feat(support): add client magicsock and agent prometheus metrics to support bundle (#12604)

* feat(codersdk): add ability to fetch prometheus metrics directly from agent * feat(support): add client magicsock and agent prometheus metrics to support bundle * refactor(support): simplify AgentInfo control flow Co-authored-by: Mathias Fredriksson <mafredri@gmail.com>
2026-09-24 15:04:27 +08:00 · 2024-03-15 15:33:49 +00:00
parent 4d9e6c0134
commit b0c4e7504c
8 changed files with 187 additions and 108 deletions
@@ -7,6 +7,7 @@ import (
 	"encoding/json"
 	"io"
 	"net/http"
+	"net/http/httptest"
 	"strings"

 	"golang.org/x/sync/errgroup"
@@ -57,14 +58,16 @@ type Workspace struct {
 }

 type Agent struct {
-	Agent           *codersdk.WorkspaceAgent                       `json:"agent"`
-	ListeningPorts  *codersdk.WorkspaceAgentListeningPortsResponse `json:"listening_ports"`
-	Logs            []byte                                         `json:"logs"`
-	MagicsockHTML   []byte                                         `json:"magicsock_html"`
-	Manifest        *agentsdk.Manifest                             `json:"manifest"`
-	PeerDiagnostics *tailnet.PeerDiagnostics                       `json:"peer_diagnostics"`
-	PingResult      *ipnstate.PingResult                           `json:"ping_result"`
-	StartupLogs     []codersdk.WorkspaceAgentLog                   `json:"startup_logs"`
+	Agent               *codersdk.WorkspaceAgent                       `json:"agent"`
+	ListeningPorts      *codersdk.WorkspaceAgentListeningPortsResponse `json:"listening_ports"`
+	Logs                []byte                                         `json:"logs"`
+	ClientMagicsockHTML []byte                                         `json:"client_magicsock_html"`
+	AgentMagicsockHTML  []byte                                         `json:"agent_magicsock_html"`
+	Manifest            *agentsdk.Manifest                             `json:"manifest"`
+	PeerDiagnostics     *tailnet.PeerDiagnostics                       `json:"peer_diagnostics"`
+	PingResult          *ipnstate.PingResult                           `json:"ping_result"`
+	Prometheus          []byte                                         `json:"prometheus"`
+	StartupLogs         []codersdk.WorkspaceAgentLog                   `json:"startup_logs"`
 }

 // Deps is a set of dependencies for discovering information
@@ -313,77 +316,10 @@ func AgentInfo(ctx context.Context, client *codersdk.Client, log slog.Logger, ag
 		return nil
 	})

-	conn, err := client.DialWorkspaceAgent(ctx, agentID, &codersdk.DialWorkspaceAgentOptions{
-		Logger:         log.Named("dial-agent"),
-		BlockEndpoints: false,
-	})
-	if err != nil {
-		log.Error(ctx, "dial agent", slog.Error(err))
-	} else {
-		defer func() {
-			if err := conn.Close(); err != nil {
-				log.Error(ctx, "failed to close agent connection", slog.Error(err))
-			}
-			<-conn.Closed()
-		}()
-		if !conn.AwaitReachable(ctx) {
-			log.Error(ctx, "timed out waiting for agent")
-		} else {
-			eg.Go(func() error {
-				_, _, pingRes, err := conn.Ping(ctx)
-				if err != nil {
-					return xerrors.Errorf("ping agent: %w", err)
-				}
-				a.PingResult = pingRes
-				return nil
-			})
-
-			eg.Go(func() error {
-				pds := conn.GetPeerDiagnostics()
-				a.PeerDiagnostics = &pds
-				return nil
-			})
-
-			eg.Go(func() error {
-				msBytes, err := conn.DebugMagicsock(ctx)
-				if err != nil {
-					return xerrors.Errorf("get agent magicsock page: %w", err)
-				}
-				a.MagicsockHTML = msBytes
-				return nil
-			})
-
-			eg.Go(func() error {
-				manifestRes, err := conn.DebugManifest(ctx)
-				if err != nil {
-					return xerrors.Errorf("fetch manifest: %w", err)
-				}
-				if err := json.NewDecoder(bytes.NewReader(manifestRes)).Decode(&a.Manifest); err != nil {
-					return xerrors.Errorf("decode agent manifest: %w", err)
-				}
-
-				return nil
-			})
-
-			eg.Go(func() error {
-				logBytes, err := conn.DebugLogs(ctx)
-				if err != nil {
-					return xerrors.Errorf("fetch coder agent logs: %w", err)
-				}
-				a.Logs = logBytes
-				return nil
-			})
-
-			eg.Go(func() error {
-				lps, err := conn.ListeningPorts(ctx)
-				if err != nil {
-					return xerrors.Errorf("get listening ports: %w", err)
-				}
-				a.ListeningPorts = &lps
-				return nil
-			})
-		}
-	}
+	// to simplify control flow, fetching information directly from
+	// the agent is handled in a separate function
+	closer := connectedAgentInfo(ctx, client, log, agentID, &eg, &a)
+	defer closer()

 	if err := eg.Wait(); err != nil {
 		log.Error(ctx, "fetch agent information", slog.Error(err))
@@ -392,6 +328,108 @@ func AgentInfo(ctx context.Context, client *codersdk.Client, log slog.Logger, ag
 	return a
 }

