fix: make GetWorkspacesEligibleForTransition return less false-positives (#15429)

Relates to https://github.com/coder/coder/issues/15082

The old implementation of `GetWorkspacesEligibleForTransition` returns
many workspaces that are not actually eligible for transition. This new
implementation reduces this number significantly (at least on our
dogfood instance).
This commit is contained in:
Danielle Maywood
2024-11-13 10:24:20 +00:00
committed by GitHub
parent fa69d1ca74
commit f2fe379bd2
7 changed files with 223 additions and 129 deletions
+1 -1
View File
@@ -2803,7 +2803,7 @@ func (q *querier) GetWorkspacesAndAgentsByOwnerID(ctx context.Context, ownerID u
return q.db.GetAuthorizedWorkspacesAndAgentsByOwnerID(ctx, ownerID, prep)
}
func (q *querier) GetWorkspacesEligibleForTransition(ctx context.Context, now time.Time) ([]database.WorkspaceTable, error) {
func (q *querier) GetWorkspacesEligibleForTransition(ctx context.Context, now time.Time) ([]database.GetWorkspacesEligibleForTransitionRow, error) {
return q.db.GetWorkspacesEligibleForTransition(ctx, now)
}
+81 -42
View File
@@ -6868,60 +6868,99 @@ func (q *FakeQuerier) GetWorkspacesAndAgentsByOwnerID(ctx context.Context, owner
return q.GetAuthorizedWorkspacesAndAgentsByOwnerID(ctx, ownerID, nil)
}
func (q *FakeQuerier) GetWorkspacesEligibleForTransition(ctx context.Context, now time.Time) ([]database.WorkspaceTable, error) {
func (q *FakeQuerier) GetWorkspacesEligibleForTransition(ctx context.Context, now time.Time) ([]database.GetWorkspacesEligibleForTransitionRow, error) {
q.mutex.RLock()
defer q.mutex.RUnlock()
workspaces := []database.WorkspaceTable{}
workspaces := []database.GetWorkspacesEligibleForTransitionRow{}
for _, workspace := range q.workspaces {
build, err := q.getLatestWorkspaceBuildByWorkspaceIDNoLock(ctx, workspace.ID)
if err != nil {
return nil, err
}
if build.Transition == database.WorkspaceTransitionStart &&
!build.Deadline.IsZero() &&
build.Deadline.Before(now) &&
!workspace.DormantAt.Valid {
workspaces = append(workspaces, workspace)
continue
}
if build.Transition == database.WorkspaceTransitionStop &&
workspace.AutostartSchedule.Valid &&
!workspace.DormantAt.Valid {
workspaces = append(workspaces, workspace)
continue
}
job, err := q.getProvisionerJobByIDNoLock(ctx, build.JobID)
if err != nil {
return nil, xerrors.Errorf("get provisioner job by ID: %w", err)
}
if codersdk.ProvisionerJobStatus(job.JobStatus) == codersdk.ProvisionerJobFailed {
workspaces = append(workspaces, workspace)
continue
}
template, err := q.getTemplateByIDNoLock(ctx, workspace.TemplateID)
if err != nil {
return nil, xerrors.Errorf("get template by ID: %w", err)
}
if !workspace.DormantAt.Valid && template.TimeTilDormant > 0 {
workspaces = append(workspaces, workspace)
continue
}
if workspace.DormantAt.Valid && template.TimeTilDormantAutoDelete > 0 {
workspaces = append(workspaces, workspace)
continue
return nil, xerrors.Errorf("get workspace build by ID: %w", err)
}
user, err := q.getUserByIDNoLock(workspace.OwnerID)
if err != nil {
return nil, xerrors.Errorf("get user by ID: %w", err)
}
if user.Status == database.UserStatusSuspended && build.Transition == database.WorkspaceTransitionStart {
workspaces = append(workspaces, workspace)
job, err := q.getProvisionerJobByIDNoLock(ctx, build.JobID)
if err != nil {
return nil, xerrors.Errorf("get provisioner job by ID: %w", err)
}
template, err := q.getTemplateByIDNoLock(ctx, workspace.TemplateID)
if err != nil {
