fix: allow lifecycle code path to retry failed stop jobs (#26277) · coder/coder@0847137

require.Equal(t, stats.Transitions[ws.ID], database.WorkspaceTransitionStop)

})

// FailureTTLStopOK verifies that a workspace whose latest build is a failed

// stop is retried by issuing another stop after the failure TTL elapses.

t.Run("FailureTTLStopOK", func(t *testing.T) {

t.Parallel()

var (

ticker = make(chan time.Time)

statCh = make(chan autobuild.Stats)

logger = slogtest.Make(t, &slogtest.Options{

// We ignore errors here since we expect to fail

// builds.

IgnoreErrors: true,

})

failureTTL = time.Minute

)

client, db, user := coderdenttest.NewWithDatabase(t, &coderdenttest.Options{

Options: &coderdtest.Options{

Logger: &logger,

AutobuildTicker: ticker,

IncludeProvisionerDaemon: true,

AutobuildStats: statCh,

TemplateScheduleStore: schedule.NewEnterpriseTemplateScheduleStore(agplUserQuietHoursScheduleStore(), notifications.NewNoopEnqueuer(), logger, nil),

},

LicenseOptions: &coderdenttest.LicenseOptions{

Features: license.Features{codersdk.FeatureAdvancedTemplateScheduling: 1},

},

})

// The start build succeeds, but the stop build fails. This leaves the

// workspace's latest build as a failed stop.

version := coderdtest.CreateTemplateVersion(t, client, user.OrganizationID, &echo.Responses{

Parse: echo.ParseComplete,

ProvisionPlan: echo.PlanComplete,

ProvisionApplyMap: map[proto.WorkspaceTransition][]*proto.Response{

proto.WorkspaceTransition_START: echo.ApplyComplete,

proto.WorkspaceTransition_STOP: echo.ApplyFailed,

},

})

template := coderdtest.CreateTemplate(t, client, user.OrganizationID, version.ID, func(ctr *codersdk.CreateTemplateRequest) {

ctr.FailureTTLMillis = ptr.Ref[int64](failureTTL.Milliseconds())

})

coderdtest.AwaitTemplateVersionJobCompleted(t, client, version.ID)

ws := coderdtest.CreateWorkspace(t, client, template.ID)

coderdtest.AwaitWorkspaceBuildJobCompleted(t, client, ws.LatestBuild.ID)

ctx := testutil.Context(t, testutil.WaitLong)

stopBuild, err := client.CreateWorkspaceBuild(ctx, ws.ID, codersdk.CreateWorkspaceBuildRequest{

Transition: codersdk.WorkspaceTransitionStop,

})

require.NoError(t, err)

build := coderdtest.AwaitWorkspaceBuildJobCompleted(t, client, stopBuild.ID)

require.Equal(t, codersdk.WorkspaceStatusFailed, build.Status)

require.Equal(t, codersdk.WorkspaceTransitionStop, build.Transition)

tickTime := build.Job.CompletedAt.Add(failureTTL * 2)

p, err := coderdtest.GetProvisionerForTags(db, time.Now(), ws.OrganizationID, nil)

require.NoError(t, err)

coderdtest.UpdateProvisionerLastSeenAt(t, db, p.ID, tickTime)

ticker <- tickTime

stats := <-statCh

// Expect the workspace to be stopped again for breaching failure TTL.

require.Len(t, stats.Transitions, 1)

require.Equal(t, stats.Transitions[ws.ID], database.WorkspaceTransitionStop)

})

t.Run("FailureTTLTooEarly", func(t *testing.T) {

t.Parallel()