feat: cancel pending prebuilds from non-active template versions (#20387)

## Description

This PR introduces an optimization to automatically cancel pending
prebuild-related jobs from non-active template versions in the
reconciliation loop.

## Problem

Currently, when a template is configured with more prebuild instances
than available provisioners, the provisioner queue can become flooded
with pending prebuild jobs. This issue is worsened when
provisioning/deprovisioning operations take a long time.

When the prebuild reconciliation loop generates jobs faster than
provisioners can process them, pending jobs accumulate in the queue.
Since prebuilt workspaces should always run the latest active template
version, pending prebuild jobs from non-active versions become obsolete
once a new version is promoted.

## Solution

The reconciliation loop cancels pending prebuild-related jobs from
non-active template versions that match the following criteria:

* Build number: 1 (initial build created by the reconciliation loop)
* Job status: `pending`
* Not yet picked up by a provisioner (`worker_id` is `NULL`)
* Owned by the prebuilds system user
* Workspace transition: `start`

This prevents the queue from being cluttered with stale prebuild jobs
that would provision workspaces on an outdated template version that
would consequently need to be deprovisioned.

## Changes

* Added new SQL query `CountPendingNonActivePrebuilds` to identify
presets with pending jobs from non-active versions
* Added new SQL query `UpdatePrebuildProvisionerJobWithCancel` to cancel
jobs for a specific preset
* New reconciliation action type `ActionTypeCancelPending` handles the
cancellation logic
* Cancellation is non-blocking: failures to cancel prebuild jobs are
logged as errors and don't prevent other reconciliation actions

## Follow-up PR

Canceling pending prebuild jobs leaves workspaces in a Canceled state.
While no Terraform resources need to be destroyed (since jobs were
canceled before provisioning started), these database records should
still be cleaned up. This will be addressed in a follow-up PR.

Closes: https://github.com/coder/coder/issues/20242
This commit is contained in:
Susana Ferreira
2025-10-24 15:27:49 +01:00
committed by GitHub
parent c301a0d804
commit f6e86c6fdb
13 changed files with 1016 additions and 63 deletions
+15
View File
@@ -1512,6 +1512,13 @@ func (q *querier) CountInProgressPrebuilds(ctx context.Context) ([]database.Coun
return q.db.CountInProgressPrebuilds(ctx)
}
func (q *querier) CountPendingNonActivePrebuilds(ctx context.Context) ([]database.CountPendingNonActivePrebuildsRow, error) {
if err := q.authorizeContext(ctx, policy.ActionRead, rbac.ResourceWorkspace.All()); err != nil {
return nil, err
}
return q.db.CountPendingNonActivePrebuilds(ctx)
}
func (q *querier) CountUnreadInboxNotificationsByUserID(ctx context.Context, userID uuid.UUID) (int64, error) {
if err := q.authorizeContext(ctx, policy.ActionRead, rbac.ResourceInboxNotification.WithOwner(userID.String())); err != nil {
return 0, err
@@ -4875,6 +4882,14 @@ func (q *querier) UpdateOrganizationDeletedByID(ctx context.Context, arg databas
return deleteQ(q.log, q.auth, q.db.GetOrganizationByID, deleteF)(ctx, arg.ID)
}
func (q *querier) UpdatePrebuildProvisionerJobWithCancel(ctx context.Context, arg database.UpdatePrebuildProvisionerJobWithCancelParams) ([]uuid.UUID, error) {
// Prebuild operation for canceling pending prebuild jobs from non-active template versions
if err := q.authorizeContext(ctx, policy.ActionUpdate, rbac.ResourcePrebuiltWorkspace); err != nil {
return []uuid.UUID{}, err
}
return q.db.UpdatePrebuildProvisionerJobWithCancel(ctx, arg)
}
func (q *querier) UpdatePresetPrebuildStatus(ctx context.Context, arg database.UpdatePresetPrebuildStatusParams) error {
preset, err := q.db.GetPresetByID(ctx, arg.PresetID)
if err != nil {
+14
View File
@@ -641,6 +641,16 @@ func (s *MethodTestSuite) TestProvisionerJob() {
dbm.EXPECT().UpdateProvisionerJobWithCancelByID(gomock.Any(), arg).Return(nil).AnyTimes()
check.Args(arg).Asserts(v.RBACObject(tpl), []policy.Action{policy.ActionRead, policy.ActionUpdate}).Returns()
}))
s.Run("UpdatePrebuildProvisionerJobWithCancel", s.Mocked(func(dbm *dbmock.MockStore, faker *gofakeit.Faker, check *expects) {
arg := database.UpdatePrebuildProvisionerJobWithCancelParams{
PresetID: uuid.NullUUID{UUID: uuid.New(), Valid: true},
Now: dbtime.Now(),
}
jobIDs := []uuid.UUID{uuid.New(), uuid.New()}
dbm.EXPECT().UpdatePrebuildProvisionerJobWithCancel(gomock.Any(), arg).Return(jobIDs, nil).AnyTimes()
check.Args(arg).Asserts(rbac.ResourcePrebuiltWorkspace, policy.ActionUpdate).Returns(jobIDs)
}))
s.Run("GetProvisionerJobsByIDs", s.Mocked(func(dbm *dbmock.MockStore, faker *gofakeit.Faker, check *expects) {
org := testutil.Fake(s.T(), faker, database.Organization{})
org2 := testutil.Fake(s.T(), faker, database.Organization{})
@@ -3758,6 +3768,10 @@ func (s *MethodTestSuite) TestPrebuilds() {
dbm.EXPECT().CountInProgressPrebuilds(gomock.Any()).Return([]database.CountInProgressPrebuildsRow{}, nil).AnyTimes()
check.Args().Asserts(rbac.ResourceWorkspace.All(), policy.ActionRead)
}))
s.Run("CountPendingNonActivePrebuilds", s.Mocked(func(dbm *dbmock.MockStore, _ *gofakeit.Faker, check *expects) {
dbm.EXPECT().CountPendingNonActivePrebuilds(gomock.Any()).Return([]database.CountPendingNonActivePrebuildsRow{}, nil).AnyTimes()
check.Args().Asserts(rbac.ResourceWorkspace.All(), policy.ActionRead)
}))
s.Run("GetPresetsAtFailureLimit", s.Mocked(func(dbm *dbmock.MockStore, _ *gofakeit.Faker, check *expects) {
dbm.EXPECT().GetPresetsAtFailureLimit(gomock.Any(), int64(0)).Return([]database.GetPresetsAtFailureLimitRow{}, nil).AnyTimes()
check.Args(int64(0)).Asserts(rbac.ResourceTemplate.All(), policy.ActionViewInsights)
+40 -9
View File
@@ -55,14 +55,10 @@ type WorkspaceBuildBuilder struct {
resources []*sdkproto.Resource
params []database.WorkspaceBuildParameter
agentToken string
dispo workspaceBuildDisposition
jobStatus database.ProvisionerJobStatus
taskAppID uuid.UUID
}
type workspaceBuildDisposition struct {
starting bool
}
// WorkspaceBuild generates a workspace build for the provided workspace.
// Pass a database.Workspace{} with a nil ID to also generate a new workspace.