return nil, xerrors.Errorf("get template by ID: %w", err)
}
if workspace.Deleted {
continue
}
if job.JobStatus != database.ProvisionerJobStatusFailed &&
!workspace.DormantAt.Valid &&
build.Transition == database.WorkspaceTransitionStart &&
(user.Status == database.UserStatusSuspended || (!build.Deadline.IsZero() && build.Deadline.Before(now))) {
workspaces = append(workspaces, database.GetWorkspacesEligibleForTransitionRow{
ID: workspace.ID,
Name: workspace.Name,
})
continue
}
if user.Status == database.UserStatusActive &&
job.JobStatus != database.ProvisionerJobStatusFailed &&
build.Transition == database.WorkspaceTransitionStop &&
workspace.AutostartSchedule.Valid {
workspaces = append(workspaces, database.GetWorkspacesEligibleForTransitionRow{
ID: workspace.ID,
Name: workspace.Name,
})
continue
}
if !workspace.DormantAt.Valid &&
template.TimeTilDormant > 0 &&
now.Sub(workspace.LastUsedAt) > time.Duration(template.TimeTilDormant) {
workspaces = append(workspaces, database.GetWorkspacesEligibleForTransitionRow{
ID: workspace.ID,
Name: workspace.Name,
})
continue
}
if workspace.DormantAt.Valid &&
workspace.DeletingAt.Valid &&
workspace.DeletingAt.Time.Before(now) &&
template.TimeTilDormantAutoDelete > 0 {
if build.Transition == database.WorkspaceTransitionDelete &&
job.JobStatus == database.ProvisionerJobStatusFailed {
if job.CanceledAt.Valid && now.Sub(job.CanceledAt.Time) <= 24*time.Hour {
continue
}
if job.CompletedAt.Valid && now.Sub(job.CompletedAt.Time) <= 24*time.Hour {
continue
}
}
workspaces = append(workspaces, database.GetWorkspacesEligibleForTransitionRow{
ID: workspace.ID,
Name: workspace.Name,
})
continue
}
if template.FailureTTL > 0 &&
build.Transition == database.WorkspaceTransitionStart &&
job.JobStatus == database.ProvisionerJobStatusFailed &&
job.CompletedAt.Valid &&
now.Sub(job.CompletedAt.Time) > time.Duration(template.FailureTTL) {
workspaces = append(workspaces, database.GetWorkspacesEligibleForTransitionRow{
ID: workspace.ID,
Name: workspace.Name,
})
continue
}
}
+1 -1
View File
@@ -1659,7 +1659,7 @@ func (m queryMetricsStore) GetWorkspacesAndAgentsByOwnerID(ctx context.Context,
return r0, r1
}
func (m queryMetricsStore) GetWorkspacesEligibleForTransition(ctx context.Context, now time.Time) ([]database.WorkspaceTable, error) {
func (m queryMetricsStore) GetWorkspacesEligibleForTransition(ctx context.Context, now time.Time) ([]database.GetWorkspacesEligibleForTransitionRow, error) {
start := time.Now()
workspaces, err := m.s.GetWorkspacesEligibleForTransition(ctx, now)
m.queryLatencies.WithLabelValues("GetWorkspacesEligibleForAutoStartStop").Observe(time.Since(start).Seconds())
+2 -2
View File
@@ -3503,10 +3503,10 @@ func (mr *MockStoreMockRecorder) GetWorkspacesAndAgentsByOwnerID(arg0, arg1 any)
}
// GetWorkspacesEligibleForTransition mocks base method.
func (m *MockStore) GetWorkspacesEligibleForTransition(arg0 context.Context, arg1 time.Time) ([]database.WorkspaceTable, error) {
func (m *MockStore) GetWorkspacesEligibleForTransition(arg0 context.Context, arg1 time.Time) ([]database.GetWorkspacesEligibleForTransitionRow, error) {
m.ctrl.T.Helper()
ret := m.ctrl.Call(m, "GetWorkspacesEligibleForTransition", arg0, arg1)
ret0, _ := ret[0].([]database.WorkspaceTable)
ret0, _ := ret[0].([]database.GetWorkspacesEligibleForTransitionRow)
ret1, _ := ret[1].(error)
return ret0, ret1
}
+1 -1
View File
@@ -346,7 +346,7 @@ type sqlcQuerier interface {
// be used in a WHERE clause.
