open-nomad/scheduler/reconcile_test.go

4065 lines
118 KiB
Go
Raw Normal View History

2017-06-01 22:16:24 +00:00
package scheduler
import (
2017-06-02 23:11:29 +00:00
"fmt"
2017-06-01 22:16:24 +00:00
"log"
"os"
"reflect"
"regexp"
"strconv"
"testing"
2017-06-02 23:11:29 +00:00
"time"
2017-06-01 22:16:24 +00:00
2017-06-02 23:11:29 +00:00
"github.com/hashicorp/nomad/helper"
"github.com/hashicorp/nomad/helper/uuid"
2017-06-01 22:16:24 +00:00
"github.com/hashicorp/nomad/nomad/mock"
"github.com/hashicorp/nomad/nomad/structs"
"github.com/kr/pretty"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
2017-06-01 22:16:24 +00:00
)
/*
Basic Tests:
2017-06-02 23:11:29 +00:00
Place when there is nothing in the cluster
Place remainder when there is some in the cluster
Scale down from n to n-m where n != m
Scale down from n to zero
Inplace upgrade test
Inplace upgrade and scale up test
Inplace upgrade and scale down test
Destructive upgrade
Destructive upgrade and scale up test
Destructive upgrade and scale down test
Handle lost nodes
Handle lost nodes and scale up
Handle lost nodes and scale down
Handle draining nodes
Handle draining nodes and scale up
Handle draining nodes and scale down
Handle task group being removed
Handle job being stopped both as .Stopped and nil
Place more that one group
Handle delayed rescheduling failed allocs for batch jobs
Handle delayed rescheduling failed allocs for service jobs
Handle eligible now rescheduling failed allocs for batch jobs
Handle eligible now rescheduling failed allocs for service jobs
Previously rescheduled allocs should not be rescheduled again
Aggregated evaluations for allocations that fail close together
2017-06-02 23:11:29 +00:00
Update stanza Tests:
2017-06-02 23:11:29 +00:00
Stopped job cancels any active deployment
Stopped job doesn't cancel terminal deployment
JobIndex change cancels any active deployment
2017-12-13 17:36:03 +00:00
JobIndex change doesn't cancels any terminal deployment
2017-06-02 23:11:29 +00:00
Destructive changes create deployment and get rolled out via max_parallelism
2017-06-06 21:08:46 +00:00
Don't create a deployment if there are no changes
Deployment created by all inplace updates
2017-06-02 23:11:29 +00:00
Paused or failed deployment doesn't create any more canaries
Paused or failed deployment doesn't do any placements unless replacing lost allocs
2017-06-02 23:11:29 +00:00
Paused or failed deployment doesn't do destructive updates
Paused does do migrations
Failed deployment doesn't do migrations
Canary that is on a draining node
Canary that is on a lost node
Stop old canaries
Create new canaries on job change
Create new canaries on job change while scaling up
Create new canaries on job change while scaling down
Fill canaries if partial placement
Promote canaries unblocks max_parallel
Promote canaries when canaries == count
Only place as many as are healthy in deployment
Limit calculation accounts for healthy allocs on migrating/lost nodes
Failed deployment should not place anything
2017-06-06 21:08:46 +00:00
Run after canaries have been promoted, new allocs have been rolled out and there is no deployment
2017-07-05 19:50:40 +00:00
Failed deployment cancels non-promoted task groups
2017-07-06 15:39:16 +00:00
Failed deployment and updated job works
Finished deployment gets marked as complete
Change job change while scaling up
Update the job when all allocations from the previous job haven't been placed yet.
2018-01-19 17:58:59 +00:00
Paused or failed deployment doesn't do any rescheduling of failed allocs
Running deployment with failed allocs doesn't do any rescheduling of failed allocs
2017-06-01 22:16:24 +00:00
*/
2017-06-02 23:11:29 +00:00
var (
canaryUpdate = &structs.UpdateStrategy{
Canary: 2,
MaxParallel: 2,
HealthCheck: structs.UpdateStrategyHealthCheck_Checks,
MinHealthyTime: 10 * time.Second,
HealthyDeadline: 10 * time.Minute,
Stagger: 31 * time.Second,
2017-06-02 23:11:29 +00:00
}
noCanaryUpdate = &structs.UpdateStrategy{
MaxParallel: 4,
HealthCheck: structs.UpdateStrategyHealthCheck_Checks,
MinHealthyTime: 10 * time.Second,
HealthyDeadline: 10 * time.Minute,
Stagger: 31 * time.Second,
2017-06-02 23:11:29 +00:00
}
)
2017-06-01 22:16:24 +00:00
func testLogger() *log.Logger {
return log.New(os.Stderr, "", log.LstdFlags)
}
func allocUpdateFnIgnore(*structs.Allocation, *structs.Job, *structs.TaskGroup) (bool, bool, *structs.Allocation) {
return true, false, nil
}
func allocUpdateFnDestructive(*structs.Allocation, *structs.Job, *structs.TaskGroup) (bool, bool, *structs.Allocation) {
return false, true, nil
}
func allocUpdateFnInplace(existing *structs.Allocation, _ *structs.Job, newTG *structs.TaskGroup) (bool, bool, *structs.Allocation) {
// Create a shallow copy
2018-03-23 23:55:21 +00:00
newAlloc := existing.CopySkipJob()
2017-06-01 22:16:24 +00:00
newAlloc.TaskResources = make(map[string]*structs.Resources)
// Use the new task resources but keep the network from the old
for _, task := range newTG.Tasks {
r := task.Resources.Copy()
r.Networks = existing.TaskResources[task.Name].Networks
newAlloc.TaskResources[task.Name] = r
}
return false, false, newAlloc
}
2017-06-02 23:11:29 +00:00
func allocUpdateFnMock(handled map[string]allocUpdateType, unhandled allocUpdateType) allocUpdateType {
return func(existing *structs.Allocation, newJob *structs.Job, newTG *structs.TaskGroup) (bool, bool, *structs.Allocation) {
if fn, ok := handled[existing.ID]; ok {
return fn(existing, newJob, newTG)
}
return unhandled(existing, newJob, newTG)
}
}
2017-06-01 22:16:24 +00:00
var (
// AllocationIndexRegex is a regular expression to find the allocation index.
allocationIndexRegex = regexp.MustCompile(".+\\[(\\d+)\\]$")
)
// allocNameToIndex returns the index of the allocation.