// Omitting the template ID on a workspace will also generate a new template
@@ -145,8 +141,17 @@ func (b WorkspaceBuildBuilder) WithTask(seed *sdkproto.App) WorkspaceBuildBuilde
}
func (b WorkspaceBuildBuilder) Starting() WorkspaceBuildBuilder {
//nolint: revive // returns modified struct
b.dispo.starting = true
b.jobStatus = database.ProvisionerJobStatusRunning
return b
}
func (b WorkspaceBuildBuilder) Pending() WorkspaceBuildBuilder {
b.jobStatus = database.ProvisionerJobStatusPending
return b
}
func (b WorkspaceBuildBuilder) Canceled() WorkspaceBuildBuilder {
b.jobStatus = database.ProvisionerJobStatusCanceled
return b
}
@@ -231,7 +236,11 @@ func (b WorkspaceBuildBuilder) Do() WorkspaceResponse {
require.NoError(b.t, err, "insert job")
b.logger.Debug(context.Background(), "inserted provisioner job", slog.F("job_id", job.ID))
if b.dispo.starting {
switch b.jobStatus {
case database.ProvisionerJobStatusPending:
// Provisioner jobs are created in 'pending' status
b.logger.Debug(context.Background(), "pending the provisioner job")
case database.ProvisionerJobStatusRunning:
// might need to do this multiple times if we got a template version
// import job as well
b.logger.Debug(context.Background(), "looping to acquire provisioner job")
@@ -255,7 +264,23 @@ func (b WorkspaceBuildBuilder) Do() WorkspaceResponse {
break
}
}
} else {
case database.ProvisionerJobStatusCanceled:
// Set provisioner job status to 'canceled'
b.logger.Debug(context.Background(), "canceling the provisioner job")
err = b.db.UpdateProvisionerJobWithCancelByID(ownerCtx, database.UpdateProvisionerJobWithCancelByIDParams{
ID: jobID,
CanceledAt: sql.NullTime{
Time: dbtime.Now(),
Valid: true,
},
CompletedAt: sql.NullTime{
Time: dbtime.Now(),
Valid: true,
},
})
require.NoError(b.t, err, "cancel job")
default:
// By default, consider jobs in 'succeeded' status
b.logger.Debug(context.Background(), "completing the provisioner job")
err = b.db.UpdateProvisionerJobWithCompleteByID(ownerCtx, database.UpdateProvisionerJobWithCompleteByIDParams{
ID: job.ID,
@@ -571,6 +596,12 @@ func (t TemplateVersionBuilder) Do() TemplateVersionResponse {
t.params[i] = dbgen.TemplateVersionParameter(t.t, t.db, param)
}
// Update response with template and version
if resp.Template.ID == uuid.Nil && version.TemplateID.Valid {
template, err := t.db.GetTemplateByID(ownerCtx, version.TemplateID.UUID)
require.NoError(t.t, err)
resp.Template = template
}
resp.TemplateVersion = version
return resp
}
+14
View File
@@ -214,6 +214,13 @@ func (m queryMetricsStore) CountInProgressPrebuilds(ctx context.Context) ([]data
return r0, r1
}
func (m queryMetricsStore) CountPendingNonActivePrebuilds(ctx context.Context) ([]database.CountPendingNonActivePrebuildsRow, error) {
start := time.Now()
r0, r1 := m.s.CountPendingNonActivePrebuilds(ctx)
m.queryLatencies.WithLabelValues("CountPendingNonActivePrebuilds").Observe(time.Since(start).Seconds())
return r0, r1
}
func (m queryMetricsStore) CountUnreadInboxNotificationsByUserID(ctx context.Context, userID uuid.UUID) (int64, error) {
start := time.Now()
r0, r1 := m.s.CountUnreadInboxNotificationsByUserID(ctx, userID)
@@ -3000,6 +3007,13 @@ func (m queryMetricsStore) UpdateOrganizationDeletedByID(ctx context.Context, ar
return r0
}
func (m queryMetricsStore) UpdatePrebuildProvisionerJobWithCancel(ctx context.Context, arg database.UpdatePrebuildProvisionerJobWithCancelParams) ([]uuid.UUID, error) {
start := time.Now()
r0, r1 := m.s.UpdatePrebuildProvisionerJobWithCancel(ctx, arg)
m.queryLatencies.WithLabelValues("UpdatePrebuildProvisionerJobWithCancel").Observe(time.Since(start).Seconds())
return r0, r1
}
func (m queryMetricsStore) UpdatePresetPrebuildStatus(ctx context.Context, arg database.UpdatePresetPrebuildStatusParams) error {
start := time.Now()
r0 := m.s.UpdatePresetPrebuildStatus(ctx, arg)
+30
View File
@@ -352,6 +352,21 @@ func (mr *MockStoreMockRecorder) CountInProgressPrebuilds(ctx any) *gomock.Call
return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "CountInProgressPrebuilds", reflect.TypeOf((*MockStore)(nil).CountInProgressPrebuilds), ctx)
}
// CountPendingNonActivePrebuilds mocks base method.
func (m *MockStore) CountPendingNonActivePrebuilds(ctx context.Context) ([]database.CountPendingNonActivePrebuildsRow, error) {
m.ctrl.T.Helper()
ret := m.ctrl.Call(m, "CountPendingNonActivePrebuilds", ctx)
ret0, _ := ret[0].([]database.CountPendingNonActivePrebuildsRow)
ret1, _ := ret[1].(error)
return ret0, ret1
}
// CountPendingNonActivePrebuilds indicates an expected call of CountPendingNonActivePrebuilds.
func (mr *MockStoreMockRecorder) CountPendingNonActivePrebuilds(ctx any) *gomock.Call {
mr.mock.ctrl.T.Helper()
return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "CountPendingNonActivePrebuilds", reflect.TypeOf((*MockStore)(nil).CountPendingNonActivePrebuilds), ctx)
}
// CountUnreadInboxNotificationsByUserID mocks base method.
func (m *MockStore) CountUnreadInboxNotificationsByUserID(ctx context.Context, userID uuid.UUID) (int64, error) {
m.ctrl.T.Helper()
@@ -6451,6 +6466,21 @@ func (mr *MockStoreMockRecorder) UpdateOrganizationDeletedByID(ctx, arg any) *go
return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "UpdateOrganizationDeletedByID", reflect.TypeOf((*MockStore)(nil).UpdateOrganizationDeletedByID), ctx, arg)
}
// UpdatePrebuildProvisionerJobWithCancel mocks base method.
func (m *MockStore) UpdatePrebuildProvisionerJobWithCancel(ctx context.Context, arg database.UpdatePrebuildProvisionerJobWithCancelParams) ([]uuid.UUID, error) {
m.ctrl.T.Helper()
ret := m.ctrl.Call(m, "UpdatePrebuildProvisionerJobWithCancel", ctx, arg)
ret0, _ := ret[0].([]uuid.UUID)
ret1, _ := ret[1].(error)
return ret0, ret1
}
// UpdatePrebuildProvisionerJobWithCancel indicates an expected call of UpdatePrebuildProvisionerJobWithCancel.
func (mr *MockStoreMockRecorder) UpdatePrebuildProvisionerJobWithCancel(ctx, arg any) *gomock.Call {
mr.mock.ctrl.T.Helper()
return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "UpdatePrebuildProvisionerJobWithCancel", reflect.TypeOf((*MockStore)(nil).UpdatePrebuildProvisionerJobWithCancel), ctx, arg)
}
// UpdatePresetPrebuildStatus mocks base method.
func (m *MockStore) UpdatePresetPrebuildStatus(ctx context.Context, arg database.UpdatePresetPrebuildStatusParams) error {
m.ctrl.T.Helper()
+7 -1
View File
@@ -68,8 +68,10 @@ type sqlcQuerier interface {
CountAuditLogs(ctx context.Context, arg CountAuditLogsParams) (int64, error)
CountConnectionLogs(ctx context.Context, arg CountConnectionLogsParams) (int64, error)
// CountInProgressPrebuilds returns the number of in-progress prebuilds, grouped by preset ID and transition.
// Prebuild considered in-progress if it's in the "starting", "stopping", or "deleting" state.
// Prebuild considered in-progress if it's in the "pending", "starting", "stopping", or "deleting" state.
CountInProgressPrebuilds(ctx context.Context) ([]CountInProgressPrebuildsRow, error)
// CountPendingNonActivePrebuilds returns the number of pending prebuilds for non-active template versions
CountPendingNonActivePrebuilds(ctx context.Context) ([]CountPendingNonActivePrebuildsRow, error)
CountUnreadInboxNotificationsByUserID(ctx context.Context, userID uuid.UUID) (int64, error)
CreateUserSecret(ctx context.Context, arg CreateUserSecretParams) (UserSecret, error)
CustomRoles(ctx context.Context, arg CustomRolesParams) ([]CustomRole, error)
@@ -647,6 +649,10 @@ type sqlcQuerier interface {
UpdateOAuth2ProviderAppSecretByID(ctx context.Context, arg UpdateOAuth2ProviderAppSecretByIDParams) (OAuth2ProviderAppSecret, error)
UpdateOrganization(ctx context.Context, arg UpdateOrganizationParams) (Organization, error)
UpdateOrganizationDeletedByID(ctx context.Context, arg UpdateOrganizationDeletedByIDParams) error
// Cancels all pending provisioner jobs for prebuilt workspaces on a specific preset from an
// inactive template version.
// This is an optimization to clean up stale pending jobs.
UpdatePrebuildProvisionerJobWithCancel(ctx context.Context, arg UpdatePrebuildProvisionerJobWithCancelParams) ([]uuid.UUID, error)
UpdatePresetPrebuildStatus(ctx context.Context, arg UpdatePresetPrebuildStatusParams) error
UpdateProvisionerDaemonLastSeenAt(ctx context.Context, arg UpdateProvisionerDaemonLastSeenAtParams) error
UpdateProvisionerJobByID(ctx context.Context, arg UpdateProvisionerJobByIDParams) error
+112 -1
View File
@@ -7924,7 +7924,7 @@ type CountInProgressPrebuildsRow struct {
}
// CountInProgressPrebuilds returns the number of in-progress prebuilds, grouped by preset ID and transition.
