0 alerts in the pendingAlerts map")
|
||||
triggeredCount, err = hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
|
||||
assert.NoError(t, err)
|
||||
assert.Zero(t, triggeredCount, "should have 0 alert triggered")
|
||||
// 4 messages sent, 2 down alerts and 2 up alerts for first 2 systems
|
||||
assert.EqualValues(t, 4, hub.TestMailer.TotalSend(), "should have 4 messages sent")
|
||||
})
|
||||
}
|
||||
|
||||
func TestAlertsHistory(t *testing.T) {
|
||||
synctest.Test(t, func(t *testing.T) {
|
||||
hub, user := beszelTests.GetHubWithUser(t)
|
||||
@@ -578,102 +496,3 @@ func TestAlertsHistory(t *testing.T) {
|
||||
assert.EqualValues(t, 2, totalHistoryCount, "Should have 2 total alert history records")
|
||||
})
|
||||
}
|
||||
func TestResolveStatusAlerts(t *testing.T) {
|
||||
hub, user := beszelTests.GetHubWithUser(t)
|
||||
defer hub.Cleanup()
|
||||
|
||||
// Create a systemUp
|
||||
systemUp, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
|
||||
"name": "test-system",
|
||||
"users": []string{user.Id},
|
||||
"host": "127.0.0.1",
|
||||
"status": "up",
|
||||
})
|
||||
assert.NoError(t, err)
|
||||
|
||||
systemDown, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
|
||||
"name": "test-system-2",
|
||||
"users": []string{user.Id},
|
||||
"host": "127.0.0.2",
|
||||
"status": "up",
|
||||
})
|
||||
assert.NoError(t, err)
|
||||
|
||||
// Create a status alertUp for the system
|
||||
alertUp, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
|
||||
"name": "Status",
|
||||
"system": systemUp.Id,
|
||||
"user": user.Id,
|
||||
"min": 1,
|
||||
})
|
||||
assert.NoError(t, err)
|
||||
|
||||
alertDown, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
|
||||
"name": "Status",
|
||||
"system": systemDown.Id,
|
||||
"user": user.Id,
|
||||
"min": 1,
|
||||
})
|
||||
assert.NoError(t, err)
|
||||
|
||||
// Verify alert is not triggered initially
|
||||
assert.False(t, alertUp.GetBool("triggered"), "Alert should not be triggered initially")
|
||||
|
||||
// Set the system to 'up' (this should not trigger the alert)
|
||||
systemUp.Set("status", "up")
|
||||
err = hub.SaveNoValidate(systemUp)
|
||||
assert.NoError(t, err)
|
||||
|
||||
systemDown.Set("status", "down")
|
||||
err = hub.SaveNoValidate(systemDown)
|
||||
assert.NoError(t, err)
|
||||
|
||||
// Wait a moment for any processing
|
||||
time.Sleep(10 * time.Millisecond)
|
||||
|
||||
// Verify alertUp is still not triggered after setting system to up
|
||||
alertUp, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": alertUp.Id})
|
||||
assert.NoError(t, err)
|
||||
assert.False(t, alertUp.GetBool("triggered"), "Alert should not be triggered when system is up")
|
||||
|
||||
// Manually set both alerts triggered to true
|
||||
alertUp.Set("triggered", true)
|
||||
err = hub.SaveNoValidate(alertUp)
|
||||
assert.NoError(t, err)
|
||||
alertDown.Set("triggered", true)
|
||||
err = hub.SaveNoValidate(alertDown)
|
||||
assert.NoError(t, err)
|
||||
|
||||
// Verify we have exactly one alert with triggered true
|
||||
triggeredCount, err := hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
|
||||
assert.NoError(t, err)
|
||||
assert.EqualValues(t, 2, triggeredCount, "Should have exactly two alerts with triggered true")
|
||||
|
||||
// Verify the specific alertUp is triggered
|
||||
alertUp, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": alertUp.Id})
|
||||
assert.NoError(t, err)
|
||||
assert.True(t, alertUp.GetBool("triggered"), "Alert should be triggered")
|
||||
|
||||
// Verify we have two unresolved alert history records
|
||||
alertHistoryCount, err := hub.CountRecords("alerts_history", dbx.HashExp{"resolved": ""})
|
||||
assert.NoError(t, err)
|
||||
assert.EqualValues(t, 2, alertHistoryCount, "Should have exactly two unresolved alert history records")
|
||||
|
||||
err = alerts.ResolveStatusAlerts(hub)
|
||||
assert.NoError(t, err)
|
||||
|
||||
// Verify alertUp is not triggered after resolving
|
||||
alertUp, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": alertUp.Id})
|
||||
assert.NoError(t, err)
|
||||
assert.False(t, alertUp.GetBool("triggered"), "Alert should not be triggered after resolving")
|
||||
// Verify alertDown is still triggered
|
||||
alertDown, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": alertDown.Id})
|
||||
assert.NoError(t, err)
|
||||
assert.True(t, alertDown.GetBool("triggered"), "Alert should still be triggered after resolving")
|
||||
|
||||
// Verify we have one unresolved alert history record
|
||||
alertHistoryCount, err = hub.CountRecords("alerts_history", dbx.HashExp{"resolved": ""})
|
||||
assert.NoError(t, err)
|
||||
assert.EqualValues(t, 1, alertHistoryCount, "Should have exactly one unresolved alert history record")
|
||||
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user