func allocNameToIndex(name string) uint {
matches := allocationIndexRegex.FindStringSubmatch(name)
if len(matches) != 2 {
return 0
}
index, err := strconv.Atoi(matches[1])
if err != nil {
return 0
}
return uint(index)
}
func assertNamesHaveIndexes(t *testing.T, indexes []int, names []string) {
t.Helper()
2017-06-01 22:16:24 +00:00
m := make(map[uint]int)
for _, i := range indexes {
m[uint(i)] += 1
}
for _, n := range names {
index := allocNameToIndex(n)
val, contained := m[index]
if !contained {
t.Fatalf("Unexpected index %d from name %s\nAll names: %v", index, n, names)
}
val--
if val < 0 {
t.Fatalf("Index %d repeated too many times\nAll names: %v", index, names)
}
m[index] = val
}
for k, remainder := range m {
if remainder != 0 {
t.Fatalf("Index %d has %d remaining uses expected\nAll names: %v", k, remainder, names)
}
}
}
2017-07-06 16:55:39 +00:00
func assertNoCanariesStopped(t *testing.T, d *structs.Deployment, stop []allocStopResult) {
t.Helper()
2017-07-06 16:55:39 +00:00
canaryIndex := make(map[string]struct{})
for _, state := range d.TaskGroups {
for _, c := range state.PlacedCanaries {
canaryIndex[c] = struct{}{}
}
}
2017-06-02 23:11:29 +00:00
for _, s := range stop {
2017-07-06 16:55:39 +00:00
if _, ok := canaryIndex[s.alloc.ID]; ok {
2017-06-02 23:11:29 +00:00
t.Fatalf("Stopping canary alloc %q %q", s.alloc.ID, s.alloc.Name)
}
}
}
func assertPlaceResultsHavePreviousAllocs(t *testing.T, numPrevious int, place []allocPlaceResult) {
t.Helper()
2017-06-02 23:11:29 +00:00
names := make(map[string]struct{}, numPrevious)
found := 0
for _, p := range place {
if _, ok := names[p.name]; ok {
t.Fatalf("Name %q already placed", p.name)
}
names[p.name] = struct{}{}
if p.previousAlloc == nil {
continue
}
if act := p.previousAlloc.Name; p.name != act {
t.Fatalf("Name mismatch on previous alloc; got %q; want %q", act, p.name)
}
found++
}
if numPrevious != found {
t.Fatalf("wanted %d; got %d placements with previous allocs", numPrevious, found)
}
}
func assertPlacementsAreRescheduled(t *testing.T, numRescheduled int, place []allocPlaceResult) {
t.Helper()
names := make(map[string]struct{}, numRescheduled)
found := 0
for _, p := range place {
if _, ok := names[p.name]; ok {
t.Fatalf("Name %q already placed", p.name)
}
names[p.name] = struct{}{}
if p.previousAlloc == nil {
continue
}
if p.reschedule {
found++
}
}
if numRescheduled != found {
t.Fatalf("wanted %d; got %d placements that are rescheduled", numRescheduled, found)
}
}
2017-06-01 22:16:24 +00:00
func intRange(pairs ...int) []int {
if len(pairs)%2 != 0 {
return nil
}
var r []int
for i := 0; i < len(pairs); i += 2 {
for j := pairs[i]; j <= pairs[i+1]; j++ {
r = append(r, j)
}
}
return r
}
func placeResultsToNames(place []allocPlaceResult) []string {
names := make([]string, 0, len(place))
for _, p := range place {
names = append(names, p.name)
}
return names
}
2017-07-15 23:31:33 +00:00
func destructiveResultsToNames(destructive []allocDestructiveResult) []string {
names := make([]string, 0, len(destructive))
for _, d := range destructive {
names = append(names, d.placeName)
}
return names
}
2017-06-01 22:16:24 +00:00
func stopResultsToNames(stop []allocStopResult) []string {
names := make([]string, 0, len(stop))
for _, s := range stop {
names = append(names, s.alloc.Name)
}
return names
}
2018-03-26 18:06:21 +00:00
func attributeUpdatesToNames(attributeUpdates map[string]*structs.Allocation) []string {
names := make([]string, 0, len(attributeUpdates))
for _, a := range attributeUpdates {
2018-03-23 23:55:21 +00:00
names = append(names, a.Name)
}
return names
}
2017-06-01 22:16:24 +00:00
func allocsToNames(allocs []*structs.Allocation) []string {
names := make([]string, 0, len(allocs))
for _, a := range allocs {
names = append(names, a.Name)
}
return names
}
type resultExpectation struct {
createDeployment *structs.Deployment
2017-06-02 23:11:29 +00:00
deploymentUpdates []*structs.DeploymentStatusUpdate
2017-06-01 22:16:24 +00:00
place int
2017-07-15 23:31:33 +00:00
destructive int
2017-06-01 22:16:24 +00:00
inplace int
2018-03-26 18:06:21 +00:00
attributeUpdates int
2017-06-01 22:16:24 +00:00
stop int
desiredTGUpdates map[string]*structs.DesiredUpdates
}
func assertResults(t *testing.T, r *reconcileResults, exp *resultExpectation) {
t.Helper()
assert := assert.New(t)
if exp.createDeployment != nil && r.deployment == nil {
2017-06-01 22:16:24 +00:00
t.Fatalf("Expect a created deployment got none")
} else if exp.createDeployment == nil && r.deployment != nil {
t.Fatalf("Expect no created deployment; got %#v", r.deployment)
} else if exp.createDeployment != nil && r.deployment != nil {
2017-06-02 23:11:29 +00:00
// Clear the deployment ID
r.deployment.ID, exp.createDeployment.ID = "", ""
if !reflect.DeepEqual(r.deployment, exp.createDeployment) {
2017-06-06 21:08:46 +00:00
t.Fatalf("Unexpected createdDeployment; got\n %#v\nwant\n%#v\nDiff: %v",
r.deployment, exp.createDeployment, pretty.Diff(r.deployment, exp.createDeployment))
2017-06-02 23:11:29 +00:00
}
2017-06-01 22:16:24 +00:00
}
assert.EqualValues(exp.deploymentUpdates, r.deploymentUpdates, "Expected Deployment Updates")
assert.Len(r.place, exp.place, "Expected Placements")
assert.Len(r.destructiveUpdate, exp.destructive, "Expected Destructive")
assert.Len(r.inplaceUpdate, exp.inplace, "Expected Inplace Updates")
assert.Len(r.attributeUpdates, exp.attributeUpdates, "Expected Attribute Updates")
assert.Len(r.stop, exp.stop, "Expected Stops")
assert.EqualValues(exp.desiredTGUpdates, r.desiredTGUpdates, "Expected Desired TG Update Annotations")
2017-06-01 22:16:24 +00:00
}
// Tests the reconciler properly handles placements for a job that has no
// existing allocations
func TestReconciler_Place_NoExisting(t *testing.T) {
job := mock.Job()
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, nil, nil, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-06-01 22:16:24 +00:00
place: 10,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 10,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 9), placeResultsToNames(r.place))
}
// Tests the reconciler properly handles placements for a job that has some
// existing allocations
func TestReconciler_Place_Existing(t *testing.T) {
job := mock.Job()
// Create 3 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 5; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-01 22:16:24 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, nil, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-06-01 22:16:24 +00:00
place: 5,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 5,
Ignore: 5,
},
},
})
assertNamesHaveIndexes(t, intRange(5, 9), placeResultsToNames(r.place))
}
// Tests the reconciler properly handles stopping allocations for a job that has
// scaled down
func TestReconciler_ScaleDown_Partial(t *testing.T) {
// Has desired 10
job := mock.Job()
// Create 20 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 20; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-01 22:16:24 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, nil, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-06-01 22:16:24 +00:00
place: 0,
inplace: 0,
stop: 10,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Ignore: 10,
Stop: 10,
},
},
})
assertNamesHaveIndexes(t, intRange(10, 19), stopResultsToNames(r.stop))
}
// Tests the reconciler properly handles stopping allocations for a job that has
// scaled down to zero desired
func TestReconciler_ScaleDown_Zero(t *testing.T) {
// Set desired 0
job := mock.Job()
job.TaskGroups[0].Count = 0
