Added custom messages for uptime alerts

README.md

@@ -188,6 +188,48 @@ specific default by adding the alert's ID or name to its
 `suppress_alert_ids` list in `cluster.yaml` (see "Edit cluster.yaml
 directly" below).
 
+## Custom alert messages
+
+Each alert can carry its own `subject_template` and `body_template`
+(Go `text/template` syntax). When set, they override the built-in
+formatting for that one alert; the default renderer is used otherwise.
+Discord ignores the subject template (it has no subject line).
+
+```sh
+qu alert add discord oncall --webhook https://... \
+    --body ':rotating_light: **{{.Check.Name}}** is now {{.Verb}}
+target: `{{.Check.Target}}`
+detail: {{.Snapshot.Detail}}'
+
+# multi-line templates are easier from a file
+qu alert add smtp ops --host ... --from ... --to ... \
+    --subject-file /etc/quptime/templates/ops.subject \
+    --body-file    /etc/quptime/templates/ops.body
+```
+
+Available template variables:
+
+| Variable | Meaning |
+|---|---|
+| `{{.Check.Name}}` | check name |
+| `{{.Check.Type}}` | `http` / `tcp` / `icmp` |
+| `{{.Check.Target}}` | URL or host:port being probed |
+| `{{.Check.ID}}` | UUID |
+| `{{.From}}` | previous state (`up` / `down` / `unknown`) |
+| `{{.To}}` | new state |
+| `{{.Verb}}` | `UP` / `DOWN` / `RECOVERED` |
+| `{{.Snapshot.Reports}}` | total per-node reports counted |
+| `{{.Snapshot.OKCount}}` | how many reported OK |
+| `{{.Snapshot.NotOK}}` | how many reported failure |
+| `{{.Snapshot.Detail}}` | first failure detail string |
+| `{{.NodeID}}` | master that dispatched |
+| `{{.When}}` | RFC3339 timestamp |
+
+`qu alert test <name>` exercises the template against a synthetic
+"homepage going DOWN" transition, so you can verify rendering before
+production traffic depends on it. A template parse or execution error
+falls back to the built-in format and is logged.
+
 ## Edit cluster.yaml directly
 
 Anything you can do through the CLI you can also do by editing
@@ -258,8 +300,8 @@ qu check add tcp   <name> <host:port>
 qu check add icmp  <name> <host>
 qu check list
 qu check remove <id-or-name>
-qu alert add smtp    <name> --host … --port … --from … --to … [--user --password --starttls] [--default]
-qu alert add discord <name> --webhook …                                                        [--default]
+qu alert add smtp    <name> --host … --port … --from … --to … [--user --password --starttls] [--default] [--subject … --body …]
+qu alert add discord <name> --webhook …                                                        [--default] [--body …]
 qu alert list / remove / test <id-or-name>
 qu alert default <id-or-name> on|off            toggle default attachment to every check
 qu trust list / remove <node-id>

internal/alerts/dispatcher.go

@@ -30,13 +30,16 @@ func (d *Dispatcher) OnTransition(check *config.Check, from, to checks.State, sn
 	if to == checks.StateUnknown {
 		return
 	}
-	msg := Render(d.selfID, check, from, to, snap)
 	alerts := d.cluster.EffectiveAlertsFor(check)
 	if len(alerts) == 0 && len(check.AlertIDs) > 0 {
 		d.logger.Printf("alerts: check %q references alerts but none resolved", check.Name)
 	}
 	for i := range alerts {
 		alert := alerts[i]
+		msg, err := RenderFor(&alert, d.selfID, check, from, to, snap)
+		if err != nil {
+			d.logger.Printf("alerts: %q template: %v — falling back to default", alert.Name, err)
+		}
 		if err := d.dispatchOne(&alert, msg); err != nil {
 			d.logger.Printf("alerts: %q via %s: %v", alert.Name, alert.Type, err)
 		}
@@ -44,18 +47,35 @@ func (d *Dispatcher) OnTransition(check *config.Check, from, to checks.State, sn
 }
 