+func connectedAgentInfo(ctx context.Context, client *codersdk.Client, log slog.Logger, agentID uuid.UUID, eg *errgroup.Group, a *Agent) (closer func()) {
+	conn, err := client.DialWorkspaceAgent(ctx, agentID, &codersdk.DialWorkspaceAgentOptions{
+		Logger:         log.Named("dial-agent"),
+		BlockEndpoints: false,
+	})
+
+	closer = func() {}
+
+	if err != nil {
+		log.Error(ctx, "dial agent", slog.Error(err))
+		return closer
+	}
+
+	if !conn.AwaitReachable(ctx) {
+		log.Error(ctx, "timed out waiting for agent")
+		return closer
+	}
+
+	closer = func() {
+		if err := conn.Close(); err != nil {
+			log.Error(ctx, "failed to close agent connection", slog.Error(err))
+		}
+		<-conn.Closed()
+	}
+
+	eg.Go(func() error {
+		req, err := http.NewRequestWithContext(ctx, http.MethodGet, "http://localhost/", nil)
+		if err != nil {
+			return xerrors.Errorf("create request: %w", err)
+		}
+		rr := httptest.NewRecorder()
+		conn.MagicsockServeHTTPDebug(rr, req)
+		a.ClientMagicsockHTML = rr.Body.Bytes()
+		return nil
+	})
+
+	eg.Go(func() error {
+		promRes, err := conn.PrometheusMetrics(ctx)
+		if err != nil {
+			return xerrors.Errorf("fetch agent prometheus metrics: %w", err)
+		}
+		a.Prometheus = promRes
+		return nil
+	})
+
+	eg.Go(func() error {
+		_, _, pingRes, err := conn.Ping(ctx)
+		if err != nil {
+			return xerrors.Errorf("ping agent: %w", err)
+		}
+		a.PingResult = pingRes
+		return nil
+	})
+
+	eg.Go(func() error {
+		pds := conn.GetPeerDiagnostics()
+		a.PeerDiagnostics = &pds
+		return nil
+	})
+
+	eg.Go(func() error {
+		msBytes, err := conn.DebugMagicsock(ctx)
+		if err != nil {
+			return xerrors.Errorf("get agent magicsock page: %w", err)
+		}
+		a.AgentMagicsockHTML = msBytes
+		return nil
+	})
+
+	eg.Go(func() error {
+		manifestRes, err := conn.DebugManifest(ctx)
+		if err != nil {
+			return xerrors.Errorf("fetch manifest: %w", err)
+		}
+		if err := json.NewDecoder(bytes.NewReader(manifestRes)).Decode(&a.Manifest); err != nil {
+			return xerrors.Errorf("decode agent manifest: %w", err)
+		}
+
+		return nil
+	})
+
+	eg.Go(func() error {
+		logBytes, err := conn.DebugLogs(ctx)
+		if err != nil {
+			return xerrors.Errorf("fetch coder agent logs: %w", err)
+		}
+		a.Logs = logBytes
+		return nil
+	})
+
+	eg.Go(func() error {
+		lps, err := conn.ListeningPorts(ctx)
+		if err != nil {
+			return xerrors.Errorf("get listening ports: %w", err)
+		}
+		a.ListeningPorts = &lps
+		return nil
+	})
+
+	return closer
+}
+
 // Run generates a support bundle with the given dependencies.
 func Run(ctx context.Context, d *Deps) (*Bundle, error) {
 	var b Bundle
@@ -75,9 +75,11 @@ func TestRun(t *testing.T) {
 		assertNotNilNotEmpty(t, bun.Agent.Agent, "agent should be present")
 		assertNotNilNotEmpty(t, bun.Agent.ListeningPorts, "agent listening ports should be present")
 		assertNotNilNotEmpty(t, bun.Agent.Logs, "agent logs should be present")
-		assertNotNilNotEmpty(t, bun.Agent.MagicsockHTML, "agent magicsock should be present")
+		assertNotNilNotEmpty(t, bun.Agent.AgentMagicsockHTML, "agent magicsock should be present")
+		assertNotNilNotEmpty(t, bun.Agent.ClientMagicsockHTML, "client magicsock should be present")
 		assertNotNilNotEmpty(t, bun.Agent.PeerDiagnostics, "agent peer diagnostics should be present")
 		assertNotNilNotEmpty(t, bun.Agent.PingResult, "agent ping result should be present")
+		assertNotNilNotEmpty(t, bun.Agent.Prometheus, "agent prometheus metrics should be present")
 		assertNotNilNotEmpty(t, bun.Agent.StartupLogs, "agent startup logs should be present")
 		assertNotNilNotEmpty(t, bun.Logs, "bundle logs should be present")
 	})