GetWorkspaces(ctx context.Context, arg GetWorkspacesParams) ([]GetWorkspacesRow, error)
GetWorkspacesAndAgentsByOwnerID(ctx context.Context, ownerID uuid.UUID) ([]GetWorkspacesAndAgentsByOwnerIDRow, error)
GetWorkspacesEligibleForTransition(ctx context.Context, now time.Time) ([]WorkspaceTable, error)
GetWorkspacesEligibleForTransition(ctx context.Context, now time.Time) ([]GetWorkspacesEligibleForTransitionRow, error)
InsertAPIKey(ctx context.Context, arg InsertAPIKeyParams) (APIKey, error)
// We use the organization_id as the id
// for simplicity since all users is
+73 -50
View File
@@ -15363,7 +15363,8 @@ func (q *sqlQuerier) GetWorkspacesAndAgentsByOwnerID(ctx context.Context, ownerI
const getWorkspacesEligibleForTransition = `-- name: GetWorkspacesEligibleForTransition :many
SELECT
workspaces.id, workspaces.created_at, workspaces.updated_at, workspaces.owner_id, workspaces.organization_id, workspaces.template_id, workspaces.deleted, workspaces.name, workspaces.autostart_schedule, workspaces.ttl, workspaces.last_used_at, workspaces.dormant_at, workspaces.deleting_at, workspaces.automatic_updates, workspaces.favorite
workspaces.id,
workspaces.name
FROM
workspaces
LEFT JOIN
@@ -15385,82 +15386,104 @@ WHERE
) AND
(
-- If the workspace build was a start transition, the workspace is
-- potentially eligible for autostop if it's past the deadline. The
-- deadline is computed at build time upon success and is bumped based
-- on activity (up the max deadline if set). We don't need to check
-- license here since that's done when the values are written to the build.
-- A workspace may be eligible for autostop if the following are true:
-- * The provisioner job has not failed.
-- * The workspace is not dormant.
-- * The workspace build was a start transition.
-- * The workspace's owner is suspended OR the workspace build deadline has passed.
(
workspace_builds.transition = 'start'::workspace_transition AND
workspace_builds.deadline IS NOT NULL AND
workspace_builds.deadline < $1 :: timestamptz
provisioner_jobs.job_status != 'failed'::provisioner_job_status AND
workspaces.dormant_at IS NULL AND
workspace_builds.transition = 'start'::workspace_transition AND (
users.status = 'suspended'::user_status OR (
workspace_builds.deadline != '0001-01-01 00:00:00+00'::timestamptz AND
workspace_builds.deadline < $1 :: timestamptz
)
)
) OR
-- If the workspace build was a stop transition, the workspace is
-- potentially eligible for autostart if it has a schedule set. The
-- caller must check if the template allows autostart in a license-aware
-- fashion as we cannot check it here.
-- A workspace may be eligible for autostart if the following are true:
-- * The workspace's owner is active.
-- * The provisioner job did not fail.
-- * The workspace build was a stop transition.
-- * The workspace has an autostart schedule.
(
users.status = 'active'::user_status AND
provisioner_jobs.job_status != 'failed'::provisioner_job_status AND
workspace_builds.transition = 'stop'::workspace_transition AND
workspaces.autostart_schedule IS NOT NULL
) OR
-- If the workspace's most recent job resulted in an error
-- it may be eligible for failed stop.
(
provisioner_jobs.error IS NOT NULL AND
provisioner_jobs.error != '' AND
workspace_builds.transition = 'start'::workspace_transition
) OR
-- If the workspace's template has an inactivity_ttl set
-- it may be eligible for dormancy.
-- A workspace may be eligible for dormant stop if the following are true:
-- * The workspace is not dormant.
-- * The template has set a time 'til dormant.
-- * The workspace has been unused for longer than the time 'til dormancy.
(
workspaces.dormant_at IS NULL AND
templates.time_til_dormant > 0 AND
workspaces.dormant_at IS NULL
($1 :: timestamptz) - workspaces.last_used_at > (INTERVAL '1 millisecond' * (templates.time_til_dormant / 1000000))
) OR
-- If the workspace's template has a time_til_dormant_autodelete set
-- and the workspace is already dormant.
-- A workspace may be eligible for deletion if the following are true:
-- * The workspace is dormant.
-- * The workspace is scheduled to be deleted.