// Create 20 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 20; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-01 22:16:24 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, nil, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-06-01 22:16:24 +00:00
place: 0,
inplace: 0,
stop: 20,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Stop: 20,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 19), stopResultsToNames(r.stop))
}
// Tests the reconciler properly handles stopping allocations for a job that has
// scaled down to zero desired where allocs have duplicate names
func TestReconciler_ScaleDown_Zero_DuplicateNames(t *testing.T) {
// Set desired 0
job := mock.Job()
job.TaskGroups[0].Count = 0
// Create 20 existing allocations
var allocs []*structs.Allocation
var expectedStopped []int
for i := 0; i < 20; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i%2))
allocs = append(allocs, alloc)
expectedStopped = append(expectedStopped, i%2)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, nil, "")
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 0,
inplace: 0,
stop: 20,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Stop: 20,
},
},
})
assertNamesHaveIndexes(t, expectedStopped, stopResultsToNames(r.stop))
}
2017-06-01 22:16:24 +00:00
// Tests the reconciler properly handles inplace upgrading allocations
func TestReconciler_Inplace(t *testing.T) {
job := mock.Job()
// Create 10 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-01 22:16:24 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnInplace, false, job.ID, job, nil, allocs, nil, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-06-01 22:16:24 +00:00
place: 0,
inplace: 10,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
InPlaceUpdate: 10,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 9), allocsToNames(r.inplaceUpdate))
}
// Tests the reconciler properly handles inplace upgrading allocations while
// scaling up
func TestReconciler_Inplace_ScaleUp(t *testing.T) {
// Set desired 15
job := mock.Job()
job.TaskGroups[0].Count = 15
// Create 10 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-01 22:16:24 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnInplace, false, job.ID, job, nil, allocs, nil, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-06-01 22:16:24 +00:00
place: 5,
inplace: 10,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 5,
InPlaceUpdate: 10,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 9), allocsToNames(r.inplaceUpdate))
assertNamesHaveIndexes(t, intRange(10, 14), placeResultsToNames(r.place))
}
// Tests the reconciler properly handles inplace upgrading allocations while
// scaling down
func TestReconciler_Inplace_ScaleDown(t *testing.T) {
// Set desired 5
job := mock.Job()
job.TaskGroups[0].Count = 5
// Create 10 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-01 22:16:24 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnInplace, false, job.ID, job, nil, allocs, nil, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-06-01 22:16:24 +00:00
place: 0,
inplace: 5,
stop: 5,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Stop: 5,
InPlaceUpdate: 5,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 4), allocsToNames(r.inplaceUpdate))
assertNamesHaveIndexes(t, intRange(5, 9), stopResultsToNames(r.stop))
}
// Tests the reconciler properly handles destructive upgrading allocations
func TestReconciler_Destructive(t *testing.T) {
job := mock.Job()
// Create 10 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-01 22:16:24 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnDestructive, false, job.ID, job, nil, allocs, nil, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-07-15 23:31:33 +00:00
destructive: 10,
2017-06-01 22:16:24 +00:00
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
DestructiveUpdate: 10,
},
},
})
2017-07-15 23:31:33 +00:00
assertNamesHaveIndexes(t, intRange(0, 9), destructiveResultsToNames(r.destructiveUpdate))
2017-06-01 22:16:24 +00:00
}
// Tests the reconciler properly handles destructive upgrading allocations while
// scaling up
func TestReconciler_Destructive_ScaleUp(t *testing.T) {
// Set desired 15
job := mock.Job()
job.TaskGroups[0].Count = 15
// Create 10 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-01 22:16:24 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnDestructive, false, job.ID, job, nil, allocs, nil, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-07-15 23:31:33 +00:00
place: 5,
destructive: 10,
2017-06-01 22:16:24 +00:00
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 5,
DestructiveUpdate: 10,
},
},
})
2017-07-15 23:31:33 +00:00
assertNamesHaveIndexes(t, intRange(0, 9), destructiveResultsToNames(r.destructiveUpdate))
assertNamesHaveIndexes(t, intRange(10, 14), placeResultsToNames(r.place))
2017-06-01 22:16:24 +00:00
}
// Tests the reconciler properly handles destructive upgrading allocations while
// scaling down
func TestReconciler_Destructive_ScaleDown(t *testing.T) {
// Set desired 5
job := mock.Job()
job.TaskGroups[0].Count = 5
// Create 10 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-01 22:16:24 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnDestructive, false, job.ID, job, nil, allocs, nil, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-07-15 23:31:33 +00:00
destructive: 5,
stop: 5,
2017-06-01 22:16:24 +00:00
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Stop: 5,
DestructiveUpdate: 5,
},
},
})
2017-07-15 23:31:33 +00:00
assertNamesHaveIndexes(t, intRange(5, 9), stopResultsToNames(r.stop))
assertNamesHaveIndexes(t, intRange(0, 4), destructiveResultsToNames(r.destructiveUpdate))
2017-06-01 22:16:24 +00:00
}
// Tests the reconciler properly handles lost nodes with allocations
func TestReconciler_LostNode(t *testing.T) {
job := mock.Job()
// Create 10 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-01 22:16:24 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
}
// Build a map of tainted nodes
tainted := make(map[string]*structs.Node, 2)
for i := 0; i < 2; i++ {
n := mock.Node()
n.ID = allocs[i].NodeID
n.Status = structs.NodeStatusDown
tainted[n.ID] = n
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, tainted, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-06-01 22:16:24 +00:00
place: 2,
inplace: 0,
stop: 2,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 2,
2017-06-02 23:11:29 +00:00
Stop: 2,
2017-06-01 22:16:24 +00:00
Ignore: 8,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 1), stopResultsToNames(r.stop))
assertNamesHaveIndexes(t, intRange(0, 1), placeResultsToNames(r.place))
}
// Tests the reconciler properly handles lost nodes with allocations while
// scaling up
func TestReconciler_LostNode_ScaleUp(t *testing.T) {
// Set desired 15
job := mock.Job()
job.TaskGroups[0].Count = 15
// Create 10 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-01 22:16:24 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
}
// Build a map of tainted nodes
tainted := make(map[string]*structs.Node, 2)
for i := 0; i < 2; i++ {
n := mock.Node()
n.ID = allocs[i].NodeID
n.Status = structs.NodeStatusDown
tainted[n.ID] = n
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, tainted, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-06-01 22:16:24 +00:00
place: 7,
inplace: 0,
stop: 2,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 7,
2017-06-02 23:11:29 +00:00
Stop: 2,
2017-06-01 22:16:24 +00:00
Ignore: 8,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 1), stopResultsToNames(r.stop))
assertNamesHaveIndexes(t, intRange(0, 1, 10, 14), placeResultsToNames(r.place))
}
// Tests the reconciler properly handles lost nodes with allocations while
// scaling down
func TestReconciler_LostNode_ScaleDown(t *testing.T) {
// Set desired 5