// Prebuild considered in-progress if it's in the "starting", "stopping", or "deleting" state.
// Prebuild considered in-progress if it's in the "pending", "starting", "stopping", or "deleting" state.
func (q *sqlQuerier) CountInProgressPrebuilds(ctx context.Context) ([]CountInProgressPrebuildsRow, error) {
rows, err := q.db.QueryContext(ctx, countInProgressPrebuilds)
if err != nil {
@@ -7954,6 +7954,58 @@ func (q *sqlQuerier) CountInProgressPrebuilds(ctx context.Context) ([]CountInPro
return items, nil
}
const countPendingNonActivePrebuilds = `-- name: CountPendingNonActivePrebuilds :many
SELECT
wpb.template_version_preset_id AS preset_id,
COUNT(*)::int AS count
FROM workspace_prebuild_builds wpb
INNER JOIN provisioner_jobs pj ON pj.id = wpb.job_id
INNER JOIN workspaces w ON w.id = wpb.workspace_id
INNER JOIN templates t ON t.id = w.template_id
WHERE
wpb.template_version_id != t.active_version_id
-- Only considers initial builds, i.e. created by the reconciliation loop
AND wpb.build_number = 1
-- Only consider 'start' transitions (provisioning), not 'stop'/'delete' (deprovisioning)
-- Deprovisioning jobs should complete naturally as they're already cleaning up resources
AND wpb.transition = 'start'::workspace_transition
-- Pending jobs that have not yet been picked up by a provisioner
AND pj.job_status = 'pending'::provisioner_job_status
AND pj.worker_id IS NULL
AND pj.canceled_at IS NULL
AND pj.completed_at IS NULL
GROUP BY wpb.template_version_preset_id
`
type CountPendingNonActivePrebuildsRow struct {
PresetID uuid.NullUUID `db:"preset_id" json:"preset_id"`
Count int32 `db:"count" json:"count"`
}
// CountPendingNonActivePrebuilds returns the number of pending prebuilds for non-active template versions
func (q *sqlQuerier) CountPendingNonActivePrebuilds(ctx context.Context) ([]CountPendingNonActivePrebuildsRow, error) {
rows, err := q.db.QueryContext(ctx, countPendingNonActivePrebuilds)
if err != nil {
return nil, err
}
defer rows.Close()
var items []CountPendingNonActivePrebuildsRow
for rows.Next() {
var i CountPendingNonActivePrebuildsRow
if err := rows.Scan(&i.PresetID, &i.Count); err != nil {
return nil, err
}
items = append(items, i)
}
if err := rows.Close(); err != nil {
return nil, err
}
if err := rows.Err(); err != nil {
return nil, err
}
return items, nil
}
const findMatchingPresetID = `-- name: FindMatchingPresetID :one
WITH provided_params AS (
SELECT
@@ -8396,6 +8448,65 @@ func (q *sqlQuerier) GetTemplatePresetsWithPrebuilds(ctx context.Context, templa
return items, nil
}
const updatePrebuildProvisionerJobWithCancel = `-- name: UpdatePrebuildProvisionerJobWithCancel :many
UPDATE provisioner_jobs
SET
canceled_at = $1::timestamptz,
completed_at = $1::timestamptz
WHERE id IN (
SELECT pj.id
FROM provisioner_jobs pj
INNER JOIN workspace_prebuild_builds wpb ON wpb.job_id = pj.id
INNER JOIN workspaces w ON w.id = wpb.workspace_id
INNER JOIN templates t ON t.id = w.template_id
WHERE
wpb.template_version_id != t.active_version_id
AND wpb.template_version_preset_id = $2
-- Only considers initial builds, i.e. created by the reconciliation loop
AND wpb.build_number = 1
-- Only consider 'start' transitions (provisioning), not 'stop'/'delete' (deprovisioning)
-- Deprovisioning jobs should complete naturally as they're already cleaning up resources
AND wpb.transition = 'start'::workspace_transition
-- Pending jobs that have not yet been picked up by a provisioner
AND pj.job_status = 'pending'::provisioner_job_status
AND pj.worker_id IS NULL
AND pj.canceled_at IS NULL
AND pj.completed_at IS NULL
)
RETURNING id
`
type UpdatePrebuildProvisionerJobWithCancelParams struct {
Now time.Time `db:"now" json:"now"`
PresetID uuid.NullUUID `db:"preset_id" json:"preset_id"`
}
// Cancels all pending provisioner jobs for prebuilt workspaces on a specific preset from an
// inactive template version.
// This is an optimization to clean up stale pending jobs.
func (q *sqlQuerier) UpdatePrebuildProvisionerJobWithCancel(ctx context.Context, arg UpdatePrebuildProvisionerJobWithCancelParams) ([]uuid.UUID, error) {
rows, err := q.db.QueryContext(ctx, updatePrebuildProvisionerJobWithCancel, arg.Now, arg.PresetID)
if err != nil {
return nil, err
}
defer rows.Close()
var items []uuid.UUID
for rows.Next() {
var id uuid.UUID
if err := rows.Scan(&id); err != nil {
return nil, err
}
items = append(items, id)
}
if err := rows.Close(); err != nil {
return nil, err
}
if err := rows.Err(); err != nil {
return nil, err
}
return items, nil
}
const getActivePresetPrebuildSchedules = `-- name: GetActivePresetPrebuildSchedules :many
SELECT
tvpps.id, tvpps.preset_id, tvpps.cron_expression, tvpps.desired_instances
+54 -1
View File
@@ -121,7 +121,7 @@ ORDER BY latest_prebuilds.id;
-- name: CountInProgressPrebuilds :many
-- CountInProgressPrebuilds returns the number of in-progress prebuilds, grouped by preset ID and transition.
-- Prebuild considered in-progress if it's in the "starting", "stopping", or "deleting" state.
-- Prebuild considered in-progress if it's in the "pending", "starting", "stopping", or "deleting" state.
SELECT t.id AS template_id, wpb.template_version_id, wpb.transition, COUNT(wpb.transition)::int AS count, wlb.template_version_preset_id as preset_id
FROM workspace_latest_builds wlb
INNER JOIN workspace_prebuild_builds wpb ON wpb.id = wlb.id
@@ -272,3 +272,56 @@ FROM preset_matches pm
WHERE pm.total_preset_params = pm.matching_params -- All preset parameters must match
ORDER BY pm.total_preset_params DESC -- Return the preset with the most parameters
LIMIT 1;
-- name: CountPendingNonActivePrebuilds :many
-- CountPendingNonActivePrebuilds returns the number of pending prebuilds for non-active template versions
SELECT
wpb.template_version_preset_id AS preset_id,
COUNT(*)::int AS count
FROM workspace_prebuild_builds wpb
INNER JOIN provisioner_jobs pj ON pj.id = wpb.job_id
INNER JOIN workspaces w ON w.id = wpb.workspace_id
INNER JOIN templates t ON t.id = w.template_id
WHERE
wpb.template_version_id != t.active_version_id
-- Only considers initial builds, i.e. created by the reconciliation loop
AND wpb.build_number = 1
-- Only consider 'start' transitions (provisioning), not 'stop'/'delete' (deprovisioning)
-- Deprovisioning jobs should complete naturally as they're already cleaning up resources
AND wpb.transition = 'start'::workspace_transition
-- Pending jobs that have not yet been picked up by a provisioner
AND pj.job_status = 'pending'::provisioner_job_status
AND pj.worker_id IS NULL
AND pj.canceled_at IS NULL
AND pj.completed_at IS NULL
GROUP BY wpb.template_version_preset_id;
-- name: UpdatePrebuildProvisionerJobWithCancel :many
-- Cancels all pending provisioner jobs for prebuilt workspaces on a specific preset from an
-- inactive template version.
-- This is an optimization to clean up stale pending jobs.