 // Test sends a one-shot test message to the named alert. Returns an
-// error so the CLI can surface failures interactively.
+// error so the CLI can surface failures interactively. If the alert
+// carries custom templates they are exercised against a synthetic
+// "homepage going DOWN" transition so the operator can confirm the
+// template renders before a real outage.
 func (d *Dispatcher) Test(alertID string) error {
 	alert := d.cluster.FindAlert(alertID)
 	if alert == nil {
 		return fmt.Errorf("alert %q not found", alertID)
 	}
+	if alert.SubjectTemplate == "" && alert.BodyTemplate == "" {
 		msg := Message{
 			Subject: "[quptime] test alert",
 			Body:    fmt.Sprintf("This is a test of alert %q from node %s.\nIf you see this, the alert channel is wired correctly.\n", alert.Name, d.selfID),
 		}
 		return d.dispatchOne(alert, msg)
 	}
+	sample := &config.Check{
+		ID:     "test-check",
+		Name:   "test-check",
+		Type:   config.CheckHTTP,
+		Target: "https://example.com",
+	}
+	snap := checks.Snapshot{Reports: 3, OKCount: 0, NotOK: 3, Detail: "synthetic test failure"}
+	msg, err := RenderFor(alert, d.selfID, sample, checks.StateUp, checks.StateDown, snap)
+	if err != nil {
+		return fmt.Errorf("render template: %w", err)
+	}
+	return d.dispatchOne(alert, msg)
+}
 
 func (d *Dispatcher) dispatchOne(a *config.Alert, msg Message) error {
 	switch a.Type {

internal/alerts/message.go

@@ -4,14 +4,29 @@
 package alerts
 
 import (
+	"bytes"
 	"fmt"
 	"strings"
+	"text/template"
 	"time"
 
 	"git.cer.sh/axodouble/quptime/internal/checks"
 	"git.cer.sh/axodouble/quptime/internal/config"
 )
 
+// TemplateContext is what user-provided subject/body templates see. It
+// is also the shape the default renderer fills in, so changing one
+// place keeps the two paths consistent.
+type TemplateContext struct {
+	Check    *config.Check
+	From     string          // previous state name
+	To       string          // new state name
+	Verb     string          // "UP" | "DOWN" | "RECOVERED"
+	Snapshot checks.Snapshot // aggregate counts and detail
+	NodeID   string          // master that rendered the message
+	When     string          // RFC3339 timestamp
+}
+
 // Message is the rendered notification ready to ship across any
 // channel. Channels may format Subject + Body differently (SMTP uses
 // both; Discord renders a single string).
@@ -20,25 +35,82 @@ type Message struct {
 	Body    string
 }
 
-// Render produces a human-readable message from one state transition.
+// Render produces a human-readable message from one state transition
+// using the built-in format. Used as the fallback when no custom
+// template is configured (or when a custom template fails to render).
 func Render(nodeID string, check *config.Check, from, to checks.State, snap checks.Snapshot) Message {
-	now := time.Now().UTC().Format(time.RFC3339)
-	verb := transitionVerb(from, to)
-	subject := fmt.Sprintf("[quptime] %s %s — %s", check.Name, verb, check.Target)
+	ctx := newContext(nodeID, check, from, to, snap)
+	subject := fmt.Sprintf("[quptime] %s %s — %s", check.Name, ctx.Verb, check.Target)
 
 	var b strings.Builder
-	fmt.Fprintf(&b, "Check %q is now %s.\n", check.Name, strings.ToUpper(string(to)))
-	fmt.Fprintf(&b, "Previous state: %s\n", from)
+	fmt.Fprintf(&b, "Check %q is now %s.\n", check.Name, strings.ToUpper(ctx.To))
+	fmt.Fprintf(&b, "Previous state: %s\n", ctx.From)
 	fmt.Fprintf(&b, "Target:         %s (%s)\n", check.Target, check.Type)
 	fmt.Fprintf(&b, "Reports:        %d (ok=%d, fail=%d)\n", snap.Reports, snap.OKCount, snap.NotOK)
 	if snap.Detail != "" {
 		fmt.Fprintf(&b, "Detail:         %s\n", snap.Detail)
 	}
 	fmt.Fprintf(&b, "Master:         %s\n", nodeID)
-	fmt.Fprintf(&b, "When:           %s\n", now)
+	fmt.Fprintf(&b, "When:           %s\n", ctx.When)
 	return Message{Subject: subject, Body: b.String()}
 }
 