-- * If there was a prior attempt to delete the workspace that failed:
-- * This attempt was at least 24 hours ago.
(
workspaces.dormant_at IS NOT NULL AND
workspaces.deleting_at IS NOT NULL AND
workspaces.deleting_at < $1 :: timestamptz AND
templates.time_til_dormant_autodelete > 0 AND
workspaces.dormant_at IS NOT NULL
CASE
WHEN (
workspace_builds.transition = 'delete'::workspace_transition AND
provisioner_jobs.job_status = 'failed'::provisioner_job_status
) THEN (
(
provisioner_jobs.canceled_at IS NOT NULL OR
provisioner_jobs.completed_at IS NOT NULL
) AND (
($1 :: timestamptz) - (CASE
WHEN provisioner_jobs.canceled_at IS NOT NULL THEN provisioner_jobs.canceled_at
ELSE provisioner_jobs.completed_at
END) > INTERVAL '24 hours'
)
)
ELSE true
END
) OR
-- If the user account is suspended, and the workspace is running.
-- A workspace may be eligible for failed stop if the following are true:
-- * The template has a failure ttl set.
-- * The workspace build was a start transition.
-- * The provisioner job failed.
-- * The provisioner job had completed.
-- * The provisioner job has been completed for longer than the failure ttl.
(
users.status = 'suspended'::user_status AND
workspace_builds.transition = 'start'::workspace_transition
templates.failure_ttl > 0 AND
workspace_builds.transition = 'start'::workspace_transition AND
provisioner_jobs.job_status = 'failed'::provisioner_job_status AND
provisioner_jobs.completed_at IS NOT NULL AND
($1 :: timestamptz) - provisioner_jobs.completed_at > (INTERVAL '1 millisecond' * (templates.failure_ttl / 1000000))
)
) AND workspaces.deleted = 'false'
`
func (q *sqlQuerier) GetWorkspacesEligibleForTransition(ctx context.Context, now time.Time) ([]WorkspaceTable, error) {
type GetWorkspacesEligibleForTransitionRow struct {
ID uuid.UUID `db:"id" json:"id"`
Name string `db:"name" json:"name"`
}
func (q *sqlQuerier) GetWorkspacesEligibleForTransition(ctx context.Context, now time.Time) ([]GetWorkspacesEligibleForTransitionRow, error) {
rows, err := q.db.QueryContext(ctx, getWorkspacesEligibleForTransition, now)
if err != nil {
return nil, err
}
defer rows.Close()
var items []WorkspaceTable
var items []GetWorkspacesEligibleForTransitionRow
for rows.Next() {
var i WorkspaceTable
if err := rows.Scan(
&i.ID,
&i.CreatedAt,
&i.UpdatedAt,
&i.OwnerID,
&i.OrganizationID,
&i.TemplateID,
&i.Deleted,
&i.Name,
&i.AutostartSchedule,
&i.Ttl,
&i.LastUsedAt,
&i.DormantAt,
&i.DeletingAt,
&i.AutomaticUpdates,
&i.Favorite,
); err != nil {
var i GetWorkspacesEligibleForTransitionRow
if err := rows.Scan(&i.ID, &i.Name); err != nil {
return nil, err
}
items = append(items, i)
+64 -32
View File
@@ -557,7 +557,8 @@ FROM pending_workspaces, building_workspaces, running_workspaces, failed_workspa
-- name: GetWorkspacesEligibleForTransition :many
SELECT
workspaces.*
workspaces.id,
workspaces.name
FROM
workspaces
LEFT JOIN
@@ -579,52 +580,85 @@ WHERE
) AND
(
-- If the workspace build was a start transition, the workspace is
-- potentially eligible for autostop if it's past the deadline. The
-- deadline is computed at build time upon success and is bumped based
-- on activity (up the max deadline if set). We don't need to check
-- license here since that's done when the values are written to the build.
-- A workspace may be eligible for autostop if the following are true:
-- * The provisioner job has not failed.
-- * The workspace is not dormant.
-- * The workspace build was a start transition.
-- * The workspace's owner is suspended OR the workspace build deadline has passed.
