Add metricsource commands and custom alert targets (metric alerts phase 2)

go.mod

@@ -3,7 +3,7 @@ module github.com/deploys-app/deploys
 go 1.27.0
 
 require (
-	github.com/deploys-app/api v0.0.0-20260902050821-02959c8ba9ca
+	github.com/deploys-app/api v0.0.0-20260903231343-5b928d9c0434
 	github.com/moonrhythm/toon v0.0.0-20260702100246-6fcdad0a6a12
 	golang.org/x/mod v0.37.0
 	golang.org/x/oauth2 v0.14.0

go.sum

@@ -9,8 +9,8 @@ github.com/asaskevich/govalidator v0.0.0-20230301143203-a9d515a09cc2/go.mod h1:W
 github.com/creack/pty v1.1.9/go.mod h1:oKZEueFk5CKHvIhNR5MUki03XCEU+Q6VDXinZuGJ33E=
 github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
-github.com/deploys-app/api v0.0.0-20260902050821-02959c8ba9ca h1:0hWen3W//N1+F/haVxmqyQrUyUZtBfw+u/8jVT/xHVA=
-github.com/deploys-app/api v0.0.0-20260902050821-02959c8ba9ca/go.mod h1:QN5lioYbGyxSstxlyrvXrERJch7S2hnNCMNOVAu/6K0=
+github.com/deploys-app/api v0.0.0-20260903231343-5b928d9c0434 h1:kLWE9MLFCICccgsY6NossFoDtdrE1qt2DJdXXc3mXvA=
+github.com/deploys-app/api v0.0.0-20260903231343-5b928d9c0434/go.mod h1:QN5lioYbGyxSstxlyrvXrERJch7S2hnNCMNOVAu/6K0=
 github.com/dustin/go-humanize v1.0.1 h1:GzkhY7T5VNhEkwH0PVJgjz+fX1rhBrR7pRT3mDkpeCY=
 github.com/dustin/go-humanize v1.0.1/go.mod h1:Mu1zIs6XwVuF/gI1OepvI0qD18qycQx+mFykh5fBlto=
 github.com/golang/protobuf v1.3.1/go.mod h1:6lQm79b+lXiMfvg/cZm0SGofjICqVBUtrP5yJMmIC1U=

internal/runner/alert.go

@@ -7,7 +7,13 @@ import (
 	"github.com/deploys-app/api"
 )
 
-const alertOpFlagHelp = "comparison operator: >=, <=, gte, or lte (default >=; quote '>=' / '<=' in shells)"
+const (
+	alertOpFlagHelp     = "comparison operator: '>=', '<=', gte, or lte (default '>='; quote '>=' / '<=' in shells)"
+	alertKindFlagHelp   = "target kind: deployment or custom (empty = deployment)"
+	alertMetricFlagHelp = "metric to watch: cpu, memory, requests, or egress (kind=deployment); value or rate (kind=custom)"
+	alertSourceFlagHelp = "metric source name (kind=custom)"
+	alertSeriesFlagHelp = "series key (kind=custom)"
+)
 
 func (rn Runner) alert(args ...string) error {
 	if len(args) == 0 || IsHelpArg(args[0]) {
@@ -43,16 +49,20 @@ func (rn Runner) alert(args ...string) error {
 		var req api.AlertCreate
 		f.StringVar(&req.Project, "project", "", "project id")
 		f.StringVar(&req.Name, "name", "", "alert rule name")
-		f.StringVar(&req.Target.Location, "location", "", "target location")
-		f.StringVar(&req.Target.Deployment, "deployment", "", "target deployment name")
-		f.StringVar(&req.Condition.Metric, "metric", "", "metric to watch: cpu, memory, requests, or egress")
+		f.StringVar(&req.Target.Kind, "kind", "", alertKindFlagHelp)
+		f.StringVar(&req.Target.Location, "location", "", "target location (kind=deployment)")
+		f.StringVar(&req.Target.Deployment, "deployment", "", "target deployment name (kind=deployment)")
+		f.StringVar(&req.Target.Source, "source", "", alertSourceFlagHelp)
+		f.StringVar(&req.Target.Series, "series", "", alertSeriesFlagHelp)
+		f.StringVar(&req.Condition.Metric, "metric", "", alertMetricFlagHelp)
 		f.StringVar(&req.Condition.Op, "op", "", alertOpFlagHelp)
-		f.Float64Var(&req.Condition.Threshold, "threshold", 0, "threshold value (percent 0-100 for cpu/memory, req/min for requests, bytes/min for egress)")
+		f.Float64Var(&req.Condition.Threshold, "threshold", 0, "threshold value (percent 0-100 for cpu/memory, req/min for requests, bytes/min for egress, gauge for value, per-minute increase for rate)")
 		f.IntVar(&req.Condition.ForMinutes, "for", 0, "minutes the condition must hold continuously (1-60)")
 		f.IntVar(&req.RenotifyMinutes, "renotify", 0, "minutes between re-notifications while still firing (0 = disabled)")
 		f.BoolVar(&req.Disabled, "disabled", false, "create the rule disabled")
 		f.Parse(args[1:])
 		req.Condition.Op = normalizeAlertOp(req.Condition.Op)
+		applyAlertTargetKind(&req.Target)
 		resp, err = s.Create(context.Background(), &req)
 