UPDATE provisioner_jobs
SET
canceled_at = @now::timestamptz,
completed_at = @now::timestamptz
WHERE id IN (
SELECT pj.id
FROM provisioner_jobs pj
INNER JOIN workspace_prebuild_builds wpb ON wpb.job_id = pj.id
INNER JOIN workspaces w ON w.id = wpb.workspace_id
INNER JOIN templates t ON t.id = w.template_id
WHERE
wpb.template_version_id != t.active_version_id
AND wpb.template_version_preset_id = @preset_id
-- Only considers initial builds, i.e. created by the reconciliation loop
AND wpb.build_number = 1
-- Only consider 'start' transitions (provisioning), not 'stop'/'delete' (deprovisioning)
-- Deprovisioning jobs should complete naturally as they're already cleaning up resources
AND wpb.transition = 'start'::workspace_transition
-- Pending jobs that have not yet been picked up by a provisioner
AND pj.job_status = 'pending'::provisioner_job_status
AND pj.worker_id IS NULL
AND pj.canceled_at IS NULL
AND pj.completed_at IS NULL
)
RETURNING id;
+15 -2
View File
@@ -8,10 +8,9 @@ import (
"cdr.dev/slog"
"github.com/coder/quartz"
"github.com/coder/coder/v2/coderd/database"
"github.com/coder/coder/v2/coderd/util/slice"
"github.com/coder/quartz"
)
// GlobalSnapshot represents a full point-in-time snapshot of state relating to prebuilds across all templates.
@@ -20,6 +19,7 @@ type GlobalSnapshot struct {
PrebuildSchedules []database.TemplateVersionPresetPrebuildSchedule
RunningPrebuilds []database.GetRunningPrebuiltWorkspacesRow
PrebuildsInProgress []database.CountInProgressPrebuildsRow
PendingPrebuilds []database.CountPendingNonActivePrebuildsRow
Backoffs []database.GetPresetsBackoffRow
HardLimitedPresetsMap map[uuid.UUID]database.GetPresetsAtFailureLimitRow
clock quartz.Clock
@@ -31,6 +31,7 @@ func NewGlobalSnapshot(
prebuildSchedules []database.TemplateVersionPresetPrebuildSchedule,
runningPrebuilds []database.GetRunningPrebuiltWorkspacesRow,
prebuildsInProgress []database.CountInProgressPrebuildsRow,
pendingPrebuilds []database.CountPendingNonActivePrebuildsRow,
backoffs []database.GetPresetsBackoffRow,
hardLimitedPresets []database.GetPresetsAtFailureLimitRow,
clock quartz.Clock,
@@ -46,6 +47,7 @@ func NewGlobalSnapshot(
PrebuildSchedules: prebuildSchedules,
RunningPrebuilds: runningPrebuilds,
PrebuildsInProgress: prebuildsInProgress,
PendingPrebuilds: pendingPrebuilds,
Backoffs: backoffs,
HardLimitedPresetsMap: hardLimitedPresetsMap,
clock: clock,
@@ -76,10 +78,20 @@ func (s GlobalSnapshot) FilterByPreset(presetID uuid.UUID) (*PresetSnapshot, err
// Separate running workspaces into non-expired and expired based on the preset's TTL
nonExpired, expired := filterExpiredWorkspaces(preset, running)
// Includes in-progress prebuilds only for active template versions.
// In-progress prebuilds correspond to workspace statuses: 'pending', 'starting', 'stopping', and 'deleting'
inProgress := slice.Filter(s.PrebuildsInProgress, func(prebuild database.CountInProgressPrebuildsRow) bool {
return prebuild.PresetID.UUID == preset.ID
})
// Includes count of pending prebuilds only for non-active template versions
pendingCount := 0
if found, ok := slice.Find(s.PendingPrebuilds, func(prebuild database.CountPendingNonActivePrebuildsRow) bool {
return prebuild.PresetID.UUID == preset.ID
}); ok {
pendingCount = int(found.Count)
}
var backoffPtr *database.GetPresetsBackoffRow
backoff, found := slice.Find(s.Backoffs, func(row database.GetPresetsBackoffRow) bool {
return row.PresetID == preset.ID
@@ -96,6 +108,7 @@ func (s GlobalSnapshot) FilterByPreset(presetID uuid.UUID) (*PresetSnapshot, err
nonExpired,
expired,
inProgress,
pendingCount,
backoffPtr,
isHardLimited,
s.clock,
+30 -12
View File
@@ -34,6 +34,9 @@ const (
// ActionTypeBackoff indicates that prebuild creation should be delayed.
ActionTypeBackoff
// ActionTypeCancelPending indicates that pending prebuilds should be canceled.
ActionTypeCancelPending
)
// PresetSnapshot is a filtered view of GlobalSnapshot focused on a single preset.
@@ -49,6 +52,7 @@ type PresetSnapshot struct {
Running []database.GetRunningPrebuiltWorkspacesRow
Expired []database.GetRunningPrebuiltWorkspacesRow
InProgress []database.CountInProgressPrebuildsRow
PendingCount int
Backoff *database.GetPresetsBackoffRow
IsHardLimited bool
clock quartz.Clock
@@ -61,6 +65,7 @@ func NewPresetSnapshot(
running []database.GetRunningPrebuiltWorkspacesRow,
expired []database.GetRunningPrebuiltWorkspacesRow,
inProgress []database.CountInProgressPrebuildsRow,
pendingCount int,
backoff *database.GetPresetsBackoffRow,
isHardLimited bool,
clock quartz.Clock,
@@ -72,6 +77,7 @@ func NewPresetSnapshot(
Running: running,
Expired: expired,
InProgress: inProgress,
PendingCount: pendingCount,
Backoff: backoff,
IsHardLimited: isHardLimited,
clock: clock,
@@ -115,7 +121,7 @@ type ReconciliationActions struct {
}
func (ra *ReconciliationActions) IsNoop() bool {
return ra.Create == 0 && len(ra.DeleteIDs) == 0 && ra.BackoffUntil.IsZero()
return ra.ActionType != ActionTypeCancelPending && ra.Create == 0 && len(ra.DeleteIDs) == 0 && ra.BackoffUntil.IsZero()
}
// MatchesCron interprets a cron spec as a continuous time range,
@@ -345,18 +351,30 @@ func (p PresetSnapshot) handleActiveTemplateVersion() (actions []*Reconciliation
return actions, nil
}
// handleInactiveTemplateVersion deletes all running prebuilds except those already being deleted
// to avoid duplicate deletion attempts.
func (p PresetSnapshot) handleInactiveTemplateVersion() ([]*ReconciliationActions, error) {
prebuildsToDelete := len(p.Running)
deleteIDs := p.getOldestPrebuildIDs(prebuildsToDelete)
// handleInactiveTemplateVersion handles prebuilds from inactive template versions:
// 1. If the preset has pending prebuild jobs from an inactive template version, create a cancel reconciliation action.
// This cancels all pending prebuild jobs for this preset's template version.
// 2. If the preset has prebuilt workspaces currently running from an inactive template version,
// create a delete reconciliation action to remove all running prebuilt workspaces.
func (p PresetSnapshot) handleInactiveTemplateVersion() (actions []*ReconciliationActions, err error) {
// Cancel pending initial prebuild jobs from inactive version
if p.PendingCount > 0 {
actions = append(actions,
&ReconciliationActions{
ActionType: ActionTypeCancelPending,
})
}
return []*ReconciliationActions{
{
ActionType: ActionTypeDelete,
DeleteIDs: deleteIDs,
},
}, nil
// Delete prebuilds running in inactive version
deleteIDs := p.getOldestPrebuildIDs(len(p.Running))
if len(deleteIDs) > 0 {
actions = append(actions,
&ReconciliationActions{
ActionType: ActionTypeDelete,
DeleteIDs: deleteIDs,
})
}
return actions, nil
}
// needsBackoffPeriod checks if we should delay prebuild creation due to recent failures.
+82 -15
View File
@@ -6,16 +6,14 @@ import (
"testing"
"time"
"github.com/coder/coder/v2/testutil"
"github.com/google/uuid"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
"github.com/coder/quartz"
"github.com/coder/coder/v2/coderd/database"
"github.com/coder/coder/v2/coderd/prebuilds"
"github.com/coder/coder/v2/testutil"
"github.com/coder/quartz"
)
type options struct {
@@ -86,7 +84,7 @@ func TestNoPrebuilds(t *testing.T) {
preset(true, 0, current),
}
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, nil, nil, nil, nil, clock, testutil.Logger(t))
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, nil, nil, nil, nil, nil, clock, testutil.Logger(t))
ps, err := snapshot.FilterByPreset(current.presetID)
require.NoError(t, err)
@@ -108,7 +106,7 @@ func TestNetNew(t *testing.T) {
preset(true, 1, current),
}
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, nil, nil, nil, nil, clock, testutil.Logger(t))
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, nil, nil, nil, nil, nil, clock, testutil.Logger(t))
ps, err := snapshot.FilterByPreset(current.presetID)
require.NoError(t, err)
@@ -150,7 +148,7 @@ func TestOutdatedPrebuilds(t *testing.T) {
var inProgress []database.CountInProgressPrebuildsRow
// WHEN: calculating the outdated preset's state.