job := mock.Job()
job.TaskGroups[0].Count = 5
// Create 10 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-01 22:16:24 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
}
// Build a map of tainted nodes
tainted := make(map[string]*structs.Node, 2)
for i := 0; i < 2; i++ {
n := mock.Node()
n.ID = allocs[i].NodeID
n.Status = structs.NodeStatusDown
tainted[n.ID] = n
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, tainted, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-06-01 22:16:24 +00:00
place: 0,
inplace: 0,
stop: 5,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Stop: 5,
Ignore: 5,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 1, 7, 9), stopResultsToNames(r.stop))
}
// Tests the reconciler properly handles draining nodes with allocations
func TestReconciler_DrainNode(t *testing.T) {
job := mock.Job()
// Create 10 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-01 22:16:24 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
}
// Build a map of tainted nodes
tainted := make(map[string]*structs.Node, 2)
for i := 0; i < 2; i++ {
n := mock.Node()
n.ID = allocs[i].NodeID
allocs[i].DesiredTransition.Migrate = helper.BoolToPtr(true)
2017-06-01 22:16:24 +00:00
n.Drain = true
tainted[n.ID] = n
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, tainted, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-06-01 22:16:24 +00:00
place: 2,
inplace: 0,
stop: 2,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Migrate: 2,
Ignore: 8,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 1), stopResultsToNames(r.stop))
assertNamesHaveIndexes(t, intRange(0, 1), placeResultsToNames(r.place))
2017-06-02 23:11:29 +00:00
assertPlaceResultsHavePreviousAllocs(t, 2, r.place)
// These should not have the reschedule field set
assertPlacementsAreRescheduled(t, 0, r.place)
2017-06-01 22:16:24 +00:00
}
// Tests the reconciler properly handles draining nodes with allocations while
// scaling up
func TestReconciler_DrainNode_ScaleUp(t *testing.T) {
// Set desired 15
job := mock.Job()
job.TaskGroups[0].Count = 15
// Create 10 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-01 22:16:24 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
}
// Build a map of tainted nodes
tainted := make(map[string]*structs.Node, 2)
for i := 0; i < 2; i++ {
n := mock.Node()
n.ID = allocs[i].NodeID
allocs[i].DesiredTransition.Migrate = helper.BoolToPtr(true)
2017-06-01 22:16:24 +00:00
n.Drain = true
tainted[n.ID] = n
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, tainted, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-06-01 22:16:24 +00:00
place: 7,
inplace: 0,
stop: 2,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 5,
Migrate: 2,
Ignore: 8,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 1), stopResultsToNames(r.stop))
assertNamesHaveIndexes(t, intRange(0, 1, 10, 14), placeResultsToNames(r.place))
2017-06-02 23:11:29 +00:00
assertPlaceResultsHavePreviousAllocs(t, 2, r.place)
// These should not have the reschedule field set
assertPlacementsAreRescheduled(t, 0, r.place)
2017-06-01 22:16:24 +00:00
}
// Tests the reconciler properly handles draining nodes with allocations while
// scaling down
func TestReconciler_DrainNode_ScaleDown(t *testing.T) {
// Set desired 8
job := mock.Job()
job.TaskGroups[0].Count = 8
// Create 10 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-01 22:16:24 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
}
// Build a map of tainted nodes
tainted := make(map[string]*structs.Node, 3)
for i := 0; i < 3; i++ {
n := mock.Node()
n.ID = allocs[i].NodeID
allocs[i].DesiredTransition.Migrate = helper.BoolToPtr(true)
2017-06-01 22:16:24 +00:00
n.Drain = true
tainted[n.ID] = n
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, tainted, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-06-01 22:16:24 +00:00
place: 1,
inplace: 0,
stop: 3,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Migrate: 1,
Stop: 2,
Ignore: 7,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 2), stopResultsToNames(r.stop))
assertNamesHaveIndexes(t, intRange(0, 0), placeResultsToNames(r.place))
2017-06-02 23:11:29 +00:00
assertPlaceResultsHavePreviousAllocs(t, 1, r.place)
// These should not have the reschedule field set
assertPlacementsAreRescheduled(t, 0, r.place)
2017-06-01 22:16:24 +00:00
}
// Tests the reconciler properly handles a task group being removed
func TestReconciler_RemovedTG(t *testing.T) {
job := mock.Job()
// Create 10 allocations for a tg that no longer exists
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-01 22:16:24 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
}
oldName := job.TaskGroups[0].Name
newName := "different"
job.TaskGroups[0].Name = newName
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, nil, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-06-01 22:16:24 +00:00
place: 10,
inplace: 0,
stop: 10,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
oldName: {
Stop: 10,
},
newName: {
Place: 10,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 9), stopResultsToNames(r.stop))
assertNamesHaveIndexes(t, intRange(0, 9), placeResultsToNames(r.place))
}
// Tests the reconciler properly handles a job in stopped states
func TestReconciler_JobStopped(t *testing.T) {
job := mock.Job()
job.Stop = true
cases := []struct {
name string
job *structs.Job
jobID, taskGroup string
}{
{
name: "stopped job",
job: job,
jobID: job.ID,
taskGroup: job.TaskGroups[0].Name,
},
{
name: "nil job",
job: nil,
jobID: "foo",
taskGroup: "bar",
},
}
for _, c := range cases {
t.Run(c.name, func(t *testing.T) {
2017-06-02 23:11:29 +00:00
// Create 10 allocations
2017-06-01 22:16:24 +00:00
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = c.job
alloc.JobID = c.jobID
alloc.NodeID = uuid.Generate()
2017-06-01 22:16:24 +00:00
alloc.Name = structs.AllocName(c.jobID, c.taskGroup, uint(i))
alloc.TaskGroup = c.taskGroup
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, c.jobID, c.job, nil, allocs, nil, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-06-01 22:16:24 +00:00
place: 0,
inplace: 0,
stop: 10,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
c.taskGroup: {
Stop: 10,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 9), stopResultsToNames(r.stop))
})
}
}
// Tests the reconciler properly handles jobs with multiple task groups
func TestReconciler_MultiTG(t *testing.T) {
job := mock.Job()
tg2 := job.TaskGroups[0].Copy()
tg2.Name = "foo"
job.TaskGroups = append(job.TaskGroups, tg2)
// Create 2 existing allocations for the first tg
var allocs []*structs.Allocation
for i := 0; i < 2; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-01 22:16:24 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, nil, "")
2017-06-01 22:16:24 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-06-02 23:11:29 +00:00
deploymentUpdates: nil,
2017-06-01 22:16:24 +00:00
place: 18,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 8,
Ignore: 2,
},
tg2.Name: {
Place: 10,
},
},
})
assertNamesHaveIndexes(t, intRange(2, 9, 0, 9), placeResultsToNames(r.place))
}
2017-06-02 23:11:29 +00:00
// Tests delayed rescheduling of failed batch allocations
func TestReconciler_RescheduleLater_Batch(t *testing.T) {
require := require.New(t)
2018-03-23 23:55:21 +00:00
// Set desired 4
2018-01-19 17:58:59 +00:00
job := mock.Job()
job.TaskGroups[0].Count = 4
now := time.Now()
2018-03-23 23:55:21 +00:00
// Set up reschedule policy
delayDur := 15 * time.Second
2018-03-26 19:45:09 +00:00
job.TaskGroups[0].ReschedulePolicy = &structs.ReschedulePolicy{Attempts: 3, Interval: 24 * time.Hour, Delay: delayDur, DelayFunction: "constant"}
tgName := job.TaskGroups[0].Name
