feat: implement HTMLDebug for PGCoord with v2 API (#10914)

Implements HTMLDebug for the PGCoordinator with the new v2 API and related DB tables.

.prettierignore

@@ -82,6 +82,7 @@ helm/**/templates/*.yaml
 
 # Testdata shouldn't be formatted.
 scripts/apitypings/testdata/**/*.ts
+enterprise/tailnet/testdata/*.golden.html
 
 # Generated files shouldn't be formatted.
 site/e2e/provisionerGenerated.ts

.prettierignore.include

@@ -8,6 +8,7 @@ helm/**/templates/*.yaml
 
 # Testdata shouldn't be formatted.
 scripts/apitypings/testdata/**/*.ts
+enterprise/tailnet/testdata/*.golden.html
 
 # Generated files shouldn't be formatted.
 site/e2e/provisionerGenerated.ts

Makefile

@@ -595,7 +595,15 @@ coderd/apidoc/swagger.json: $(shell find ./scripts/apidocgen $(FIND_EXCLUSIONS)
 	./scripts/apidocgen/generate.sh
 	pnpm run format:write:only ./docs/api ./docs/manifest.json ./coderd/apidoc/swagger.json
 
-update-golden-files: cli/testdata/.gen-golden helm/coder/tests/testdata/.gen-golden helm/provisioner/tests/testdata/.gen-golden scripts/ci-report/testdata/.gen-golden enterprise/cli/testdata/.gen-golden coderd/.gen-golden provisioner/terraform/testdata/.gen-golden
+update-golden-files: \
+	cli/testdata/.gen-golden \
+	helm/coder/tests/testdata/.gen-golden \
+	helm/provisioner/tests/testdata/.gen-golden \
+	scripts/ci-report/testdata/.gen-golden \
+	enterprise/cli/testdata/.gen-golden \
+	enterprise/tailnet/testdata/.gen-golden \
+	coderd/.gen-golden \
+	provisioner/terraform/testdata/.gen-golden
 .PHONY: update-golden-files
 
 cli/testdata/.gen-golden: $(wildcard cli/testdata/*.golden) $(wildcard cli/*.tpl) $(GO_SRC_FILES) $(wildcard cli/*_test.go)
@@ -606,6 +614,10 @@ enterprise/cli/testdata/.gen-golden: $(wildcard enterprise/cli/testdata/*.golden
 	go test ./enterprise/cli -run="TestEnterpriseCommandHelp" -update
 	touch "$@"
 
+enterprise/tailnet/testdata/.gen-golden: $(wildcard enterprise/tailnet/testdata/*.golden.html) $(GO_SRC_FILES) $(wildcard enterprise/tailnet/*_test.go)
+	go test ./enterprise/tailnet -run="TestDebugTemplate" -update
+	touch "$@"
+
 helm/coder/tests/testdata/.gen-golden: $(wildcard helm/coder/tests/testdata/*.yaml) $(wildcard helm/coder/tests/testdata/*.golden) $(GO_SRC_FILES) $(wildcard helm/coder/tests/*_test.go)
 	go test ./helm/coder/tests -run=TestUpdateGoldenFiles -update
 	touch "$@"