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, running, inProgress, nil, nil, quartz.NewMock(t), testutil.Logger(t))
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, running, inProgress, nil, nil, nil, quartz.NewMock(t), testutil.Logger(t))
ps, err := snapshot.FilterByPreset(outdated.presetID)
require.NoError(t, err)
@@ -216,7 +214,7 @@ func TestDeleteOutdatedPrebuilds(t *testing.T) {
}
// WHEN: calculating the outdated preset's state.
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, running, inProgress, nil, nil, quartz.NewMock(t), testutil.Logger(t))
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, running, inProgress, nil, nil, nil, quartz.NewMock(t), testutil.Logger(t))
ps, err := snapshot.FilterByPreset(outdated.presetID)
require.NoError(t, err)
@@ -238,6 +236,74 @@ func TestDeleteOutdatedPrebuilds(t *testing.T) {
}, actions)
}
func TestCancelPendingPrebuilds(t *testing.T) {
t.Parallel()
// Setup
current := opts[optionSet3]
clock := quartz.NewMock(t)
t.Run("CancelPendingPrebuildsNonActiveVersion", func(t *testing.T) {
t.Parallel()
// Given: a preset from a non-active version
defaultPreset := preset(false, 0, current)
presets := []database.GetTemplatePresetsWithPrebuildsRow{
defaultPreset,
}
// Given: 2 pending prebuilt workspaces for the preset
pending := []database.CountPendingNonActivePrebuildsRow{{
PresetID: uuid.NullUUID{
UUID: defaultPreset.ID,
Valid: true,
},
Count: 2,
}}
// When: calculating the current preset's state
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, nil, nil, pending, nil, nil, clock, testutil.Logger(t))
ps, err := snapshot.FilterByPreset(current.presetID)
require.NoError(t, err)
// Then: it should create a cancel reconciliation action
actions, err := ps.CalculateActions(backoffInterval)
require.NoError(t, err)
expectedAction := []*prebuilds.ReconciliationActions{{ActionType: prebuilds.ActionTypeCancelPending}}
require.Equal(t, expectedAction, actions)
})
t.Run("NotCancelPendingPrebuildsActiveVersion", func(t *testing.T) {
t.Parallel()
// Given: a preset from an active version
defaultPreset := preset(true, 0, current)
presets := []database.GetTemplatePresetsWithPrebuildsRow{
defaultPreset,
}
// Given: 2 pending prebuilt workspaces for the preset
pending := []database.CountPendingNonActivePrebuildsRow{{
PresetID: uuid.NullUUID{
UUID: defaultPreset.ID,
Valid: true,
},
Count: 2,
}}
// When: calculating the current preset's state
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, nil, nil, pending, nil, nil, clock, testutil.Logger(t))
ps, err := snapshot.FilterByPreset(current.presetID)
require.NoError(t, err)
// Then: it should not create a cancel reconciliation action
actions, err := ps.CalculateActions(backoffInterval)
require.NoError(t, err)
var expectedAction []*prebuilds.ReconciliationActions
require.Equal(t, expectedAction, actions)
})
}
// A new template version is created with a preset with prebuilds configured; while a prebuild is provisioning up or down,
// the calculated actions should indicate the state correctly.
func TestInProgressActions(t *testing.T) {
@@ -460,7 +526,7 @@ func TestInProgressActions(t *testing.T) {
}
// WHEN: calculating the current preset's state.
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, running, inProgress, nil, nil, quartz.NewMock(t), testutil.Logger(t))
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, running, inProgress, nil, nil, nil, quartz.NewMock(t), testutil.Logger(t))
ps, err := snapshot.FilterByPreset(current.presetID)
require.NoError(t, err)
@@ -503,7 +569,7 @@ func TestExtraneous(t *testing.T) {
var inProgress []database.CountInProgressPrebuildsRow
// WHEN: calculating the current preset's state.
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, running, inProgress, nil, nil, quartz.NewMock(t), testutil.Logger(t))
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, running, inProgress, nil, nil, nil, quartz.NewMock(t), testutil.Logger(t))
ps, err := snapshot.FilterByPreset(current.presetID)
require.NoError(t, err)
@@ -683,7 +749,7 @@ func TestExpiredPrebuilds(t *testing.T) {
}
// WHEN: calculating the current preset's state.
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, running, nil, nil, nil, clock, testutil.Logger(t))
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, running, nil, nil, nil, nil, clock, testutil.Logger(t))
ps, err := snapshot.FilterByPreset(current.presetID)
require.NoError(t, err)
@@ -719,7 +785,7 @@ func TestDeprecated(t *testing.T) {
var inProgress []database.CountInProgressPrebuildsRow
// WHEN: calculating the current preset's state.
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, running, inProgress, nil, nil, quartz.NewMock(t), testutil.Logger(t))
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, running, inProgress, nil, nil, nil, quartz.NewMock(t), testutil.Logger(t))
ps, err := snapshot.FilterByPreset(current.presetID)
require.NoError(t, err)
@@ -772,7 +838,7 @@ func TestLatestBuildFailed(t *testing.T) {
}
// WHEN: calculating the current preset's state.
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, running, inProgress, backoffs, nil, clock, testutil.Logger(t))
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, running, inProgress, nil, backoffs, nil, clock, testutil.Logger(t))
psCurrent, err := snapshot.FilterByPreset(current.presetID)
require.NoError(t, err)
@@ -865,7 +931,7 @@ func TestMultiplePresetsPerTemplateVersion(t *testing.T) {
},
}
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, nil, inProgress, nil, nil, clock, testutil.Logger(t))
snapshot := prebuilds.NewGlobalSnapshot(presets, nil, nil, inProgress, nil, nil, nil, clock, testutil.Logger(t))
// Nothing has to be created for preset 1.
{
@@ -985,7 +1051,7 @@ func TestPrebuildScheduling(t *testing.T) {
schedule(presets[1].ID, "* 14-16 * * 1-5", 5),
}
snapshot := prebuilds.NewGlobalSnapshot(presets, schedules, nil, nil, nil, nil, clock, testutil.Logger(t))
snapshot := prebuilds.NewGlobalSnapshot(presets, schedules, nil, nil, nil, nil, nil, clock, testutil.Logger(t))
// Check 1st preset.
{
@@ -1093,6 +1159,7 @@ func TestCalculateDesiredInstances(t *testing.T) {
nil,
nil,
nil,
0,
nil,
false,
quartz.NewMock(t),
+43 -7
View File
@@ -12,10 +12,13 @@ import (
"sync/atomic"
"time"
"github.com/google/uuid"
"github.com/hashicorp/go-multierror"
"github.com/prometheus/client_golang/prometheus"
"golang.org/x/sync/errgroup"
"golang.org/x/xerrors"
"github.com/coder/quartz"
"cdr.dev/slog"
"github.com/coder/coder/v2/coderd/audit"
"github.com/coder/coder/v2/coderd/database"
@@ -30,12 +33,7 @@ import (
"github.com/coder/coder/v2/coderd/wsbuilder"
"github.com/coder/coder/v2/codersdk"
sdkproto "github.com/coder/coder/v2/provisionersdk/proto"
"cdr.dev/slog"
"github.com/google/uuid"
"golang.org/x/sync/errgroup"
"golang.org/x/xerrors"
"github.com/coder/quartz"
)
type StoreReconciler struct {
@@ -412,6 +410,11 @@ func (c *StoreReconciler) SnapshotState(ctx context.Context, store database.Stor
return xerrors.Errorf("failed to get prebuilds in progress: %w", err)
}
allPendingPrebuilds, err := db.CountPendingNonActivePrebuilds(ctx)
if err != nil {
return xerrors.Errorf("failed to get pending prebuilds: %w", err)
}
presetsBackoff, err := db.GetPresetsBackoff(ctx, c.clock.Now().Add(-c.cfg.ReconciliationBackoffLookback.Value()))
if err != nil {
return xerrors.Errorf("failed to get backoffs for presets: %w", err)
@@ -427,6 +430,7 @@ func (c *StoreReconciler) SnapshotState(ctx context.Context, store database.Stor
presetPrebuildSchedules,
allRunningPrebuilds,
allPrebuildsInProgress,
allPendingPrebuilds,
presetsBackoff,
hardLimitedPresets,
c.clock,
@@ -581,6 +585,8 @@ func (c *StoreReconciler) executeReconciliationAction(ctx context.Context, logge
levelFn = logger.Info
case action.ActionType == prebuilds.ActionTypeDelete && len(action.DeleteIDs) > 0:
levelFn = logger.Info
case action.ActionType == prebuilds.ActionTypeCancelPending:
levelFn = logger.Info
}
switch action.ActionType {
@@ -635,6 +641,36 @@ func (c *StoreReconciler) executeReconciliationAction(ctx context.Context, logge
return multiErr.ErrorOrNil()
case prebuilds.ActionTypeCancelPending:
// Cancel pending prebuild jobs from non-active template versions to avoid
// provisioning obsolete workspaces that would immediately be deprovisioned.