 	case "update":
@@ -62,8 +72,11 @@ func (rn Runner) alert(args ...string) error {
 		// before applying overrides.
 		var (
 			req        api.AlertUpdate
+			kind       string
 			location   string
 			deployment string
+			source     string
+			series     string
 			metric     string
 			op         string
 			threshold  float64
@@ -73,11 +86,14 @@ func (rn Runner) alert(args ...string) error {
 		)
 		f.StringVar(&req.Project, "project", "", "project id")
 		f.StringVar(&req.Name, "name", "", "alert rule name")
-		f.StringVar(&location, "location", "", "target location")
-		f.StringVar(&deployment, "deployment", "", "target deployment name")
-		f.StringVar(&metric, "metric", "", "metric to watch: cpu, memory, requests, or egress")
+		f.StringVar(&kind, "kind", "", alertKindFlagHelp)
+		f.StringVar(&location, "location", "", "target location (kind=deployment)")
+		f.StringVar(&deployment, "deployment", "", "target deployment name (kind=deployment)")
+		f.StringVar(&source, "source", "", alertSourceFlagHelp)
+		f.StringVar(&series, "series", "", alertSeriesFlagHelp)
+		f.StringVar(&metric, "metric", "", alertMetricFlagHelp)
 		f.StringVar(&op, "op", "", alertOpFlagHelp)
-		f.Float64Var(&threshold, "threshold", 0, "threshold value (percent 0-100 for cpu/memory, req/min for requests, bytes/min for egress)")
+		f.Float64Var(&threshold, "threshold", 0, "threshold value (percent 0-100 for cpu/memory, req/min for requests, bytes/min for egress, gauge for value, per-minute increase for rate)")
 		f.IntVar(&forMin, "for", 0, "minutes the condition must hold continuously (1-60)")
 		f.IntVar(&renotify, "renotify", 0, "minutes between re-notifications while still firing (0 = disabled)")
 		f.BoolVar(&disabled, "disabled", false, "disable the rule")
@@ -95,12 +111,21 @@ func (rn Runner) alert(args ...string) error {
 		req.RenotifyMinutes = cur.RenotifyMinutes
 		req.Disabled = cur.Disabled
 
+		if set["kind"] {
+			req.Target.Kind = kind
+		}
 		if set["location"] {
 			req.Target.Location = location
 		}
 		if set["deployment"] {
 			req.Target.Deployment = deployment
 		}
+		if set["source"] {
+			req.Target.Source = source
+		}
+		if set["series"] {
+			req.Target.Series = series
+		}
 		if set["metric"] {
 			req.Condition.Metric = metric
 		}
@@ -119,6 +144,7 @@ func (rn Runner) alert(args ...string) error {
 		if set["disabled"] {
 			req.Disabled = disabled
 		}
+		applyAlertTargetKind(&req.Target)
 		resp, err = s.Update(context.Background(), &req)
 
 	case "delete":
@@ -142,6 +168,19 @@ func (rn Runner) alert(args ...string) error {
 	return rn.print(resp)
 }
 
+// applyAlertTargetKind enforces the API shape after flags are applied: kind=custom
+// requires Location/Deployment empty; kind=deployment (or empty) requires Source/
+// Series empty. Empty Kind is left empty (the API treats it as deployment).
+func applyAlertTargetKind(t *api.AlertTarget) {
+	if t.Kind == api.AlertTargetKindCustom {
+		t.Location = ""
+		t.Deployment = ""
+		return
+	}
+	t.Source = ""
+	t.Series = ""
+}
+
 // normalizeAlertOp maps CLI-friendly aliases to the API operators.
 // Shells treat unquoted >= and <= as redirections, so gte/lte are accepted too.
 func normalizeAlertOp(op string) string {