enterprise/tailnet/htmldebug.go

@@ -0,0 +1,199 @@
+package tailnet
+
+import (
+	"context"
+	"database/sql"
+	"html/template"
+	"net/http"
+	"time"
+
+	"github.com/google/uuid"
+	"golang.org/x/xerrors"
+	gProto "google.golang.org/protobuf/proto"
+
+	"github.com/coder/coder/v2/coderd/database"
+	"github.com/coder/coder/v2/tailnet/proto"
+)
+
+type HTMLDebug struct {
+	Coordinators []*HTMLCoordinator
+	Peers        []*HTMLPeer
+	Tunnels      []*HTMLTunnel
+}
+
+type HTMLPeer struct {
+	ID            uuid.UUID
+	CoordinatorID uuid.UUID
+	LastWriteAge  time.Duration
+	Node          string
+	Status        database.TailnetStatus
+}
+
+type HTMLCoordinator struct {
+	ID           uuid.UUID
+	HeartbeatAge time.Duration
+}
+
+type HTMLTunnel struct {
+	CoordinatorID uuid.UUID
+	SrcID         uuid.UUID
+	DstID         uuid.UUID
+	LastWriteAge  time.Duration
+}
+
+func (c *pgCoord) ServeHTTPDebug(w http.ResponseWriter, r *http.Request) {
+	ctx := r.Context()
+	debug, err := getDebug(ctx, c.store)
+	if err != nil {
+		w.WriteHeader(http.StatusInternalServerError)
+		_, _ = w.Write([]byte(err.Error()))
+		return
+	}
+
+	w.Header().Set("Content-Type", "text/html; charset=utf-8")
+
+	err = debugTempl.Execute(w, debug)
+	if err != nil {
+		w.WriteHeader(http.StatusInternalServerError)
+		_, _ = w.Write([]byte(err.Error()))
+		return
+	}
+}
+
+func getDebug(ctx context.Context, store database.Store) (HTMLDebug, error) {
+	out := HTMLDebug{}
+	coords, err := store.GetAllTailnetCoordinators(ctx)
+	if err != nil && !xerrors.Is(err, sql.ErrNoRows) {
+		return HTMLDebug{}, xerrors.Errorf("failed to query coordinators: %w", err)
+	}
+	peers, err := store.GetAllTailnetPeers(ctx)
+	if err != nil && !xerrors.Is(err, sql.ErrNoRows) {
+		return HTMLDebug{}, xerrors.Errorf("failed to query peers: %w", err)
+	}
+	tunnels, err := store.GetAllTailnetTunnels(ctx)
+	if err != nil && !xerrors.Is(err, sql.ErrNoRows) {
+		return HTMLDebug{}, xerrors.Errorf("failed to query tunnels: %w", err)
+	}
+	now := time.Now() // call this once so all our ages are on the same timebase
+	for _, coord := range coords {
+		out.Coordinators = append(out.Coordinators, coordToHTML(coord, now))
+	}
+	for _, peer := range peers {
+		ph, err := peerToHTML(peer, now)
+		if err != nil {
+			return HTMLDebug{}, err
+		}
+		out.Peers = append(out.Peers, ph)
+	}
+	for _, tunnel := range tunnels {
+		out.Tunnels = append(out.Tunnels, tunnelToHTML(tunnel, now))
+	}
+	return out, nil
+}
+
+func coordToHTML(d database.TailnetCoordinator, now time.Time) *HTMLCoordinator {
