refactor: convert workspacesdk.AgentConn to an interface #19392
Conversation
| if options.PrometheusRegistry == nil { | ||
| options.PrometheusRegistry = prometheus.NewRegistry() | ||
| } | ||
| if options.Authorizer == nil { | ||
| options.Authorizer = rbac.NewCachingAuthorizer(options.PrometheusRegistry) | ||
| if buildinfo.IsDev() { |
There was a problem hiding this comment.
This re-order fixes a nil pointer access when no PrometheusRegistry and Authorizer are passed through options.
| t.Run("PayloadTooLarge", func(t *testing.T) { | ||
| t.Parallel() | ||
|
|
||
| var ( | ||
| ctx = testutil.Context(t, testutil.WaitSuperLong) | ||
| logger = slogtest.Make(t, &slogtest.Options{IgnoreErrors: true}).Leveled(slog.LevelDebug) | ||
| mClock = quartz.NewMock(t) | ||
| updaterTickerTrap = mClock.Trap().TickerFunc("updaterLoop") | ||
| mCtrl = gomock.NewController(t) | ||
| mCCLI = acmock.NewMockContainerCLI(mCtrl) | ||
|
|
||
| client, db = coderdtest.NewWithDatabase(t, &coderdtest.Options{Logger: &logger}) | ||
| user = coderdtest.CreateFirstUser(t, client) | ||
| r = dbfake.WorkspaceBuild(t, db, database.WorkspaceTable{ | ||
| OrganizationID: user.OrganizationID, | ||
| OwnerID: user.UserID, | ||
| }).WithAgent(func(agents []*proto.Agent) []*proto.Agent { | ||
| return agents | ||
| }).Do() | ||
| ) | ||
|
|
||
| // WebSocket limit is 4MiB, so we want to ensure we create _more_ than 4MiB worth of payload. | ||
| // Creating 20,000 fake containers creates a payload of roughly 7MiB. | ||
| var fakeContainers []codersdk.WorkspaceAgentContainer | ||
| for range 20_000 { | ||
| fakeContainers = append(fakeContainers, codersdk.WorkspaceAgentContainer{ | ||
| CreatedAt: time.Now(), | ||
| ID: uuid.NewString(), | ||
| FriendlyName: uuid.NewString(), | ||
| Image: "busybox:latest", | ||
| Labels: map[string]string{ | ||
| agentcontainers.DevcontainerLocalFolderLabel: "/home/coder/project", | ||
| agentcontainers.DevcontainerConfigFileLabel: "/home/coder/project/.devcontainer/devcontainer.json", | ||
| }, | ||
| Running: false, | ||
| Ports: []codersdk.WorkspaceAgentContainerPort{}, | ||
| Status: string(codersdk.WorkspaceAgentDevcontainerStatusRunning), | ||
| Volumes: map[string]string{}, | ||
| }) | ||
| } | ||
|
|
||
| mCCLI.EXPECT().List(gomock.Any()).Return(codersdk.WorkspaceAgentListContainersResponse{Containers: fakeContainers}, nil) | ||
| mCCLI.EXPECT().DetectArchitecture(gomock.Any(), gomock.Any()).Return("<none>", nil).AnyTimes() | ||
|
|
||
| _ = agenttest.New(t, client.URL, r.AgentToken, func(o *agent.Options) { | ||
| o.Logger = logger.Named("agent") | ||
| o.Devcontainers = true | ||
| o.DevcontainerAPIOptions = []agentcontainers.Option{ | ||
| agentcontainers.WithClock(mClock), | ||
| agentcontainers.WithContainerCLI(mCCLI), | ||
| agentcontainers.WithWatcher(watcher.NewNoop()), | ||
| } | ||
| }) | ||
|
|
||
| resources := coderdtest.NewWorkspaceAgentWaiter(t, client, r.Workspace.ID).Wait() | ||
| require.Len(t, resources, 1, "expected one resource") | ||
| require.Len(t, resources[0].Agents, 1, "expected one agent") | ||
| agentID := resources[0].Agents[0].ID | ||
|
|
||
| updaterTickerTrap.MustWait(ctx).MustRelease(ctx) | ||
| defer updaterTickerTrap.Close() | ||
|
|
||
| containers, closer, err := client.WatchWorkspaceAgentContainers(ctx, agentID) | ||
| require.NoError(t, err) | ||
| defer func() { | ||
| closer.Close() | ||
| }() | ||
|
|
||
| select { | ||
| case <-ctx.Done(): | ||
| t.Fail() | ||
| case _, ok := <-containers: | ||
| require.False(t, ok) | ||
| } | ||
| }) |
There was a problem hiding this comment.
This test case is now covered by the test in coderd/workspaceagents_internal_test.go.
We convert `workspacesdk.AgentConn` to an interface and generate a mock for it. This allows writing `coderd` tests that rely on the agent's HTTP api to not have to set up an entire tailnet networking stack.
d579e1d to
ea4de03
Compare
| decodeCh := decoder.Chan() | ||
|
|
||
| // When: We close the `containersCh` | ||
| close(containersCh) |
There was a problem hiding this comment.
Should we have at least one test message sent first to confirm the base case works and then the exit?
There was a problem hiding this comment.
Sounds good to me
| Conn: conn, | ||
| opts: opts, | ||
| } | ||
| } | ||
|
|
||
| // AgentConn represents a connection to a workspace agent. | ||
| // @typescript-ignore AgentConn | ||
| type AgentConn struct { | ||
| type AgentConn interface { |
There was a problem hiding this comment.
Given this is an SDK are we really ready to commit to this just to fix a flake? It feels a little heavy handed
There was a problem hiding this comment.
Without this or something like it it's basically impossible to test Coderd routes that need to interact with Agents without the whole baggage of a real tailnet and a real agent.
Our testing is straining under the weight of spinning up all this stuff to test individual API handlers.
SDKs that we develop need to be designed with mocking in mind, otherwise we can't effectively unit test code that depends on them.
|
@DanielleMaywood could you elaborate on the underlying cause of the flake? |
Sure thing. The underlying cause of the flake is that it is just incredibly slow. On my workspace it could take up to around 10 seconds, and that is dedicated hardware without contention. Running it in CI means results in it often exceeding the You can see in this flake that it took almost 8 seconds just for the tailnet to begin accepting TCP connections (test started at As you can see in the logs, most of the time is spent setting up the agent and the networking stack. We don't need any of this running to test for a specific behavior on The idea behind the change to make |
Fixes coder/internal#907
We convert
workspacesdk.AgentConnto an interface and generate a mock for it. This allows writingcoderdtests that rely on the agent's HTTP api to not have to set up an entire tailnet networking stack.