2018-03-23 23:55:21 +00:00
// Create 6 existing allocations - 2 running, 1 complete and 3 failed
var allocs []*structs.Allocation
for i := 0; i < 6; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
alloc.ClientStatus = structs.AllocClientStatusRunning
}
2018-03-23 23:55:21 +00:00
// Mark 3 as failed with restart tracking info
allocs[0].ClientStatus = structs.AllocClientStatusFailed
allocs[0].NextAllocation = allocs[1].ID
allocs[1].ClientStatus = structs.AllocClientStatusFailed
allocs[1].RescheduleTracker = &structs.RescheduleTracker{Events: []*structs.RescheduleEvent{
{RescheduleTime: time.Now().Add(-1 * time.Hour).UTC().UnixNano(),
PrevAllocID: allocs[0].ID,
PrevNodeID: uuid.Generate(),
},
}}
allocs[1].NextAllocation = allocs[2].ID
allocs[2].ClientStatus = structs.AllocClientStatusFailed
allocs[2].TaskStates = map[string]*structs.TaskState{tgName: {State: "start",
StartedAt: now.Add(-1 * time.Hour),
FinishedAt: now}}
allocs[2].RescheduleTracker = &structs.RescheduleTracker{Events: []*structs.RescheduleEvent{
{RescheduleTime: time.Now().Add(-2 * time.Hour).UTC().UnixNano(),
PrevAllocID: allocs[0].ID,
PrevNodeID: uuid.Generate(),
},
{RescheduleTime: time.Now().Add(-1 * time.Hour).UTC().UnixNano(),
PrevAllocID: allocs[1].ID,
PrevNodeID: uuid.Generate(),
},
}}
2018-03-23 23:55:21 +00:00
// Mark one as complete
allocs[5].ClientStatus = structs.AllocClientStatusComplete
2018-01-19 17:58:59 +00:00
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, true, job.ID, job, nil, allocs, nil, uuid.Generate())
r := reconciler.Compute()
// Two reschedule attempts were already made, one more can be made at a future time
// Verify that the follow up eval has the expected waitUntil time
evals := r.desiredFollowupEvals[tgName]
require.NotNil(evals)
require.Equal(1, len(evals))
require.Equal(now.Add(delayDur), evals[0].WaitUntil)
// Alloc 5 should not be replaced because it is terminal
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 0,
2018-03-23 23:55:21 +00:00
inplace: 0,
2018-03-26 18:06:21 +00:00
attributeUpdates: 1,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 0,
InPlaceUpdate: 0,
Ignore: 4,
},
},
})
2018-03-26 18:06:21 +00:00
assertNamesHaveIndexes(t, intRange(2, 2), attributeUpdatesToNames(r.attributeUpdates))
2018-03-23 23:55:21 +00:00
// Verify that the followup evalID field is set correctly
var annotated *structs.Allocation
2018-03-26 18:06:21 +00:00
for _, a := range r.attributeUpdates {
2018-03-23 23:55:21 +00:00
annotated = a
}
require.Equal(evals[0].ID, annotated.FollowupEvalID)
}
// Tests delayed rescheduling of failed batch allocations and batching of allocs
// with fail times that are close together
func TestReconciler_RescheduleLaterWithBatchedEvals_Batch(t *testing.T) {
require := require.New(t)
2018-03-23 23:55:21 +00:00
// Set desired 4
job := mock.Job()
job.TaskGroups[0].Count = 10
now := time.Now()
2018-03-23 23:55:21 +00:00
2018-01-19 17:58:59 +00:00
// Set up reschedule policy
delayDur := 15 * time.Second
2018-03-26 19:45:09 +00:00
job.TaskGroups[0].ReschedulePolicy = &structs.ReschedulePolicy{Attempts: 3, Interval: 24 * time.Hour, Delay: delayDur, DelayFunction: "constant"}
tgName := job.TaskGroups[0].Name
2018-03-23 23:55:21 +00:00
// Create 10 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
alloc.ClientStatus = structs.AllocClientStatusRunning
}
2018-03-23 23:55:21 +00:00
// Mark 5 as failed with fail times very close together
for i := 0; i < 5; i++ {
allocs[i].ClientStatus = structs.AllocClientStatusFailed
allocs[i].TaskStates = map[string]*structs.TaskState{tgName: {State: "start",
StartedAt: now.Add(-1 * time.Hour),
FinishedAt: now.Add(time.Duration(50*i) * time.Millisecond)}}
}
// Mark two more as failed several seconds later
for i := 5; i < 7; i++ {
allocs[i].ClientStatus = structs.AllocClientStatusFailed
allocs[i].TaskStates = map[string]*structs.TaskState{tgName: {State: "start",
StartedAt: now.Add(-1 * time.Hour),
FinishedAt: now.Add(10 * time.Second)}}
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, true, job.ID, job, nil, allocs, nil, uuid.Generate())
r := reconciler.Compute()
// Verify that two follow up evals were created
evals := r.desiredFollowupEvals[tgName]
require.NotNil(evals)
require.Equal(2, len(evals))
// Verify expected WaitUntil values for both batched evals
require.Equal(now.Add(delayDur), evals[0].WaitUntil)
secondBatchDuration := delayDur + 10*time.Second
require.Equal(now.Add(secondBatchDuration), evals[1].WaitUntil)
2018-01-19 17:58:59 +00:00
// Alloc 5 should not be replaced because it is terminal
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 0,
2018-03-23 23:55:21 +00:00
inplace: 0,
2018-03-26 18:06:21 +00:00
attributeUpdates: 7,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 0,
InPlaceUpdate: 0,
Ignore: 10,
},
},
})
2018-03-26 18:06:21 +00:00
assertNamesHaveIndexes(t, intRange(0, 6), attributeUpdatesToNames(r.attributeUpdates))
2018-03-23 23:55:21 +00:00
// Verify that the followup evalID field is set correctly
2018-03-26 18:06:21 +00:00
for _, alloc := range r.attributeUpdates {
if allocNameToIndex(alloc.Name) < 5 {
require.Equal(evals[0].ID, alloc.FollowupEvalID)
} else if allocNameToIndex(alloc.Name) < 7 {
require.Equal(evals[1].ID, alloc.FollowupEvalID)
} else {
t.Fatalf("Unexpected alloc name in Inplace results %v", alloc.Name)
}
}
}
// Tests rescheduling failed batch allocations
func TestReconciler_RescheduleNow_Batch(t *testing.T) {
require := require.New(t)
// Set desired 4
job := mock.Job()
job.TaskGroups[0].Count = 4
now := time.Now()
// Set up reschedule policy
2018-03-26 19:45:09 +00:00
job.TaskGroups[0].ReschedulePolicy = &structs.ReschedulePolicy{Attempts: 3, Interval: 24 * time.Hour, Delay: 5 * time.Second, DelayFunction: "constant"}
tgName := job.TaskGroups[0].Name
// Create 6 existing allocations - 2 running, 1 complete and 3 failed
2018-01-19 17:58:59 +00:00
var allocs []*structs.Allocation
for i := 0; i < 6; i++ {
2018-01-19 17:58:59 +00:00
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
alloc.ClientStatus = structs.AllocClientStatusRunning
}
// Mark 3 as failed with restart tracking info
2018-01-19 17:58:59 +00:00
allocs[0].ClientStatus = structs.AllocClientStatusFailed
allocs[0].NextAllocation = allocs[1].ID
allocs[1].ClientStatus = structs.AllocClientStatusFailed
allocs[1].RescheduleTracker = &structs.RescheduleTracker{Events: []*structs.RescheduleEvent{
{RescheduleTime: time.Now().Add(-1 * time.Hour).UTC().UnixNano(),
PrevAllocID: allocs[0].ID,
PrevNodeID: uuid.Generate(),
},
}}
allocs[1].NextAllocation = allocs[2].ID
allocs[2].ClientStatus = structs.AllocClientStatusFailed
allocs[2].TaskStates = map[string]*structs.TaskState{tgName: {State: "start",
StartedAt: now.Add(-1 * time.Hour),
FinishedAt: now.Add(-5 * time.Second)}}
allocs[2].FollowupEvalID = uuid.Generate()
allocs[2].RescheduleTracker = &structs.RescheduleTracker{Events: []*structs.RescheduleEvent{
{RescheduleTime: time.Now().Add(-2 * time.Hour).UTC().UnixNano(),
PrevAllocID: allocs[0].ID,
PrevNodeID: uuid.Generate(),
},
{RescheduleTime: time.Now().Add(-1 * time.Hour).UTC().UnixNano(),
PrevAllocID: allocs[1].ID,
PrevNodeID: uuid.Generate(),
},
}}
2018-01-19 17:58:59 +00:00
// Mark one as complete
allocs[5].ClientStatus = structs.AllocClientStatusComplete
2018-01-19 17:58:59 +00:00
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, true, job.ID, job, nil, allocs, nil, "")
reconciler.now = now
2018-01-19 17:58:59 +00:00
r := reconciler.Compute()
// Verify that no follow up evals were created
evals := r.desiredFollowupEvals[tgName]
require.Nil(evals)
// Two reschedule attempts were made, one more can be made now
