fix: cache disconnected agent names in tailnet coordinator debug (#5870)

coadler · web-flow · commit dd8eab56754a · 2023-01-25T21:23:14.000-06:00
diff --git a/coderd/workspaceagents.go b/coderd/workspaceagents.go
@@ -531,6 +531,15 @@ func (api *API) workspaceAgentCoordinate(rw http.ResponseWriter, r *http.Request
 		return
 	}
 
+	owner, err := api.Database.GetUserByID(ctx, workspace.OwnerID)
+	if err != nil {
+		httpapi.Write(ctx, rw, http.StatusBadRequest, codersdk.Response{
+			Message: "Internal error fetching user.",
+			Detail:  err.Error(),
+		})
+		return
+	}
+
 	// Ensure the resource is still valid!
 	// We only accept agents for resources on the latest build.
 	ensureLatestBuild := func() error {
@@ -628,7 +637,9 @@ func (api *API) workspaceAgentCoordinate(rw http.ResponseWriter, r *http.Request
 	closeChan := make(chan struct{})
 	go func() {
 		defer close(closeChan)
-		err := (*api.TailnetCoordinator.Load()).ServeAgent(wsNetConn, workspaceAgent.ID, fmt.Sprintf("%s-%s", workspace.Name, workspaceAgent.Name))
+		err := (*api.TailnetCoordinator.Load()).ServeAgent(wsNetConn, workspaceAgent.ID,
+			fmt.Sprintf("%s-%s-%s", owner.Username, workspace.Name, workspaceAgent.Name),
+		)
 		if err != nil {
 			api.Logger.Warn(ctx, "tailnet coordinator agent error", slog.Error(err))
 			_ = conn.Close(websocket.StatusInternalError, err.Error())
diff --git a/go.mod b/go.mod
@@ -96,6 +96,7 @@ require (
 	github.com/google/uuid v1.3.0
 	github.com/hashicorp/go-reap v0.0.0-20170704170343-bf58d8a43e7b
 	github.com/hashicorp/go-version v1.6.0
+	github.com/hashicorp/golang-lru/v2 v2.0.1
 	github.com/hashicorp/hc-install v0.4.1-0.20220912074615-4487b02cbcbb
 	github.com/hashicorp/hcl/v2 v2.14.0
 	github.com/hashicorp/terraform-config-inspect v0.0.0-20211115214459-90acf1ca460f
diff --git a/go.sum b/go.sum
@@ -1020,6 +1020,8 @@ github.com/hashicorp/go.net v0.0.1/go.mod h1:hjKkEWcCURg++eb33jQU7oqQcI9XDCnUzHA
 github.com/hashicorp/golang-lru v0.5.0/go.mod h1:/m3WP610KZHVQ1SGc6re/UDhFvYD7pJ4Ao+sR/qLZy8=
 github.com/hashicorp/golang-lru v0.5.1/go.mod h1:/m3WP610KZHVQ1SGc6re/UDhFvYD7pJ4Ao+sR/qLZy8=
 github.com/hashicorp/golang-lru v0.5.4/go.mod h1:iADmTwqILo4mZ8BN3D2Q6+9jd8WM5uGBxy+E8yxSoD4=
+github.com/hashicorp/golang-lru/v2 v2.0.1 h1:5pv5N1lT1fjLg2VQ5KWc7kmucp2x/kvFOnxuVTqZ6x4=
+github.com/hashicorp/golang-lru/v2 v2.0.1/go.mod h1:QeFd9opnmA6QUJc5vARoKUSoFhyfM2/ZepoAG6RGpeM=
 github.com/hashicorp/hc-install v0.4.1-0.20220912074615-4487b02cbcbb h1:0AmumMAu6gi5zXEyXvLKDu/HALK+rIcVBZU5XJNyjRM=
 github.com/hashicorp/hc-install v0.4.1-0.20220912074615-4487b02cbcbb/go.mod h1:b3vG+IG40BBISnWiQb9/nHqZI/N3oiunwTtyTDaMGOA=
 github.com/hashicorp/hcl v0.0.0-20170504190234-a4b07c25de5f/go.mod h1:oZtUIOe8dh44I2q6ScRibXws4Ajl+d+nod3AaR9vL5w=
diff --git a/tailnet/coordinator.go b/tailnet/coordinator.go
@@ -14,6 +14,7 @@ import (
 	"time"
 
 	"github.com/google/uuid"
+	lru "github.com/hashicorp/golang-lru/v2"
 	"golang.org/x/exp/slices"
 	"golang.org/x/xerrors"
 	"tailscale.com/tailcfg"
@@ -109,11 +110,17 @@ func ServeCoordinator(conn net.Conn, updateNodes func(node []*Node) error) (func
 // coordinator is incompatible with multiple Coder replicas as all node data is
 // in-memory.
 func NewCoordinator() Coordinator {
+	cache, err := lru.New[uuid.UUID, string](512)
+	if err != nil {
+		panic("make lru cache: " + err.Error())
+	}
+
 	return &coordinator{
 		closed:                   false,
 		nodes:                    map[uuid.UUID]*Node{},
 		agentSockets:             map[uuid.UUID]*trackedConn{},
 		agentToConnectionSockets: map[uuid.UUID]map[uuid.UUID]*trackedConn{},
+		agentNameCache:           cache,
 	}
 }
 
@@ -135,6 +142,10 @@ type coordinator struct {
 	// agentToConnectionSockets maps agent IDs to connection IDs of conns that
 	// are subscribed to updates for that agent.
 	agentToConnectionSockets map[uuid.UUID]map[uuid.UUID]*trackedConn
+
+	// agentNameCache holds a cache of agent names. If one of them disappears,
+	// it's helpful to have a name cached for debugging.
+	agentNameCache *lru.Cache[uuid.UUID, string]
 }
 
 type trackedConn struct {
@@ -288,6 +299,8 @@ func (c *coordinator) ServeAgent(conn net.Conn, id uuid.UUID, name string) error
 		return xerrors.New("coordinator is closed")
 	}
 
+	c.agentNameCache.Add(id, name)
+
 	sockets, ok := c.agentToConnectionSockets[id]
 	if ok {
 		// Publish all nodes that want to connect to the
@@ -532,7 +545,13 @@ func (c *coordinator) ServeHTTPDebug(w http.ResponseWriter, _ *http.Request) {
 		fmt.Fprintln(w, "<ul>")
 
 		for _, agentConns := range missingAgents {
-			fmt.Fprintf(w, "<li style=\"margin-top:4px\"><b>unknown</b> (<code>%s</code>): created ? ago, write ? ago, overwrites ? </li>\n",
+			agentName, ok := c.agentNameCache.Get(agentConns.id)
+			if !ok {
+				agentName = "unknown"
+			}
+
+			fmt.Fprintf(w, "<li style=\"margin-top:4px\"><b>%s</b> (<code>%s</code>): created ? ago, write ? ago, overwrites ? </li>\n",
+				agentName,
 				agentConns.id.String(),
 			)
 