internal/runner/alert_test.go

@@ -43,3 +43,57 @@ func TestAlertCreateHelpMentionsShellSafeOp(t *testing.T) {
 		}
 	}
 }
+
+func TestAlertCreateHelpMentionsKindAndSource(t *testing.T) {
+	c := lookupCommand("alert")
+	if c == nil {
+		t.Fatal("alert group missing")
+	}
+	sub := c.lookupSub("create")
+	if sub == nil {
+		t.Fatal("alert create missing")
+	}
+	for _, want := range []string{"kind", "source", "value", "rate"} {
+		if !strings.Contains(sub.args, want) {
+			t.Errorf("create args %q missing %q", sub.args, want)
+		}
+	}
+}
+
+func TestApplyAlertTargetKind(t *testing.T) {
+	custom := api.AlertTarget{
+		Kind:       api.AlertTargetKindCustom,
+		Location:   "gke.cluster-rcf2",
+		Deployment: "web",
+		Source:     "app",
+		Series:     `queue_depth{queue="email"}`,
+	}
+	applyAlertTargetKind(&custom)
+	if custom.Kind != api.AlertTargetKindCustom {
+		t.Errorf("custom kind = %q; want %q", custom.Kind, api.AlertTargetKindCustom)
+	}
+	if custom.Location != "" || custom.Deployment != "" {
+		t.Errorf("custom target kept deployment fields: %+v", custom)
+	}
+	if custom.Source != "app" || custom.Series != `queue_depth{queue="email"}` {
+		t.Errorf("custom target dropped source/series: %+v", custom)
+	}
+
+	dep := api.AlertTarget{
+		Kind:       "",
+		Location:   "gke.cluster-rcf2",
+		Deployment: "web",
+		Source:     "app",
+		Series:     `queue_depth{queue="email"}`,
+	}
+	applyAlertTargetKind(&dep)
+	if dep.Kind != "" {
+		t.Errorf("empty kind rewritten to %q; want empty", dep.Kind)
+	}
+	if dep.Location != "gke.cluster-rcf2" || dep.Deployment != "web" {
+		t.Errorf("deployment target dropped location/deployment: %+v", dep)
+	}
+	if dep.Source != "" || dep.Series != "" {
+		t.Errorf("deployment target kept custom fields: %+v", dep)
+	}
+}

internal/runner/help.go

@@ -380,11 +380,23 @@ var commands = []command{
 			{name: "pull", args: "-name [-ack -limit -follow -poll -interval]", short: "fetch a pull channel's change events (ack to advance; -follow streams over SSE)"},
 		},
 	},
+	{
+		name:  "metricsource",
+		short: "Prometheus scrape sources for custom metrics and alerts",
+		subs: []subcommand{
+			{name: "set", args: "-name -location -deployment -port [-path /metrics -disabled]", short: "create or replace a scrape source"},
+			{name: "get", args: "-name", short: "show a scrape source"},
+			{name: "list", short: "list scrape sources"},
+			{name: "delete", args: "-name", short: "delete a scrape source"},
+			{name: "series", args: "-name", short: "list discovered series for a scrape source"},
+			{name: "query", args: "-name [-series s] [-timerange 1h|6h|12h|1d|7d|30d]", short: "query series for a scrape source"},
+		},
+	},
 	{
 		name:  "alert",
-		short: "metric alert rules on a deployment's CPU, memory, requests, or egress",
+		short: "metric alert rules on a deployment or a custom metric source",
 		subs: []subcommand{
-			{name: "create", args: "-name -location -deployment -metric cpu|memory|requests|egress -threshold n -for n [-op gte|lte|'>='|'<=' -renotify n -disabled]", short: "create a metric alert rule"},
+			{name: "create", args: "-name [-kind deployment|custom] [-location -deployment | -source -series] -metric cpu|memory|requests|egress|value|rate -threshold n -for n [-op gte|lte|'>='|'<=' -renotify n -disabled]", short: "create a metric alert rule"},
 			{name: "get", args: "-name", short: "show an alert rule"},
 			{name: "list", short: "list alert rules"},
 			{name: "update", args: "-name [flags]", short: "update an alert rule (omitted flags are preserved)"},

internal/runner/help_test.go

@@ -128,6 +128,9 @@ func TestRunHelpNeedsNoAPI(t *testing.T) {
 		{[]string{"alert"}, "Subcommands:"},
 		{[]string{"alert", "help"}, "metric alert rules"},
 		{[]string{"alert", "-h"}, "metric alert rules"},
+		{[]string{"metricsource"}, "Subcommands:"},
+		{[]string{"metricsource", "help"}, "Prometheus scrape sources"},
+		{[]string{"metricsource", "-h"}, "Prometheus scrape sources"},
 	}
 	for _, tc := range cases {
 		if err := tmp.Truncate(0); err != nil {