+	return &HTMLCoordinator{
+		ID:           d.ID,
+		HeartbeatAge: now.Sub(d.HeartbeatAt),
+	}
+}
+
+func peerToHTML(d database.TailnetPeer, now time.Time) (*HTMLPeer, error) {
+	node := &proto.Node{}
+	err := gProto.Unmarshal(d.Node, node)
+	if err != nil {
+		return nil, xerrors.Errorf("unmarshal node: %w", err)
+	}
+	return &HTMLPeer{
+		ID:            d.ID,
+		CoordinatorID: d.CoordinatorID,
+		LastWriteAge:  now.Sub(d.UpdatedAt),
+		Status:        d.Status,
+		Node:          node.String(),
+	}, nil
+}
+
+func tunnelToHTML(d database.TailnetTunnel, now time.Time) *HTMLTunnel {
+	return &HTMLTunnel{
+		CoordinatorID: d.CoordinatorID,
+		SrcID:         d.SrcID,
+		DstID:         d.DstID,
+		LastWriteAge:  now.Sub(d.UpdatedAt),
+	}
+}
+
+var coordinatorDebugTmpl = `
+<!DOCTYPE html>
+<html>
+	<head>
+		<meta charset="UTF-8">
+		<style>
+th, td {
+  padding-top: 6px;
+  padding-bottom: 6px;
+  padding-left: 10px;
+  padding-right: 10px;
+  text-align: left;
+}
+tr {
+  border-bottom: 1px solid #ddd;
+}
+		</style>
+	</head>
+	<body>
+		<h2 id=coordinators><a href=#coordinators>#</a> coordinators: total {{ len .Coordinators }}</h2>
+		<table>
+			<tr style="margin-top:4px">
+				<th>ID</th>
+				<th>Heartbeat Age</th>
+			</tr>
+		{{- range .Coordinators}}
+			<tr style="margin-top:4px">
+				<td>{{ .ID }}</td>
+				<td>{{ .HeartbeatAge }} ago</td>
+			</tr>
+		{{- end }}
+		</table>
+
+		<h2 id=peers> <a href=#peers>#</a> peers: total {{ len .Peers }} </h2>
+		<table>
+			<tr style="margin-top:4px">
+				<th>ID</th>
+				<th>CoordinatorID</th>
+				<th>Status</th>
+				<th>Last Write Age</th>
+				<th>Node</th>
+			</tr>
+		{{- range .Peers }}
+			<tr style="margin-top:4px">
+				<td>{{ .ID }}</td>
+				<td>{{ .CoordinatorID }}</td>
+				<td>{{ .Status }}</td>
+				<td>{{ .LastWriteAge }} ago</td>
+				<td style="white-space: pre;"><code>{{ .Node }}</code></td>
+			</tr>
+		{{- end }}
+		</table>
+
+		<h2 id=tunnels><a href=#tunnels>#</a> tunnels: total {{ len .Tunnels }}</h2>
+		<table>
+			<tr style="margin-top:4px">
+				<th>SrcID</th>
+				<th>DstID</th>
+				<th>CoordinatorID</th>
+				<th>Last Write Age</th>
+			</tr>
+		{{- range .Tunnels }}
+			<tr style="margin-top:4px">
+				<td>{{ .SrcID }}</td>
+				<td>{{ .DstID }}</td>
+				<td>{{ .CoordinatorID }}</td>
+				<td>{{ .LastWriteAge }} ago</td>
+			</tr>
+		{{- end }}
+		</table>
+	</body>
+</html>
+`
+
+var debugTempl = template.Must(template.New("coordinator_debug").Parse(coordinatorDebugTmpl))