(
workspace_builds.transition = 'start'::workspace_transition AND
workspace_builds.deadline IS NOT NULL AND
workspace_builds.deadline < @now :: timestamptz
provisioner_jobs.job_status != 'failed'::provisioner_job_status AND
workspaces.dormant_at IS NULL AND
workspace_builds.transition = 'start'::workspace_transition AND (
users.status = 'suspended'::user_status OR (
workspace_builds.deadline != '0001-01-01 00:00:00+00'::timestamptz AND
workspace_builds.deadline < @now :: timestamptz
)
)
) OR
-- If the workspace build was a stop transition, the workspace is
-- potentially eligible for autostart if it has a schedule set. The
-- caller must check if the template allows autostart in a license-aware
-- fashion as we cannot check it here.
-- A workspace may be eligible for autostart if the following are true:
-- * The workspace's owner is active.
-- * The provisioner job did not fail.
-- * The workspace build was a stop transition.
-- * The workspace has an autostart schedule.
(
users.status = 'active'::user_status AND
provisioner_jobs.job_status != 'failed'::provisioner_job_status AND
workspace_builds.transition = 'stop'::workspace_transition AND
workspaces.autostart_schedule IS NOT NULL
) OR
-- If the workspace's most recent job resulted in an error
-- it may be eligible for failed stop.
(
provisioner_jobs.error IS NOT NULL AND
provisioner_jobs.error != '' AND
workspace_builds.transition = 'start'::workspace_transition
) OR
-- If the workspace's template has an inactivity_ttl set
-- it may be eligible for dormancy.
-- A workspace may be eligible for dormant stop if the following are true:
-- * The workspace is not dormant.
-- * The template has set a time 'til dormant.
-- * The workspace has been unused for longer than the time 'til dormancy.
(
workspaces.dormant_at IS NULL AND
templates.time_til_dormant > 0 AND
workspaces.dormant_at IS NULL
(@now :: timestamptz) - workspaces.last_used_at > (INTERVAL '1 millisecond' * (templates.time_til_dormant / 1000000))
) OR
-- If the workspace's template has a time_til_dormant_autodelete set
-- and the workspace is already dormant.
-- A workspace may be eligible for deletion if the following are true:
-- * The workspace is dormant.
-- * The workspace is scheduled to be deleted.
-- * If there was a prior attempt to delete the workspace that failed:
-- * This attempt was at least 24 hours ago.
(
workspaces.dormant_at IS NOT NULL AND
workspaces.deleting_at IS NOT NULL AND
workspaces.deleting_at < @now :: timestamptz AND
templates.time_til_dormant_autodelete > 0 AND
workspaces.dormant_at IS NOT NULL
CASE
WHEN (
workspace_builds.transition = 'delete'::workspace_transition AND
provisioner_jobs.job_status = 'failed'::provisioner_job_status
) THEN (
(
provisioner_jobs.canceled_at IS NOT NULL OR
provisioner_jobs.completed_at IS NOT NULL
) AND (
(@now :: timestamptz) - (CASE
WHEN provisioner_jobs.canceled_at IS NOT NULL THEN provisioner_jobs.canceled_at
ELSE provisioner_jobs.completed_at
END) > INTERVAL '24 hours'
)
)
ELSE true
END
) OR
-- If the user account is suspended, and the workspace is running.
-- A workspace may be eligible for failed stop if the following are true:
-- * The template has a failure ttl set.
-- * The workspace build was a start transition.
-- * The provisioner job failed.
-- * The provisioner job had completed.
-- * The provisioner job has been completed for longer than the failure ttl.
(
users.status = 'suspended'::user_status AND
workspace_builds.transition = 'start'::workspace_transition
templates.failure_ttl > 0 AND
workspace_builds.transition = 'start'::workspace_transition AND
provisioner_jobs.job_status = 'failed'::provisioner_job_status AND
provisioner_jobs.completed_at IS NOT NULL AND
(@now :: timestamptz) - provisioner_jobs.completed_at > (INTERVAL '1 millisecond' * (templates.failure_ttl / 1000000))
)
) AND workspaces.deleted = 'false';
@@ -727,5 +761,3 @@ WHERE
-- Authorize Filter clause will be injected below in GetAuthorizedWorkspacesAndAgentsByOwnerID
-- @authorize_filter
GROUP BY workspaces.id, workspaces.name, latest_build.job_status, latest_build.job_id, latest_build.transition;