// Alloc 5 should not be replaced because it is terminal
2018-01-19 17:58:59 +00:00
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 1,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 1,
Ignore: 3,
2018-01-19 17:58:59 +00:00
},
},
})
assertNamesHaveIndexes(t, intRange(2, 2), placeResultsToNames(r.place))
2018-01-19 17:58:59 +00:00
assertPlaceResultsHavePreviousAllocs(t, 1, r.place)
assertPlacementsAreRescheduled(t, 1, r.place)
2018-01-19 17:58:59 +00:00
}
// Tests rescheduling failed service allocations with desired state stop
func TestReconciler_RescheduleLater_Service(t *testing.T) {
require := require.New(t)
2018-03-23 23:55:21 +00:00
2018-01-19 17:58:59 +00:00
// Set desired 5
job := mock.Job()
job.TaskGroups[0].Count = 5
tgName := job.TaskGroups[0].Name
now := time.Now()
2018-03-23 23:55:21 +00:00
2018-01-19 17:58:59 +00:00
// Set up reschedule policy
delayDur := 15 * time.Second
2018-03-13 15:06:26 +00:00
job.TaskGroups[0].ReschedulePolicy = &structs.ReschedulePolicy{Attempts: 1, Interval: 24 * time.Hour, Delay: delayDur, MaxDelay: 1 * time.Hour}
2018-01-19 17:58:59 +00:00
// Create 5 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 5; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
alloc.ClientStatus = structs.AllocClientStatusRunning
}
2018-03-23 23:55:21 +00:00
2018-01-19 17:58:59 +00:00
// Mark two as failed
allocs[0].ClientStatus = structs.AllocClientStatusFailed
2018-03-23 23:55:21 +00:00
// Mark one of them as already rescheduled once
allocs[0].RescheduleTracker = &structs.RescheduleTracker{Events: []*structs.RescheduleEvent{
{RescheduleTime: time.Now().Add(-1 * time.Hour).UTC().UnixNano(),
PrevAllocID: uuid.Generate(),
PrevNodeID: uuid.Generate(),
},
}}
allocs[1].TaskStates = map[string]*structs.TaskState{tgName: {State: "start",
StartedAt: now.Add(-1 * time.Hour),
FinishedAt: now}}
2018-01-19 17:58:59 +00:00
allocs[1].ClientStatus = structs.AllocClientStatusFailed
// Mark one as desired state stop
allocs[4].DesiredStatus = structs.AllocDesiredStatusStop
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, nil, uuid.Generate())
r := reconciler.Compute()
// Should place a new placement and create a follow up eval for the delayed reschedule
// Verify that the follow up eval has the expected waitUntil time
evals := r.desiredFollowupEvals[tgName]
require.NotNil(evals)
require.Equal(1, len(evals))
require.Equal(now.Add(delayDur), evals[0].WaitUntil)
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 1,
2018-03-23 23:55:21 +00:00
inplace: 0,
2018-03-26 18:06:21 +00:00
attributeUpdates: 1,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 1,
InPlaceUpdate: 0,
Ignore: 4,
},
},
})
assertNamesHaveIndexes(t, intRange(4, 4), placeResultsToNames(r.place))
2018-03-26 18:06:21 +00:00
assertNamesHaveIndexes(t, intRange(1, 1), attributeUpdatesToNames(r.attributeUpdates))
2018-03-23 23:55:21 +00:00
// Verify that the followup evalID field is set correctly
var annotated *structs.Allocation
2018-03-26 18:06:21 +00:00
for _, a := range r.attributeUpdates {
2018-03-23 23:55:21 +00:00
annotated = a
}
require.Equal(evals[0].ID, annotated.FollowupEvalID)
}
// Tests service allocations with client status complete
func TestReconciler_Service_ClientStatusComplete(t *testing.T) {
// Set desired 5
job := mock.Job()
job.TaskGroups[0].Count = 5
// Set up reschedule policy
delayDur := 15 * time.Second
job.TaskGroups[0].ReschedulePolicy = &structs.ReschedulePolicy{
Attempts: 1,
Interval: 24 * time.Hour,
Delay: delayDur,
MaxDelay: 1 * time.Hour,
}
// Create 5 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 5; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
alloc.ClientStatus = structs.AllocClientStatusRunning
alloc.DesiredStatus = structs.AllocDesiredStatusRun
}
// Mark one as client status complete
allocs[4].ClientStatus = structs.AllocClientStatusComplete
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, nil, "")
r := reconciler.Compute()
// Should place a new placement for the alloc that was marked complete
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 1,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 1,
InPlaceUpdate: 0,
Ignore: 4,
},
},
})
assertNamesHaveIndexes(t, intRange(4, 4), placeResultsToNames(r.place))
}
// Tests service job placement with desired stop and client status complete
func TestReconciler_Service_DesiredStop_ClientStatusComplete(t *testing.T) {
// Set desired 5
job := mock.Job()
job.TaskGroups[0].Count = 5
// Set up reschedule policy
delayDur := 15 * time.Second
job.TaskGroups[0].ReschedulePolicy = &structs.ReschedulePolicy{
Attempts: 1,
Interval: 24 * time.Hour,
Delay: delayDur,
MaxDelay: 1 * time.Hour,
}
// Create 5 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 5; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
alloc.ClientStatus = structs.AllocClientStatusRunning
alloc.DesiredStatus = structs.AllocDesiredStatusRun
}
// Mark one as failed but with desired status stop
// Should not trigger rescheduling logic but should trigger a placement
allocs[4].ClientStatus = structs.AllocClientStatusFailed
allocs[4].DesiredStatus = structs.AllocDesiredStatusStop
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, nil, "")
r := reconciler.Compute()
// Should place a new placement for the alloc that was marked stopped
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 1,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 1,
InPlaceUpdate: 0,
Ignore: 4,
},
},
})
assertNamesHaveIndexes(t, intRange(4, 4), placeResultsToNames(r.place))
// Should not have any follow up evals created
require := require.New(t)
require.Equal(0, len(r.desiredFollowupEvals))
}
// Tests rescheduling failed service allocations with desired state stop
func TestReconciler_RescheduleNow_Service(t *testing.T) {
require := require.New(t)
2018-03-26 18:06:21 +00:00
// Set desired 5
job := mock.Job()
job.TaskGroups[0].Count = 5
tgName := job.TaskGroups[0].Name
now := time.Now()
2018-03-26 18:06:21 +00:00
// Set up reschedule policy and update stanza
job.TaskGroups[0].ReschedulePolicy = &structs.ReschedulePolicy{
Attempts: 1,
Interval: 24 * time.Hour,
Delay: 5 * time.Second,
DelayFunction: "",
MaxDelay: 1 * time.Hour,
Unlimited: false,
}
job.TaskGroups[0].Update = noCanaryUpdate
// Create 5 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 5; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
alloc.ClientStatus = structs.AllocClientStatusRunning
}
2018-03-26 18:06:21 +00:00
// Mark two as failed
allocs[0].ClientStatus = structs.AllocClientStatusFailed
2018-03-26 18:06:21 +00:00
2018-01-19 21:20:00 +00:00
// Mark one of them as already rescheduled once
allocs[0].RescheduleTracker = &structs.RescheduleTracker{Events: []*structs.RescheduleEvent{
2018-01-19 21:20:00 +00:00
{RescheduleTime: time.Now().Add(-1 * time.Hour).UTC().UnixNano(),
PrevAllocID: uuid.Generate(),
PrevNodeID: uuid.Generate(),
},
}}
allocs[1].TaskStates = map[string]*structs.TaskState{tgName: {State: "start",
StartedAt: now.Add(-1 * time.Hour),
FinishedAt: now.Add(-10 * time.Second)}}
allocs[1].ClientStatus = structs.AllocClientStatusFailed
2018-01-19 21:20:00 +00:00
2018-01-19 17:58:59 +00:00
// Mark one as desired state stop
allocs[4].DesiredStatus = structs.AllocDesiredStatusStop
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, nil, "")
2018-01-19 17:58:59 +00:00
r := reconciler.Compute()
// Verify that no follow up evals were created
evals := r.desiredFollowupEvals[tgName]
require.Nil(evals)
// Verify that one rescheduled alloc and one replacement for terminal alloc were placed
2018-01-19 17:58:59 +00:00
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
2018-01-19 21:20:00 +00:00
place: 2,