enterprise/tailnet/pgcoord.go

@@ -6,7 +6,6 @@ import (
 	"encoding/json"
 	"io"
 	"net"
-	"net/http"
 	"net/netip"
 	"strings"
 	"sync"
@@ -19,7 +18,6 @@ import (
 
 	"github.com/cenkalti/backoff/v4"
 	"github.com/google/uuid"
-	"golang.org/x/exp/slices"
 	"golang.org/x/xerrors"
 	gProto "google.golang.org/protobuf/proto"
 
@@ -28,7 +26,6 @@ import (
 	"github.com/coder/coder/v2/coderd/database/dbauthz"
 	"github.com/coder/coder/v2/coderd/database/pubsub"
 	"github.com/coder/coder/v2/coderd/rbac"
-	"github.com/coder/coder/v2/coderd/util/slice"
 	agpl "github.com/coder/coder/v2/tailnet"
 )
 
@@ -1296,29 +1293,6 @@ func (q *querier) setHealthy() {
 	q.healthy = true
 }
 
-func (q *querier) getAll(ctx context.Context) (map[uuid.UUID]database.TailnetAgent, map[uuid.UUID][]database.TailnetClient, error) {
-	agents, err := q.store.GetAllTailnetAgents(ctx)
-	if err != nil {
-		return nil, nil, xerrors.Errorf("get all tailnet agents: %w", err)
-	}
-	agentsMap := map[uuid.UUID]database.TailnetAgent{}
-	for _, agent := range agents {
-		agentsMap[agent.ID] = agent
-	}
-	clients, err := q.store.GetAllTailnetClients(ctx)
-	if err != nil {
-		return nil, nil, xerrors.Errorf("get all tailnet clients: %w", err)
-	}
-	clientsMap := map[uuid.UUID][]database.TailnetClient{}
-	for _, client := range clients {
-		for _, agentID := range client.AgentIds {
-			clientsMap[agentID] = append(clientsMap[agentID], client.TailnetClient)
-		}
-	}
-
-	return agentsMap, clientsMap, nil
-}
-
 func parseTunnelUpdate(msg string) ([]uuid.UUID, error) {
 	parts := strings.Split(msg, ",")
 	if len(parts) != 2 {
@@ -1721,91 +1695,3 @@ func (h *heartbeats) cleanup() {
 	}
 	h.logger.Debug(h.ctx, "cleaned up old coordinators")
 }
-
-func (c *pgCoord) ServeHTTPDebug(w http.ResponseWriter, r *http.Request) {
-	ctx := r.Context()
-	debug, err := c.htmlDebug(ctx)
-	if err != nil {
-		w.WriteHeader(http.StatusInternalServerError)
-		_, _ = w.Write([]byte(err.Error()))
-		return
-	}
-
-	agpl.CoordinatorHTTPDebug(debug)(w, r)
-}
-
-func (c *pgCoord) htmlDebug(ctx context.Context) (agpl.HTMLDebug, error) {
-	now := time.Now()
-	data := agpl.HTMLDebug{}
-	agents, clients, err := c.querier.getAll(ctx)
-	if err != nil {
-		return data, xerrors.Errorf("get all agents and clients: %w", err)
-	}
-
-	for _, agent := range agents {
-		htmlAgent := &agpl.HTMLAgent{
-			ID: agent.ID,
-			// Name: ??, TODO: get agent names
-			LastWriteAge: now.Sub(agent.UpdatedAt).Round(time.Second),
-		}
-		for _, conn := range clients[agent.ID] {
-			htmlAgent.Connections = append(htmlAgent.Connections, &agpl.HTMLClient{
-				ID:           conn.ID,
-				Name:         conn.ID.String(),
-				LastWriteAge: now.Sub(conn.UpdatedAt).Round(time.Second),
-			})
-			data.Nodes = append(data.Nodes, &agpl.HTMLNode{
-				ID:   conn.ID,
-				Node: conn.Node,
-			})
-		}
-		slices.SortFunc(htmlAgent.Connections, func(a, b *agpl.HTMLClient) int {
-			return slice.Ascending(a.Name, b.Name)
-		})
-
-		data.Agents = append(data.Agents, htmlAgent)
-		data.Nodes = append(data.Nodes, &agpl.HTMLNode{
-			ID: agent.ID,
-			// Name: ??, TODO: get agent names
-			Node: agent.Node,
-		})
-	}
-	slices.SortFunc(data.Agents, func(a, b *agpl.HTMLAgent) int {
-		return slice.Ascending(a.Name, b.Name)
-	})
-
-	for agentID, conns := range clients {
-		if len(conns) == 0 {
-			continue
-		}
-
-		if _, ok := agents[agentID]; ok {
-			continue
-		}
-		agent := &agpl.HTMLAgent{
-			Name: "unknown",
-			ID:   agentID,
-		}
-		for _, conn := range conns {
-			agent.Connections = append(agent.Connections, &agpl.HTMLClient{
-				Name:         conn.ID.String(),
-				ID:           conn.ID,
-				LastWriteAge: now.Sub(conn.UpdatedAt).Round(time.Second),
-			})
-			data.Nodes = append(data.Nodes, &agpl.HTMLNode{
-				ID:   conn.ID,
-				Node: conn.Node,
-			})
-		}
-		slices.SortFunc(agent.Connections, func(a, b *agpl.HTMLClient) int {
-			return slice.Ascending(a.Name, b.Name)
-		})
-
-		data.MissingAgents = append(data.MissingAgents, agent)
-	}
-	slices.SortFunc(data.MissingAgents, func(a, b *agpl.HTMLAgent) int {
-		return slice.Ascending(a.Name, b.Name)
-	})
-
-	return data, nil
-}

enterprise/tailnet/pgcoord_internal_test.go

@@ -1,19 +1,35 @@
 package tailnet
 
 import (
+	"bytes"
 	"context"
+	"flag"
+	"os"
+	"path/filepath"
+	"runtime"
 	"testing"
 	"time"
 
 	"github.com/golang/mock/gomock"
+	"github.com/google/uuid"
+	"github.com/stretchr/testify/require"
+	gProto "google.golang.org/protobuf/proto"
 
 	"cdr.dev/slog"
 	"cdr.dev/slog/sloggers/slogtest"
 
+	"github.com/coder/coder/v2/coderd/database"
 	"github.com/coder/coder/v2/coderd/database/dbmock"
+	"github.com/coder/coder/v2/coderd/database/dbtestutil"
+	"github.com/coder/coder/v2/tailnet/proto"
 	"github.com/coder/coder/v2/testutil"
 )
 