internal/runner/metricsource.go

@@ -0,0 +1,84 @@
+package runner
+
+import (
+	"context"
+
+	"github.com/deploys-app/api"
+)
+
+func (rn Runner) metricSource(args ...string) error {
+	if len(args) == 0 || IsHelpArg(args[0]) {
+		return rn.groupUsage("metricsource")
+	}
+
+	s := rn.API.MetricSource()
+
+	var (
+		resp any
+		err  error
+	)
+
+	f := rn.subFlagSet("metricsource", args[0])
+	switch args[0] {
+	default:
+		return rn.unknownSub("metricsource", args[0])
+
+	case "set":
+		var req api.MetricSourceSet
+		f.StringVar(&req.Project, "project", "", "project id")
+		f.StringVar(&req.Name, "name", "", "metric source name")
+		f.StringVar(&req.Location, "location", "", "location")
+		f.StringVar(&req.Deployment, "deployment", "", "deployment to scrape")
+		f.IntVar(&req.Port, "port", 0, "scrape port")
+		f.StringVar(&req.Path, "path", "/metrics", "scrape path (must start with /)")
+		f.BoolVar(&req.Disabled, "disabled", false, "disable scraping")
+		f.Parse(args[1:])
+		resp, err = s.Set(context.Background(), &req)
+
+	case "get":
+		var req api.MetricSourceGet
+		f.StringVar(&req.Project, "project", "", "project id")
+		f.StringVar(&req.Name, "name", "", "metric source name")
+		f.Parse(args[1:])
+		resp, err = s.Get(context.Background(), &req)
+
+	case "list":
+		var req api.MetricSourceList
+		f.StringVar(&req.Project, "project", "", "project id")
+		f.Parse(args[1:])
+		resp, err = s.List(context.Background(), &req)
+
+	case "delete":
+		var req api.MetricSourceDelete
+		f.StringVar(&req.Project, "project", "", "project id")
+		f.StringVar(&req.Name, "name", "", "metric source name")
+		f.Parse(args[1:])
+		resp, err = s.Delete(context.Background(), &req)
+
+	case "series":
+		var req api.MetricSourceSeries
+		f.StringVar(&req.Project, "project", "", "project id")
+		f.StringVar(&req.Name, "name", "", "metric source name")
+		f.Parse(args[1:])
+		resp, err = s.Series(context.Background(), &req)
+
+	case "query":
+		var (
+			req       api.MetricSourceQuery
+			series    multiFlag
+			timeRange string
+		)
+		f.StringVar(&req.Project, "project", "", "project id")
+		f.StringVar(&req.Name, "name", "", "metric source name")
+		f.Var(&series, "series", "series key (repeatable; omit to let the server pick)")
+		f.StringVar(&timeRange, "timerange", "1h", "time range (1h, 6h, 12h, 1d, 7d, 30d)")
+		f.Parse(args[1:])
+		req.Series = series
+		req.TimeRange = timeRange
+		resp, err = s.Query(context.Background(), &req)
+	}
+	if err != nil {
+		return err
+	}
+	return rn.print(resp)
+}

internal/runner/metricsource_test.go

@@ -0,0 +1,34 @@
+package runner
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestMetricSourceHelp(t *testing.T) {
+	c := lookupCommand("metricsource")
+	if c == nil {
+		t.Fatal("metricsource group missing")
+	}
+	for _, name := range []string{"set", "get", "list", "delete", "series", "query"} {
+		if c.lookupSub(name) == nil {
+			t.Errorf("metricsource missing subcommand %q", name)
+		}
+	}
+	query := c.lookupSub("query")
+	if query == nil {
+		t.Fatal("metricsource query missing")
+	}
+	if !strings.Contains(query.args, "timerange") {
+		t.Errorf("query args %q missing timerange", query.args)
+	}
+	set := c.lookupSub("set")
+	if set == nil {
+		t.Fatal("metricsource set missing")
+	}
+	for _, want := range []string{"location", "deployment", "port", "path"} {
+		if !strings.Contains(set.args, want) {
+			t.Errorf("set args %q missing %q", set.args, want)
+		}
+	}
+}

internal/runner/runner.go

@@ -179,6 +179,8 @@ func (rn Runner) Run(args ...string) error {
 		return rn.scheduler(args[1:]...)
 	case "notification":
 		return rn.notification(args[1:]...)
+	case "metricsource":
+		return rn.metricSource(args[1:]...)
 	case "alert":
 		return rn.alert(args[1:]...)
 	case "check-update":