// This uses a criteria-based update to ensure only jobs that are still pending
// at execution time are canceled, avoiding race conditions where jobs may have
// transitioned to running status between query and update.
canceledJobs, err := c.store.UpdatePrebuildProvisionerJobWithCancel(
ctx,
database.UpdatePrebuildProvisionerJobWithCancelParams{
Now: c.clock.Now(),
PresetID: uuid.NullUUID{
UUID: ps.Preset.ID,
Valid: true,
},
})
if err != nil {
logger.Error(ctx, "failed to cancel pending prebuild jobs",
slog.F("template_version_id", ps.Preset.TemplateVersionID.String()),
slog.F("preset_id", ps.Preset.ID),
slog.Error(err))
return err
}
if len(canceledJobs) > 0 {
logger.Info(ctx, "canceled pending prebuild jobs for inactive version",
slog.F("template_version_id", ps.Preset.TemplateVersionID.String()),
slog.F("preset_id", ps.Preset.ID),
slog.F("count", len(canceledJobs)))
}
return nil
default:
return xerrors.Errorf("unknown action type: %v", action.ActionType)
}
+560 -15
View File
@@ -9,36 +9,35 @@ import (
"testing"
"time"
"github.com/google/uuid"
"github.com/prometheus/client_golang/prometheus"
"github.com/stretchr/testify/assert"
"golang.org/x/xerrors"
"github.com/coder/coder/v2/coderd/coderdtest"
"github.com/coder/coder/v2/coderd/database/dbtime"
"github.com/coder/coder/v2/coderd/files"
"github.com/coder/coder/v2/coderd/notifications"
"github.com/coder/coder/v2/coderd/notifications/notificationstest"
"github.com/coder/coder/v2/coderd/util/slice"
"github.com/coder/coder/v2/coderd/wsbuilder"
sdkproto "github.com/coder/coder/v2/provisionersdk/proto"
"github.com/google/uuid"
"github.com/stretchr/testify/require"
"golang.org/x/xerrors"
"tailscale.com/types/ptr"
"cdr.dev/slog"
"cdr.dev/slog/sloggers/slogtest"
"github.com/coder/quartz"
"github.com/coder/serpent"
"github.com/coder/coder/v2/coderd/coderdtest"
"github.com/coder/coder/v2/coderd/database"
"github.com/coder/coder/v2/coderd/database/dbfake"
"github.com/coder/coder/v2/coderd/database/dbgen"
"github.com/coder/coder/v2/coderd/database/dbtestutil"
"github.com/coder/coder/v2/coderd/database/dbtime"
"github.com/coder/coder/v2/coderd/database/pubsub"
"github.com/coder/coder/v2/coderd/files"
"github.com/coder/coder/v2/coderd/notifications"
"github.com/coder/coder/v2/coderd/notifications/notificationstest"
"github.com/coder/coder/v2/coderd/rbac"
"github.com/coder/coder/v2/coderd/util/slice"
"github.com/coder/coder/v2/coderd/wsbuilder"
"github.com/coder/coder/v2/codersdk"
"github.com/coder/coder/v2/enterprise/coderd/prebuilds"
sdkproto "github.com/coder/coder/v2/provisionersdk/proto"
"github.com/coder/coder/v2/testutil"
"github.com/coder/quartz"
"github.com/coder/serpent"
)
func TestNoReconciliationActionsIfNoPresets(t *testing.T) {
@@ -1783,6 +1782,552 @@ func TestExpiredPrebuildsMultipleActions(t *testing.T) {
}
}
func TestCancelPendingPrebuilds(t *testing.T) {
t.Parallel()
t.Run("CancelPendingPrebuilds", func(t *testing.T) {
t.Parallel()
for _, tt := range []struct {
name string
setupBuild func(
t *testing.T,
db database.Store,
client *codersdk.Client,
orgID uuid.UUID,
templateID uuid.UUID,
templateVersionID uuid.UUID,
presetID uuid.NullUUID,
) dbfake.WorkspaceResponse
activeTemplateVersion bool
previouslyCanceled bool
previouslyCompleted bool
shouldCancel bool
}{
// Should cancel pending prebuild-related jobs from a non-active template version
{
name: "CancelsPendingPrebuildJobNonActiveVersion",
// Given: a pending prebuild job
setupBuild: func(t *testing.T,
db database.Store,
client *codersdk.Client,
orgID uuid.UUID,
templateID uuid.UUID,
templateVersionID uuid.UUID,
presetID uuid.NullUUID,
) dbfake.WorkspaceResponse {
return dbfake.WorkspaceBuild(t, db, database.WorkspaceTable{
OwnerID: database.PrebuildsSystemUserID,
OrganizationID: orgID,
TemplateID: templateID,
}).Pending().Seed(database.WorkspaceBuild{
InitiatorID: database.PrebuildsSystemUserID,
TemplateVersionID: templateVersionID,
TemplateVersionPresetID: presetID,
}).Do()
},
activeTemplateVersion: false,
previouslyCanceled: false,
previouslyCompleted: false,
shouldCancel: true,
},
// Should not cancel pending prebuild-related jobs from an active template version
{
name: "DoesNotCancelPendingPrebuildJobActiveVersion",
// Given: a pending prebuild job
setupBuild: func(t *testing.T,
db database.Store,
client *codersdk.Client,
orgID uuid.UUID,
templateID uuid.UUID,
templateVersionID uuid.UUID,
presetID uuid.NullUUID,
) dbfake.WorkspaceResponse {
return dbfake.WorkspaceBuild(t, db, database.WorkspaceTable{
OwnerID: database.PrebuildsSystemUserID,
OrganizationID: orgID,
TemplateID: templateID,
}).Pending().Seed(database.WorkspaceBuild{
InitiatorID: database.PrebuildsSystemUserID,
TemplateVersionID: templateVersionID,
TemplateVersionPresetID: presetID,
}).Do()
},
activeTemplateVersion: true,
previouslyCanceled: false,
previouslyCompleted: false,
shouldCancel: false,
},
// Should not cancel pending prebuild-related jobs associated to a second workspace build
{
name: "DoesNotCancelPendingPrebuildJobSecondBuild",
// Given: a pending prebuild job associated to a second workspace build
setupBuild: func(t *testing.T,
db database.Store,
client *codersdk.Client,
orgID uuid.UUID,
templateID uuid.UUID,
templateVersionID uuid.UUID,
presetID uuid.NullUUID,
) dbfake.WorkspaceResponse {
return dbfake.WorkspaceBuild(t, db, database.WorkspaceTable{
OwnerID: database.PrebuildsSystemUserID,
OrganizationID: orgID,
TemplateID: templateID,
}).Pending().Seed(database.WorkspaceBuild{
InitiatorID: database.PrebuildsSystemUserID,
BuildNumber: int32(2),
TemplateVersionID: templateVersionID,
TemplateVersionPresetID: presetID,
}).Do()
},
activeTemplateVersion: false,
previouslyCanceled: false,
previouslyCompleted: false,
shouldCancel: false,
},
// Should not cancel pending prebuild-related jobs of a different template
{
name: "DoesNotCancelPrebuildJobDifferentTemplate",
// Given: a pending prebuild job belonging to a different template
setupBuild: func(
t *testing.T,
db database.Store,
client *codersdk.Client,
orgID uuid.UUID,
templateID uuid.UUID,
templateVersionID uuid.UUID,
presetID uuid.NullUUID,
) dbfake.WorkspaceResponse {
return dbfake.WorkspaceBuild(t, db, database.WorkspaceTable{
OwnerID: database.PrebuildsSystemUserID,
OrganizationID: orgID,
TemplateID: uuid.Nil,
}).Pending().Seed(database.WorkspaceBuild{
InitiatorID: database.PrebuildsSystemUserID,
TemplateVersionID: templateVersionID,
TemplateVersionPresetID: presetID,
}).Do()
},
activeTemplateVersion: false,
previouslyCanceled: false,
previouslyCompleted: false,
shouldCancel: false,
},
// Should not cancel pending user workspace build jobs
{
name: "DoesNotCancelUserWorkspaceJob",
// Given: a pending user workspace build job
setupBuild: func(
t *testing.T,
db database.Store,
client *codersdk.Client,
orgID uuid.UUID,
templateID uuid.UUID,
templateVersionID uuid.UUID,
presetID uuid.NullUUID,
) dbfake.WorkspaceResponse {
_, member := coderdtest.CreateAnotherUser(t, client, orgID, rbac.RoleMember())
return dbfake.WorkspaceBuild(t, db, database.WorkspaceTable{
OwnerID: member.ID,
OrganizationID: orgID,
TemplateID: uuid.Nil,