+// UpdateGoldenFiles indicates golden files should be updated.
+// To update the golden files:
+// make update-golden-files
+var UpdateGoldenFiles = flag.Bool("update", false, "update .golden files")
+
 // TestHeartbeat_Cleanup is internal so that we can overwrite the cleanup period and not wait an hour for the timed
 // cleanup.
 func TestHeartbeat_Cleanup(t *testing.T) {
@@ -50,3 +66,122 @@ func TestHeartbeat_Cleanup(t *testing.T) {
 	}
 	close(waitForCleanup)
 }
+
+func TestDebugTemplate(t *testing.T) {
+	t.Parallel()
+	if runtime.GOOS == "windows" {
+		t.Skip("newlines screw up golden files on windows")
+	}
+	c1 := uuid.MustParse("01000000-1111-1111-1111-111111111111")
+	c2 := uuid.MustParse("02000000-1111-1111-1111-111111111111")
+	p1 := uuid.MustParse("01000000-2222-2222-2222-222222222222")
+	p2 := uuid.MustParse("02000000-2222-2222-2222-222222222222")
+	in := HTMLDebug{
+		Coordinators: []*HTMLCoordinator{
+			{
+				ID:           c1,
+				HeartbeatAge: 2 * time.Second,
+			},
+			{
+				ID:           c2,
+				HeartbeatAge: time.Second,
+			},
+		},
+		Peers: []*HTMLPeer{
+			{
+				ID:            p1,
+				CoordinatorID: c1,
+				LastWriteAge:  5 * time.Second,
+				Status:        database.TailnetStatusOk,
+				Node:          `id:1 preferred_derp:999 endpoints:"192.168.0.49:4449"`,
+			},
+			{
+				ID:            p2,
+				CoordinatorID: c2,
+				LastWriteAge:  7 * time.Second,
+				Status:        database.TailnetStatusLost,
+				Node:          `id:2 preferred_derp:999 endpoints:"192.168.0.33:4449"`,
+			},
+		},
+		Tunnels: []*HTMLTunnel{
+			{
+				CoordinatorID: c1,
+				SrcID:         p1,
+				DstID:         p2,
+				LastWriteAge:  3 * time.Second,
+			},
+		},
+	}
+	buf := new(bytes.Buffer)
+	err := debugTempl.Execute(buf, in)
+	require.NoError(t, err)
+	actual := buf.Bytes()
+
+	goldenPath := filepath.Join("testdata", "debug.golden.html")
+	if *UpdateGoldenFiles {
+		t.Logf("update golden file %s", goldenPath)
+		err := os.WriteFile(goldenPath, actual, 0o600)
+		require.NoError(t, err, "update golden file")
+	}
+
+	expected, err := os.ReadFile(goldenPath)
+	require.NoError(t, err, "read golden file, run \"make update-golden-files\" and commit the changes")
+
+	require.Equal(
+		t, string(expected), string(actual),
+		"golden file mismatch: %s, run \"make update-golden-files\", verify and commit the changes",
+		goldenPath,
+	)
+}
+
+func TestGetDebug(t *testing.T) {
+	t.Parallel()
+	if !dbtestutil.WillUsePostgres() {
+		t.Skip("test only with postgres")
+	}
+	store, _ := dbtestutil.NewDB(t)
+	ctx, cancel := context.WithTimeout(context.Background(), testutil.WaitShort)
+	defer cancel()
+
+	coordID := uuid.New()
+	_, err := store.UpsertTailnetCoordinator(ctx, coordID)
+	require.NoError(t, err)
+
+	peerID := uuid.New()
+	node := &proto.Node{PreferredDerp: 44}
+	nodeb, err := gProto.Marshal(node)
+	require.NoError(t, err)
+	_, err = store.UpsertTailnetPeer(ctx, database.UpsertTailnetPeerParams{
+		ID:            peerID,
+		CoordinatorID: coordID,
+		Node:          nodeb,
+		Status:        database.TailnetStatusLost,
+	})
+	require.NoError(t, err)
+
+	dstID := uuid.New()
+	_, err = store.UpsertTailnetTunnel(ctx, database.UpsertTailnetTunnelParams{
+		CoordinatorID: coordID,
+		SrcID:         peerID,
+		DstID:         dstID,
+	})
+	require.NoError(t, err)
+
+	debug, err := getDebug(ctx, store)
+	require.NoError(t, err)
+
+	require.Len(t, debug.Coordinators, 1)
+	require.Len(t, debug.Peers, 1)
+	require.Len(t, debug.Tunnels, 1)
+
+	require.Equal(t, coordID, debug.Coordinators[0].ID)
+
+	require.Equal(t, peerID, debug.Peers[0].ID)
+	require.Equal(t, coordID, debug.Peers[0].CoordinatorID)
+	require.Equal(t, database.TailnetStatusLost, debug.Peers[0].Status)
+	require.Equal(t, node.String(), debug.Peers[0].Node)
+
+	require.Equal(t, coordID, debug.Tunnels[0].CoordinatorID)
+	require.Equal(t, peerID, debug.Tunnels[0].SrcID)
+	require.Equal(t, dstID, debug.Tunnels[0].DstID)
+}