2018-01-19 17:58:59 +00:00
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
2018-01-19 21:20:00 +00:00
Place: 2,
Ignore: 3,
2018-01-19 17:58:59 +00:00
},
},
})
// Rescheduled allocs should have previous allocs
2018-03-26 18:06:21 +00:00
assertNamesHaveIndexes(t, intRange(1, 1, 4, 4), placeResultsToNames(r.place))
2018-01-19 21:20:00 +00:00
assertPlaceResultsHavePreviousAllocs(t, 1, r.place)
assertPlacementsAreRescheduled(t, 1, r.place)
2018-01-19 17:58:59 +00:00
}
// Tests rescheduling failed service allocations when there's clock drift (upto a second)
func TestReconciler_RescheduleNow_WithinAllowedTimeWindow(t *testing.T) {
require := require.New(t)
// Set desired 5
job := mock.Job()
job.TaskGroups[0].Count = 5
tgName := job.TaskGroups[0].Name
now := time.Now()
// Set up reschedule policy and update stanza
job.TaskGroups[0].ReschedulePolicy = &structs.ReschedulePolicy{
Attempts: 1,
Interval: 24 * time.Hour,
Delay: 5 * time.Second,
DelayFunction: "",
MaxDelay: 1 * time.Hour,
Unlimited: false,
}
job.TaskGroups[0].Update = noCanaryUpdate
// Create 5 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 5; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
alloc.ClientStatus = structs.AllocClientStatusRunning
}
// Mark one as failed
allocs[0].ClientStatus = structs.AllocClientStatusFailed
// Mark one of them as already rescheduled once
allocs[0].RescheduleTracker = &structs.RescheduleTracker{Events: []*structs.RescheduleEvent{
{RescheduleTime: time.Now().Add(-1 * time.Hour).UTC().UnixNano(),
PrevAllocID: uuid.Generate(),
PrevNodeID: uuid.Generate(),
},
}}
// Set fail time to 4 seconds ago which falls within the reschedule window
allocs[1].TaskStates = map[string]*structs.TaskState{tgName: {State: "start",
StartedAt: now.Add(-1 * time.Hour),
FinishedAt: now.Add(-4 * time.Second)}}
allocs[1].ClientStatus = structs.AllocClientStatusFailed
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, nil, "")
reconciler.now = now
r := reconciler.Compute()
// Verify that no follow up evals were created
evals := r.desiredFollowupEvals[tgName]
require.Nil(evals)
// Verify that one rescheduled alloc was placed
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 1,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 1,
Ignore: 4,
},
},
})
// Rescheduled allocs should have previous allocs
assertNamesHaveIndexes(t, intRange(1, 1), placeResultsToNames(r.place))
assertPlaceResultsHavePreviousAllocs(t, 1, r.place)
assertPlacementsAreRescheduled(t, 1, r.place)
}
// Tests rescheduling failed service allocations when the eval ID matches and there's a large clock drift
func TestReconciler_RescheduleNow_EvalIDMatch(t *testing.T) {
require := require.New(t)
// Set desired 5
job := mock.Job()
job.TaskGroups[0].Count = 5
tgName := job.TaskGroups[0].Name
now := time.Now()
// Set up reschedule policy and update stanza
job.TaskGroups[0].ReschedulePolicy = &structs.ReschedulePolicy{
Attempts: 1,
Interval: 24 * time.Hour,
Delay: 5 * time.Second,
DelayFunction: "",
MaxDelay: 1 * time.Hour,
Unlimited: false,
}
job.TaskGroups[0].Update = noCanaryUpdate
// Create 5 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 5; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
alloc.ClientStatus = structs.AllocClientStatusRunning
}
// Mark one as failed
allocs[0].ClientStatus = structs.AllocClientStatusFailed
// Mark one of them as already rescheduled once
allocs[0].RescheduleTracker = &structs.RescheduleTracker{Events: []*structs.RescheduleEvent{
{RescheduleTime: time.Now().Add(-1 * time.Hour).UTC().UnixNano(),
PrevAllocID: uuid.Generate(),
PrevNodeID: uuid.Generate(),
},
}}
// Set fail time to 5 seconds ago and eval ID
evalID := uuid.Generate()
allocs[1].TaskStates = map[string]*structs.TaskState{tgName: {State: "start",
StartedAt: now.Add(-1 * time.Hour),
FinishedAt: now.Add(-5 * time.Second)}}
allocs[1].ClientStatus = structs.AllocClientStatusFailed
allocs[1].FollowupEvalID = evalID
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, nil, evalID)
reconciler.now = now.Add(-30 * time.Second)
r := reconciler.Compute()
// Verify that no follow up evals were created
evals := r.desiredFollowupEvals[tgName]
require.Nil(evals)
// Verify that one rescheduled alloc was placed
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 1,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 1,
Ignore: 4,
},
},
})
// Rescheduled allocs should have previous allocs
assertNamesHaveIndexes(t, intRange(1, 1), placeResultsToNames(r.place))
assertPlaceResultsHavePreviousAllocs(t, 1, r.place)
assertPlacementsAreRescheduled(t, 1, r.place)
}
// Tests failed service allocations that were already rescheduled won't be rescheduled again
func TestReconciler_DontReschedule_PreviouslyRescheduled(t *testing.T) {
// Set desired 5
job := mock.Job()
job.TaskGroups[0].Count = 5
// Set up reschedule policy
job.TaskGroups[0].ReschedulePolicy = &structs.ReschedulePolicy{Attempts: 5, Interval: 24 * time.Hour}
// Create 7 existing allocations
var allocs []*structs.Allocation
for i := 0; i < 7; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
allocs = append(allocs, alloc)
alloc.ClientStatus = structs.AllocClientStatusRunning
}
// Mark two as failed and rescheduled
allocs[0].ClientStatus = structs.AllocClientStatusFailed
allocs[0].ID = allocs[1].ID
allocs[1].ClientStatus = structs.AllocClientStatusFailed
allocs[1].RescheduleTracker = &structs.RescheduleTracker{Events: []*structs.RescheduleEvent{
{RescheduleTime: time.Now().Add(-1 * time.Hour).UTC().UnixNano(),
PrevAllocID: uuid.Generate(),
PrevNodeID: uuid.Generate(),
},
}}
allocs[1].NextAllocation = allocs[2].ID
// Mark one as desired state stop
allocs[4].DesiredStatus = structs.AllocDesiredStatusStop
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, nil, "")
r := reconciler.Compute()
// Should place 1 - one is a new placement to make up the desired count of 5
// failing allocs are not rescheduled
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 1,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 1,
Ignore: 4,
},
},
})
// name index 0 is used for the replacement because its
assertNamesHaveIndexes(t, intRange(0, 0), placeResultsToNames(r.place))
}
2017-06-02 23:11:29 +00:00
// Tests the reconciler cancels an old deployment when the job is being stopped
func TestReconciler_CancelDeployment_JobStop(t *testing.T) {
job := mock.Job()
job.Stop = true
running := structs.NewDeployment(job)
failed := structs.NewDeployment(job)
failed.Status = structs.DeploymentStatusFailed
cases := []struct {
name string
job *structs.Job
jobID, taskGroup string
deployment *structs.Deployment
cancel bool
}{
{
name: "stopped job, running deployment",
job: job,
jobID: job.ID,
taskGroup: job.TaskGroups[0].Name,
deployment: running,
cancel: true,
},
{
name: "nil job, running deployment",
job: nil,
jobID: "foo",
taskGroup: "bar",
deployment: running,
cancel: true,
},
{
name: "stopped job, failed deployment",
job: job,
jobID: job.ID,
taskGroup: job.TaskGroups[0].Name,
deployment: failed,
cancel: false,
},
{
name: "nil job, failed deployment",
job: nil,
jobID: "foo",
taskGroup: "bar",
deployment: failed,
cancel: false,
},
}
for _, c := range cases {
t.Run(c.name, func(t *testing.T) {
// Create 10 allocations
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = c.job
alloc.JobID = c.jobID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(c.jobID, c.taskGroup, uint(i))
alloc.TaskGroup = c.taskGroup
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, c.jobID, c.job, c.deployment, allocs, nil, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
var updates []*structs.DeploymentStatusUpdate
if c.cancel {
updates = []*structs.DeploymentStatusUpdate{