}).Pending().Seed(database.WorkspaceBuild{
InitiatorID: member.ID,
TemplateVersionID: templateVersionID,
TemplateVersionPresetID: presetID,
}).Do()
},
activeTemplateVersion: false,
previouslyCanceled: false,
previouslyCompleted: false,
shouldCancel: false,
},
// Should not cancel pending prebuild-related jobs with a delete transition
{
name: "DoesNotCancelPrebuildJobDeleteTransition",
// Given: a pending prebuild job with a delete transition
setupBuild: func(
t *testing.T,
db database.Store,
client *codersdk.Client,
orgID uuid.UUID,
templateID uuid.UUID,
templateVersionID uuid.UUID,
presetID uuid.NullUUID,
) dbfake.WorkspaceResponse {
return dbfake.WorkspaceBuild(t, db, database.WorkspaceTable{
OwnerID: database.PrebuildsSystemUserID,
OrganizationID: orgID,
TemplateID: templateID,
}).Pending().Seed(database.WorkspaceBuild{
InitiatorID: database.PrebuildsSystemUserID,
Transition: database.WorkspaceTransitionDelete,
TemplateVersionID: templateVersionID,
TemplateVersionPresetID: presetID,
}).Do()
},
activeTemplateVersion: false,
previouslyCanceled: false,
previouslyCompleted: false,
shouldCancel: false,
},
// Should not cancel prebuild-related jobs already being processed by a provisioner
{
name: "DoesNotCancelRunningPrebuildJob",
// Given: a running prebuild job
setupBuild: func(
t *testing.T,
db database.Store,
client *codersdk.Client,
orgID uuid.UUID,
templateID uuid.UUID,
templateVersionID uuid.UUID,
presetID uuid.NullUUID,
) dbfake.WorkspaceResponse {
return dbfake.WorkspaceBuild(t, db, database.WorkspaceTable{
OwnerID: database.PrebuildsSystemUserID,
OrganizationID: orgID,
TemplateID: templateID,
}).Starting().Seed(database.WorkspaceBuild{
InitiatorID: database.PrebuildsSystemUserID,
TemplateVersionID: templateVersionID,
TemplateVersionPresetID: presetID,
}).Do()
},
activeTemplateVersion: false,
previouslyCanceled: false,
previouslyCompleted: false,
shouldCancel: false,
},
// Should not cancel already canceled prebuild-related jobs
{
name: "DoesNotCancelCanceledPrebuildJob",
// Given: a canceled prebuild job
setupBuild: func(
t *testing.T,
db database.Store,
client *codersdk.Client,
orgID uuid.UUID,
templateID uuid.UUID,
templateVersionID uuid.UUID,
presetID uuid.NullUUID,
) dbfake.WorkspaceResponse {
return dbfake.WorkspaceBuild(t, db, database.WorkspaceTable{
OwnerID: database.PrebuildsSystemUserID,
OrganizationID: orgID,
TemplateID: templateID,
}).Canceled().Seed(database.WorkspaceBuild{
InitiatorID: database.PrebuildsSystemUserID,
TemplateVersionID: templateVersionID,
TemplateVersionPresetID: presetID,
}).Do()
},
activeTemplateVersion: false,
shouldCancel: false,
previouslyCanceled: true,
previouslyCompleted: true,
},
// Should not cancel completed prebuild-related jobs
{
name: "DoesNotCancelCompletedPrebuildJob",
// Given: a completed prebuild job
setupBuild: func(
t *testing.T,
db database.Store,
client *codersdk.Client,
orgID uuid.UUID,
templateID uuid.UUID,
templateVersionID uuid.UUID,
presetID uuid.NullUUID,
) dbfake.WorkspaceResponse {
return dbfake.WorkspaceBuild(t, db, database.WorkspaceTable{
OwnerID: database.PrebuildsSystemUserID,
OrganizationID: orgID,
TemplateID: templateID,
}).Seed(database.WorkspaceBuild{
InitiatorID: database.PrebuildsSystemUserID,
TemplateVersionID: templateVersionID,
TemplateVersionPresetID: presetID,
}).Do()
},
activeTemplateVersion: false,
shouldCancel: false,
previouslyCanceled: false,
previouslyCompleted: true,
},
} {
t.Run(tt.name, func(t *testing.T) {
t.Parallel()
// Set the clock to Monday, January 1st, 2024 at 8:00 AM UTC to keep the test deterministic
clock := quartz.NewMock(t)
clock.Set(time.Date(2024, 1, 1, 8, 0, 0, 0, time.UTC))
ctx, cancel := context.WithTimeout(context.Background(), testutil.WaitLong)
defer cancel()
// Setup
db, ps := dbtestutil.NewDB(t)
client, _, _ := coderdtest.NewWithAPI(t, &coderdtest.Options{
// Explicitly not including provisioner daemons, as we don't want the jobs to be processed
// Jobs operations will be simulated via the database model
IncludeProvisionerDaemon: false,
Database: db,
Pubsub: ps,
Clock: clock,
})
fakeEnqueuer := newFakeEnqueuer()
registry := prometheus.NewRegistry()
cache := files.New(registry, &coderdtest.FakeAuthorizer{})
logger := slogtest.Make(t, &slogtest.Options{IgnoreErrors: false}).Leveled(slog.LevelDebug)
reconciler := prebuilds.NewStoreReconciler(db, ps, cache, codersdk.PrebuildsConfig{}, logger, clock, registry, fakeEnqueuer, newNoopUsageCheckerPtr())
owner := coderdtest.CreateFirstUser(t, client)
// Given: a template with a version containing a preset with 1 prebuild instance
nonActivePresetID := uuid.NullUUID{
UUID: uuid.New(),
Valid: true,
}
nonActiveTemplateVersion := dbfake.TemplateVersion(t, db).Seed(database.TemplateVersion{
OrganizationID: owner.OrganizationID,
CreatedBy: owner.UserID,
}).Preset(database.TemplateVersionPreset{
ID: nonActivePresetID.UUID,
DesiredInstances: sql.NullInt32{
Int32: 1,
Valid: true,
},
}).Do()
templateID := nonActiveTemplateVersion.Template.ID
// Given: a new active template version
activePresetID := uuid.NullUUID{
UUID: uuid.New(),
Valid: true,
}
activeTemplateVersion := dbfake.TemplateVersion(t, db).Seed(database.TemplateVersion{
OrganizationID: owner.OrganizationID,
CreatedBy: owner.UserID,
TemplateID: uuid.NullUUID{
UUID: templateID,
Valid: true,
},
}).Preset(database.TemplateVersionPreset{
ID: activePresetID.UUID,
DesiredInstances: sql.NullInt32{
Int32: 1,
Valid: true,
},
}).SkipCreateTemplate().Do()
var workspace dbfake.WorkspaceResponse
if tt.activeTemplateVersion {
// Given: a prebuilt workspace, workspace build and respective provisioner job from an
// active template version
workspace = tt.setupBuild(t, db, client,
owner.OrganizationID, templateID, activeTemplateVersion.TemplateVersion.ID, activePresetID)
} else {
// Given: a prebuilt workspace, workspace build and respective provisioner job from a
// non-active template version
workspace = tt.setupBuild(t, db, client,
owner.OrganizationID, templateID, nonActiveTemplateVersion.TemplateVersion.ID, nonActivePresetID)
}
// Given: the new template version is promoted to active
err := db.UpdateTemplateActiveVersionByID(ctx, database.UpdateTemplateActiveVersionByIDParams{
ID: templateID,
ActiveVersionID: activeTemplateVersion.TemplateVersion.ID,
})
require.NoError(t, err)
// When: the reconciliation loop is triggered
require.NoError(t, reconciler.ReconcileAll(ctx))
if tt.shouldCancel {
// Then: the prebuild related jobs from non-active version should be canceled
cancelledJob, err := db.GetProvisionerJobByID(ctx, workspace.Build.JobID)
require.NoError(t, err)
require.Equal(t, clock.Now().UTC(), cancelledJob.CanceledAt.Time.UTC())
require.Equal(t, clock.Now().UTC(), cancelledJob.CompletedAt.Time.UTC())
require.Equal(t, database.ProvisionerJobStatusCanceled, cancelledJob.JobStatus)
} else {
// Then: the provisioner job should not be canceled
job, err := db.GetProvisionerJobByID(ctx, workspace.Build.JobID)
require.NoError(t, err)
if !tt.previouslyCanceled {
require.Zero(t, job.CanceledAt.Time.UTC())
require.NotEqual(t, database.ProvisionerJobStatusCanceled, job.JobStatus)
}