+// RenderFor produces a message for one specific alert. If the alert
+// defines SubjectTemplate or BodyTemplate, those override the
+// corresponding field from the default render. A template error falls
+// back to the default for that field and is reported via the returned
+// error (the caller is expected to log but still ship the message).
+func RenderFor(alert *config.Alert, nodeID string, check *config.Check, from, to checks.State, snap checks.Snapshot) (Message, error) {
+	def := Render(nodeID, check, from, to, snap)
+	if alert == nil || (alert.SubjectTemplate == "" && alert.BodyTemplate == "") {
+		return def, nil
+	}
+	ctx := newContext(nodeID, check, from, to, snap)
+	msg := def
+	var firstErr error
+	if alert.SubjectTemplate != "" {
+		s, err := execTemplate("subject", alert.SubjectTemplate, ctx)
+		if err != nil {
+			firstErr = err
+		} else {
+			msg.Subject = s
+		}
+	}
+	if alert.BodyTemplate != "" {
+		s, err := execTemplate("body", alert.BodyTemplate, ctx)
+		if err != nil && firstErr == nil {
+			firstErr = err
+		} else if err == nil {
+			msg.Body = s
+		}
+	}
+	return msg, firstErr
+}
+
+func newContext(nodeID string, check *config.Check, from, to checks.State, snap checks.Snapshot) TemplateContext {
+	return TemplateContext{
+		Check:    check,
+		From:     string(from),
+		To:       string(to),
+		Verb:     transitionVerb(from, to),
+		Snapshot: snap,
+		NodeID:   nodeID,
+		When:     time.Now().UTC().Format(time.RFC3339),
+	}
+}
+
+func execTemplate(name, src string, ctx TemplateContext) (string, error) {
+	tmpl, err := template.New(name).Option("missingkey=zero").Parse(src)
+	if err != nil {
+		return "", fmt.Errorf("parse %s template: %w", name, err)
+	}
+	var b bytes.Buffer
+	if err := tmpl.Execute(&b, ctx); err != nil {
+		return "", fmt.Errorf("execute %s template: %w", name, err)
+	}
+	return b.String(), nil
+}
+
 func transitionVerb(from, to checks.State) string {
 	switch to {
 	case checks.StateDown:

internal/alerts/message_test.go

@@ -50,3 +50,61 @@ func TestRenderUpInitialTransition(t *testing.T) {
 		t.Error("first-time UP should not be tagged RECOVERED")
 	}
 }
+
+func TestRenderForUsesAlertTemplates(t *testing.T) {
+	check := &config.Check{Name: "homepage", Target: "https://example.com", Type: config.CheckHTTP}
+	snap := checks.Snapshot{Reports: 3, OKCount: 0, NotOK: 3, Detail: "connection refused"}
+	alert := &config.Alert{
+		SubjectTemplate: "{{.Check.Name}} is {{.Verb}}",
+		BodyTemplate:    "{{.Check.Target}} :: {{.Snapshot.Detail}}",
+	}
+	msg, err := RenderFor(alert, "master", check, checks.StateUp, checks.StateDown, snap)
+	if err != nil {
+		t.Fatalf("unexpected error: %v", err)
+	}
+	if msg.Subject != "homepage is DOWN" {
+		t.Errorf("subject = %q", msg.Subject)
+	}
+	if msg.Body != "https://example.com :: connection refused" {
+		t.Errorf("body = %q", msg.Body)
+	}
+}
+
+func TestRenderForFallsBackToDefaultPerField(t *testing.T) {
+	check := &config.Check{Name: "homepage", Target: "https://example.com", Type: config.CheckHTTP}
+	snap := checks.Snapshot{Reports: 3, OKCount: 0, NotOK: 3}
+	// only body overridden; subject should match default.
+	alert := &config.Alert{BodyTemplate: "custom body"}
+	msg, err := RenderFor(alert, "master", check, checks.StateUp, checks.StateDown, snap)
+	if err != nil {
+		t.Fatalf("unexpected error: %v", err)
+	}
+	if !strings.Contains(msg.Subject, "DOWN") {
+		t.Errorf("subject should be default rendering, got %q", msg.Subject)
+	}
+	if msg.Body != "custom body" {
+		t.Errorf("body = %q", msg.Body)
+	}
+}
+
+func TestRenderForReportsTemplateError(t *testing.T) {
+	check := &config.Check{Name: "homepage", Target: "https://example.com"}
+	snap := checks.Snapshot{}
+	alert := &config.Alert{BodyTemplate: "{{.Check.MissingField"} // unbalanced
+	_, err := RenderFor(alert, "master", check, checks.StateUp, checks.StateDown, snap)
+	if err == nil {
+		t.Fatal("expected parse error for malformed template")
+	}
+}
+
+func TestRenderForNilAlertReturnsDefault(t *testing.T) {
+	check := &config.Check{Name: "homepage", Target: "https://example.com"}
+	snap := checks.Snapshot{Reports: 1, OKCount: 1}
+	msg, err := RenderFor(nil, "master", check, checks.StateUp, checks.StateUp, snap)
+	if err != nil {
+		t.Fatalf("unexpected error: %v", err)
+	}
+	if !strings.Contains(msg.Subject, "homepage") {
+		t.Errorf("default subject should mention check, got %q", msg.Subject)
+	}
+}

internal/cli/alert.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
+	"os"
 	"text/tabwriter"
 	"time"
 
@@ -15,6 +16,41 @@ import (
 	"git.cer.sh/axodouble/quptime/internal/transport"
 )
 