enterprise/tailnet/testdata/debug.golden.html

@@ -0,0 +1,77 @@
+
+<!DOCTYPE html>
+<html>
+	<head>
+		<meta charset="UTF-8">
+		<style>
+th, td {
+  padding-top: 6px;
+  padding-bottom: 6px;
+  padding-left: 10px;
+  padding-right: 10px;
+  text-align: left;
+}
+tr {
+  border-bottom: 1px solid #ddd;
+}
+		</style>
+	</head>
+	<body>
+		<h2 id=coordinators><a href=#coordinators>#</a> coordinators: total 2</h2>
+		<table>
+			<tr style="margin-top:4px">
+				<th>ID</th>
+				<th>Heartbeat Age</th>
+			</tr>
+			<tr style="margin-top:4px">
+				<td>01000000-1111-1111-1111-111111111111</td>
+				<td>2s ago</td>
+			</tr>
+			<tr style="margin-top:4px">
+				<td>02000000-1111-1111-1111-111111111111</td>
+				<td>1s ago</td>
+			</tr>
+		</table>
+
+		<h2 id=peers> <a href=#peers>#</a> peers: total 2 </h2>
+		<table>
+			<tr style="margin-top:4px">
+				<th>ID</th>
+				<th>CoordinatorID</th>
+				<th>Status</th>
+				<th>Last Write Age</th>
+				<th>Node</th>
+			</tr>
+			<tr style="margin-top:4px">
+				<td>01000000-2222-2222-2222-222222222222</td>
+				<td>01000000-1111-1111-1111-111111111111</td>
+				<td>ok</td>
+				<td>5s ago</td>
+				<td style="white-space: pre;"><code>id:1 preferred_derp:999 endpoints:&#34;192.168.0.49:4449&#34;</code></td>
+			</tr>
+			<tr style="margin-top:4px">
+				<td>02000000-2222-2222-2222-222222222222</td>
+				<td>02000000-1111-1111-1111-111111111111</td>
+				<td>lost</td>
+				<td>7s ago</td>
+				<td style="white-space: pre;"><code>id:2 preferred_derp:999 endpoints:&#34;192.168.0.33:4449&#34;</code></td>
+			</tr>
+		</table>
+
+		<h2 id=tunnels><a href=#tunnels>#</a> tunnels: total 1</h2>
+		<table>
+			<tr style="margin-top:4px">
+				<th>SrcID</th>
+				<th>DstID</th>
+				<th>CoordinatorID</th>
+				<th>Last Write Age</th>
+			</tr>
+			<tr style="margin-top:4px">
+				<td>01000000-2222-2222-2222-222222222222</td>
+				<td>02000000-2222-2222-2222-222222222222</td>
+				<td>01000000-1111-1111-1111-111111111111</td>
+				<td>3s ago</td>
+			</tr>
+		</table>
+	</body>
+</html>

site/.eslintignore

@@ -82,6 +82,7 @@ result
 
 # Testdata shouldn't be formatted.
 ../scripts/apitypings/testdata/**/*.ts
+../enterprise/tailnet/testdata/*.golden.html
 
 # Generated files shouldn't be formatted.
 e2e/provisionerGenerated.ts

site/.prettierignore

@@ -82,6 +82,7 @@ result
 
 # Testdata shouldn't be formatted.
 ../scripts/apitypings/testdata/**/*.ts
+../enterprise/tailnet/testdata/*.golden.html
 
 # Generated files shouldn't be formatted.
 e2e/provisionerGenerated.ts