if !tt.previouslyCompleted {
require.Zero(t, job.CompletedAt.Time.UTC())
}
}
})
}
})
t.Run("CancelPendingPrebuildsMultipleTemplates", func(t *testing.T) {
t.Parallel()
createTemplateVersionWithPreset := func(
t *testing.T,
db database.Store,
orgID uuid.UUID,
userID uuid.UUID,
templateID uuid.UUID,
prebuiltInstances int32,
) (uuid.UUID, uuid.UUID, uuid.UUID) {
templatePreset := uuid.NullUUID{
UUID: uuid.New(),
Valid: true,
}
templateVersion := dbfake.TemplateVersion(t, db).Seed(database.TemplateVersion{
OrganizationID: orgID,
CreatedBy: userID,
TemplateID: uuid.NullUUID{
UUID: templateID,
Valid: true,
},
}).Preset(database.TemplateVersionPreset{
ID: templatePreset.UUID,
DesiredInstances: sql.NullInt32{
Int32: prebuiltInstances,
Valid: true,
},
}).Do()
return templateVersion.Template.ID, templateVersion.TemplateVersion.ID, templatePreset.UUID
}
setupPrebuilds := func(
t *testing.T,
db database.Store,
orgID uuid.UUID,
templateID uuid.UUID,
versionID uuid.UUID,
presetID uuid.UUID,
count int,
pending bool,
) []dbfake.WorkspaceResponse {
prebuilds := make([]dbfake.WorkspaceResponse, count)
for i := range count {
builder := dbfake.WorkspaceBuild(t, db, database.WorkspaceTable{
OwnerID: database.PrebuildsSystemUserID,
OrganizationID: orgID,
TemplateID: templateID,
})
if pending {
builder = builder.Pending()
}
prebuilds[i] = builder.Seed(database.WorkspaceBuild{
InitiatorID: database.PrebuildsSystemUserID,
TemplateVersionID: versionID,
TemplateVersionPresetID: uuid.NullUUID{
UUID: presetID,
Valid: true,
},
}).Do()
}
return prebuilds
}
checkIfJobCanceled := func(
t *testing.T,
clock *quartz.Mock,
ctx context.Context,
db database.Store,
shouldBeCanceled bool,
prebuilds []dbfake.WorkspaceResponse,
) {
for _, prebuild := range prebuilds {
job, err := db.GetProvisionerJobByID(ctx, prebuild.Build.JobID)
require.NoError(t, err)
if shouldBeCanceled {
require.Equal(t, database.ProvisionerJobStatusCanceled, job.JobStatus)
require.Equal(t, clock.Now().UTC(), job.CanceledAt.Time.UTC())
require.Equal(t, clock.Now().UTC(), job.CompletedAt.Time.UTC())
} else {
require.NotEqual(t, database.ProvisionerJobStatusCanceled, job.JobStatus)
require.Zero(t, job.CanceledAt.Time.UTC())
}
}
}
// Set the clock to Monday, January 1st, 2024 at 8:00 AM UTC to keep the test deterministic
clock := quartz.NewMock(t)
clock.Set(time.Date(2024, 1, 1, 8, 0, 0, 0, time.UTC))
ctx, cancel := context.WithTimeout(context.Background(), testutil.WaitLong)
defer cancel()
// Setup
db, ps := dbtestutil.NewDB(t)
client, _, _ := coderdtest.NewWithAPI(t, &coderdtest.Options{
// Explicitly not including provisioner daemons, as we don't want the jobs to be processed
// Jobs operations will be simulated via the database model
IncludeProvisionerDaemon: false,
Database: db,
Pubsub: ps,
Clock: clock,
})
fakeEnqueuer := newFakeEnqueuer()
registry := prometheus.NewRegistry()
cache := files.New(registry, &coderdtest.FakeAuthorizer{})
logger := slogtest.Make(t, &slogtest.Options{IgnoreErrors: false}).Leveled(slog.LevelDebug)
reconciler := prebuilds.NewStoreReconciler(db, ps, cache, codersdk.PrebuildsConfig{}, logger, clock, registry, fakeEnqueuer, newNoopUsageCheckerPtr())
owner := coderdtest.CreateFirstUser(t, client)
// Given: template A with 2 versions
// Given: template A version v1: with a preset with 5 instances (2 running, 3 pending)
templateAID, templateAVersion1ID, templateAVersion1PresetID := createTemplateVersionWithPreset(t, db, owner.OrganizationID, owner.UserID, uuid.Nil, 5)
templateAVersion1Running := setupPrebuilds(t, db, owner.OrganizationID, templateAID, templateAVersion1ID, templateAVersion1PresetID, 2, false)
templateAVersion1Pending := setupPrebuilds(t, db, owner.OrganizationID, templateAID, templateAVersion1ID, templateAVersion1PresetID, 3, true)
// Given: template A version v2 (active version): with a preset with 2 instances (1 running, 1 pending)
_, templateAVersion2ID, templateAVersion2PresetID := createTemplateVersionWithPreset(t, db, owner.OrganizationID, owner.UserID, templateAID, 2)
templateAVersion2Running := setupPrebuilds(t, db, owner.OrganizationID, templateAID, templateAVersion2ID, templateAVersion2PresetID, 1, false)
templateAVersion2Pending := setupPrebuilds(t, db, owner.OrganizationID, templateAID, templateAVersion2ID, templateAVersion2PresetID, 1, true)
// Given: template B with 3 versions
// Given: template B version v1: with a preset with 3 instances (1 running, 2 pending)
templateBID, templateBVersion1ID, templateBVersion1PresetID := createTemplateVersionWithPreset(t, db, owner.OrganizationID, owner.UserID, uuid.Nil, 3)
templateBVersion1Running := setupPrebuilds(t, db, owner.OrganizationID, templateBID, templateBVersion1ID, templateBVersion1PresetID, 1, false)
templateBVersion1Pending := setupPrebuilds(t, db, owner.OrganizationID, templateBID, templateBVersion1ID, templateBVersion1PresetID, 2, true)
// Given: template B version v2: with a preset with 2 instances (2 pending)
_, templateBVersion2ID, templateBVersion2PresetID := createTemplateVersionWithPreset(t, db, owner.OrganizationID, owner.UserID, templateBID, 2)
templateBVersion2Pending := setupPrebuilds(t, db, owner.OrganizationID, templateBID, templateBVersion2ID, templateBVersion2PresetID, 2, true)
// Given: template B version v3 (active version): with a preset with 2 instances (1 running, 1 pending)
_, templateBVersion3ID, templateBVersion3PresetID := createTemplateVersionWithPreset(t, db, owner.OrganizationID, owner.UserID, templateBID, 2)
templateBVersion3Running := setupPrebuilds(t, db, owner.OrganizationID, templateBID, templateBVersion3ID, templateBVersion3PresetID, 1, false)
templateBVersion3Pending := setupPrebuilds(t, db, owner.OrganizationID, templateBID, templateBVersion3ID, templateBVersion3PresetID, 1, true)
// When: the reconciliation loop is executed
require.NoError(t, reconciler.ReconcileAll(ctx))
// Then: template A version 1 running workspaces should not be canceled
checkIfJobCanceled(t, clock, ctx, db, false, templateAVersion1Running)
// Then: template A version 1 pending workspaces should be canceled
checkIfJobCanceled(t, clock, ctx, db, true, templateAVersion1Pending)
// Then: template A version 2 running and pending workspaces should not be canceled
checkIfJobCanceled(t, clock, ctx, db, false, templateAVersion2Running)
checkIfJobCanceled(t, clock, ctx, db, false, templateAVersion2Pending)
// Then: template B version 1 running workspaces should not be canceled
checkIfJobCanceled(t, clock, ctx, db, false, templateBVersion1Running)
// Then: template B version 1 pending workspaces should be canceled
checkIfJobCanceled(t, clock, ctx, db, true, templateBVersion1Pending)
// Then: template B version 2 pending workspaces should be canceled
checkIfJobCanceled(t, clock, ctx, db, true, templateBVersion2Pending)
// Then: template B version 3 running and pending workspaces should not be canceled
checkIfJobCanceled(t, clock, ctx, db, false, templateBVersion3Running)
checkIfJobCanceled(t, clock, ctx, db, false, templateBVersion3Pending)
})
}
func newNoopEnqueuer() *notifications.NoopEnqueuer {
return notifications.NewNoopEnqueuer()
}