+// bindTemplateFlags attaches --subject / --subject-file / --body /
+// --body-file to a cobra command. resolveTemplateFlags reads the file
+// variants (if non-empty) and returns the effective subject + body
+// template strings. Inline flags take precedence over file flags.
+func bindTemplateFlags(cmd *cobra.Command) {
+	cmd.Flags().String("subject", "", "subject template (text/template syntax — SMTP only)")
+	cmd.Flags().String("subject-file", "", "path to a file containing the subject template")
+	cmd.Flags().String("body", "", "body template (text/template syntax)")
+	cmd.Flags().String("body-file", "", "path to a file containing the body template")
+}
+
+func resolveTemplateFlags(cmd *cobra.Command) (subject, body string, err error) {
+	subject, _ = cmd.Flags().GetString("subject")
+	body, _ = cmd.Flags().GetString("body")
+	if subject == "" {
+		if p, _ := cmd.Flags().GetString("subject-file"); p != "" {
+			raw, e := os.ReadFile(p)
+			if e != nil {
+				return "", "", fmt.Errorf("read --subject-file %s: %w", p, e)
+			}
+			subject = string(raw)
+		}
+	}
+	if body == "" {
+		if p, _ := cmd.Flags().GetString("body-file"); p != "" {
+			raw, e := os.ReadFile(p)
+			if e != nil {
+				return "", "", fmt.Errorf("read --body-file %s: %w", p, e)
+			}
+			body = string(raw)
+		}
+	}
+	return subject, body, nil
+}
+
 func addAlertCmd(root *cobra.Command) {
 	alert := &cobra.Command{
 		Use:   "alert",
@@ -136,11 +172,17 @@ func buildSMTPAddCmd() *cobra.Command {
 		RunE: func(cmd *cobra.Command, args []string) error {
 			ctx, cancel := context.WithTimeout(cmd.Context(), 10*time.Second)
 			defer cancel()
+			subj, body, err := resolveTemplateFlags(cmd)
+			if err != nil {
+				return err
+			}
 			a := config.Alert{
 				ID:              uuid.NewString(),
 				Name:            args[0],
 				Type:            config.AlertSMTP,
 				Default:         makeDefault,
+				SubjectTemplate: subj,
+				BodyTemplate:    body,
 				SMTPHost:        host,
 				SMTPPort:        port,
 				SMTPUser:        user,
@@ -150,8 +192,8 @@ func buildSMTPAddCmd() *cobra.Command {
 				SMTPStartTLS:    startTLS,
 			}
 			payload, _ := json.Marshal(a)
-			body := daemon.MutateBody{Kind: transport.MutationAddAlert, Payload: payload}
-			raw, err := callDaemon(ctx, daemon.CtrlMutate, body)
+			mb := daemon.MutateBody{Kind: transport.MutationAddAlert, Payload: payload}
+			raw, err := callDaemon(ctx, daemon.CtrlMutate, mb)
 			if err != nil {
 				return err
 			}
@@ -170,6 +212,7 @@ func buildSMTPAddCmd() *cobra.Command {
 	cmd.Flags().StringSliceVar(&to, "to", nil, "recipient address (repeat or comma-separate)")
 	cmd.Flags().BoolVar(&startTLS, "starttls", true, "negotiate STARTTLS")
 	cmd.Flags().BoolVar(&makeDefault, "default", false, "attach this alert to every check automatically")
+	bindTemplateFlags(cmd)
 	_ = cmd.MarkFlagRequired("host")
 	_ = cmd.MarkFlagRequired("from")
 	_ = cmd.MarkFlagRequired("to")
@@ -186,16 +229,22 @@ func buildDiscordAddCmd() *cobra.Command {
 		RunE: func(cmd *cobra.Command, args []string) error {
 			ctx, cancel := context.WithTimeout(cmd.Context(), 10*time.Second)
 			defer cancel()
+			subj, body, err := resolveTemplateFlags(cmd)
+			if err != nil {
+				return err
+			}
 			a := config.Alert{
 				ID:              uuid.NewString(),
 				Name:            args[0],
 				Type:            config.AlertDiscord,
 				Default:         makeDefault,
+				SubjectTemplate: subj,
+				BodyTemplate:    body,
 				DiscordWebhook:  webhook,
 			}
 			payload, _ := json.Marshal(a)
-			body := daemon.MutateBody{Kind: transport.MutationAddAlert, Payload: payload}
-			raw, err := callDaemon(ctx, daemon.CtrlMutate, body)
+			mb := daemon.MutateBody{Kind: transport.MutationAddAlert, Payload: payload}
+			raw, err := callDaemon(ctx, daemon.CtrlMutate, mb)
 			if err != nil {
 				return err
 			}
@@ -208,6 +257,7 @@ func buildDiscordAddCmd() *cobra.Command {
 	}
 	cmd.Flags().StringVar(&webhook, "webhook", "", "discord webhook URL")
 	cmd.Flags().BoolVar(&makeDefault, "default", false, "attach this alert to every check automatically")
+	bindTemplateFlags(cmd)
 	_ = cmd.MarkFlagRequired("webhook")
 	return cmd
 }

internal/config/cluster.go

@@ -85,6 +85,16 @@ type Alert struct {
 
 	// Discord options.
 	DiscordWebhook string `yaml:"discord_webhook,omitempty"`
+
+	// SubjectTemplate / BodyTemplate are optional text/template strings
+	// that override the default rendering. Empty means use the built-in
+	// format. Discord ignores SubjectTemplate (it has no subject line);
+	// SMTP uses both. Available variables: {{.Check.Name}},
+	// {{.Check.Type}}, {{.Check.Target}}, {{.Check.ID}}, {{.From}},
+	// {{.To}}, {{.Verb}}, {{.Snapshot.Reports}}, {{.Snapshot.OKCount}},
+	// {{.Snapshot.NotOK}}, {{.Snapshot.Detail}}, {{.NodeID}}, {{.When}}.
+	SubjectTemplate string `yaml:"subject_template,omitempty"`
+	BodyTemplate    string `yaml:"body_template,omitempty"`
 }
 
 // ClusterConfig is the replicated cluster state. The Version field