{
DeploymentID: c.deployment.ID,
Status: structs.DeploymentStatusCancelled,
StatusDescription: structs.DeploymentStatusDescriptionStoppedJob,
},
}
}
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: updates,
place: 0,
inplace: 0,
stop: 10,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
c.taskGroup: {
Stop: 10,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 9), stopResultsToNames(r.stop))
})
}
}
// Tests the reconciler cancels an old deployment when the job is updated
func TestReconciler_CancelDeployment_JobUpdate(t *testing.T) {
// Create a base job
job := mock.Job()
// Create two deployments
running := structs.NewDeployment(job)
failed := structs.NewDeployment(job)
failed.Status = structs.DeploymentStatusFailed
// Make the job newer than the deployment
job.Version += 10
2017-06-02 23:11:29 +00:00
cases := []struct {
name string
deployment *structs.Deployment
cancel bool
}{
{
name: "running deployment",
deployment: running,
cancel: true,
},
{
name: "failed deployment",
deployment: failed,
cancel: false,
},
}
for _, c := range cases {
t.Run(c.name, func(t *testing.T) {
// Create 10 allocations
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, c.deployment, allocs, nil, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
var updates []*structs.DeploymentStatusUpdate
if c.cancel {
updates = []*structs.DeploymentStatusUpdate{
{
DeploymentID: c.deployment.ID,
Status: structs.DeploymentStatusCancelled,
StatusDescription: structs.DeploymentStatusDescriptionNewerJob,
},
}
}
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: updates,
place: 0,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Ignore: 10,
},
},
})
})
}
}
2017-06-06 21:08:46 +00:00
// Tests the reconciler creates a deployment and does a rolling upgrade with
// destructive changes
func TestReconciler_CreateDeployment_RollingUpgrade_Destructive(t *testing.T) {
2017-06-02 23:11:29 +00:00
job := mock.Job()
job.TaskGroups[0].Update = noCanaryUpdate
// Create 10 allocations from the old job
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnDestructive, false, job.ID, job, nil, allocs, nil, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
d := structs.NewDeployment(job)
d.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
DesiredTotal: 10,
}
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: d,
deploymentUpdates: nil,
2017-07-15 23:31:33 +00:00
destructive: 4,
2017-06-02 23:11:29 +00:00
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
DestructiveUpdate: 4,
Ignore: 6,
},
},
})
2017-07-15 23:31:33 +00:00
assertNamesHaveIndexes(t, intRange(0, 3), destructiveResultsToNames(r.destructiveUpdate))
2017-06-02 23:11:29 +00:00
}
2017-06-06 21:08:46 +00:00
// Tests the reconciler creates a deployment for inplace updates
func TestReconciler_CreateDeployment_RollingUpgrade_Inplace(t *testing.T) {
2018-03-23 23:55:21 +00:00
jobOld := mock.Job()
job := jobOld.Copy()
job.Version++
2017-06-06 21:08:46 +00:00
job.TaskGroups[0].Update = noCanaryUpdate
// Create 10 allocations from the old job
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
2018-03-23 23:55:21 +00:00
alloc.Job = jobOld
2017-06-06 21:08:46 +00:00
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-06 21:08:46 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnInplace, false, job.ID, job, nil, allocs, nil, "")
2017-06-06 21:08:46 +00:00
r := reconciler.Compute()
d := structs.NewDeployment(job)
d.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
DesiredTotal: 10,
}
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: d,
deploymentUpdates: nil,
place: 0,
inplace: 10,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
InPlaceUpdate: 10,
},
},
})
}
// Tests the reconciler doesn't creates a deployment if there are no changes
func TestReconciler_DontCreateDeployment_NoChanges(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = noCanaryUpdate
// Create 10 allocations from the job
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-06 21:08:46 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, nil, allocs, nil, "")
2017-06-06 21:08:46 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 0,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
DestructiveUpdate: 0,
Ignore: 10,
},
},
})
}
2017-06-02 23:11:29 +00:00
// Tests the reconciler doesn't place any more canaries when the deployment is
// paused or failed
func TestReconciler_PausedOrFailedDeployment_NoMoreCanaries(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = canaryUpdate
cases := []struct {
name string
deploymentStatus string
2017-07-05 19:50:40 +00:00
stop uint64
2017-06-02 23:11:29 +00:00
}{
{
name: "paused deployment",
deploymentStatus: structs.DeploymentStatusPaused,
2017-07-05 19:50:40 +00:00
stop: 0,
2017-06-02 23:11:29 +00:00
},
{
name: "failed deployment",
deploymentStatus: structs.DeploymentStatusFailed,
2017-07-05 19:50:40 +00:00
stop: 1,
2017-06-02 23:11:29 +00:00
},
}
for _, c := range cases {
t.Run(c.name, func(t *testing.T) {
2017-07-05 19:50:40 +00:00
// Create a deployment that is paused/failed and has placed some canaries
2017-06-02 23:11:29 +00:00
d := structs.NewDeployment(job)
d.Status = c.deploymentStatus
d.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
Promoted: false,
DesiredCanaries: 2,
DesiredTotal: 10,
PlacedAllocs: 1,
}
// Create 10 allocations for the original job
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
// Create one canary
canary := mock.Alloc()
canary.Job = job
canary.JobID = job.ID
canary.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
canary.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, 0)
canary.TaskGroup = job.TaskGroups[0].Name
canary.DeploymentID = d.ID
allocs = append(allocs, canary)
2017-07-05 19:50:40 +00:00
d.TaskGroups[canary.TaskGroup].PlacedCanaries = []string{canary.ID}
2017-06-02 23:11:29 +00:00
mockUpdateFn := allocUpdateFnMock(map[string]allocUpdateType{canary.ID: allocUpdateFnIgnore}, allocUpdateFnDestructive)
reconciler := NewAllocReconciler(testLogger(), mockUpdateFn, false, job.ID, job, d, allocs, nil, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 0,
inplace: 0,
2017-07-05 19:50:40 +00:00
stop: int(c.stop),
2017-06-02 23:11:29 +00:00
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
2017-07-05 19:50:40 +00:00
Ignore: 11 - c.stop,
Stop: c.stop,
2017-06-02 23:11:29 +00:00
},
},
})
})
}
}
// Tests the reconciler doesn't place any more allocs when the deployment is
// paused or failed
func TestReconciler_PausedOrFailedDeployment_NoMorePlacements(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = noCanaryUpdate
job.TaskGroups[0].Count = 15
cases := []struct {
name string
deploymentStatus string
}{
{
name: "paused deployment",
deploymentStatus: structs.DeploymentStatusPaused,
},
{
name: "failed deployment",
deploymentStatus: structs.DeploymentStatusFailed,
},
}
for _, c := range cases {
t.Run(c.name, func(t *testing.T) {
// Create a deployment that is paused and has placed some canaries
d := structs.NewDeployment(job)
d.Status = c.deploymentStatus
d.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
Promoted: false,
DesiredTotal: 15,
PlacedAllocs: 10,
}
// Create 10 allocations for the new job
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, d, allocs, nil, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 0,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Ignore: 10,
},
},
})
})
}
}
// Tests the reconciler doesn't do any more destructive updates when the
// deployment is paused or failed
func TestReconciler_PausedOrFailedDeployment_NoMoreDestructiveUpdates(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = noCanaryUpdate
cases := []struct {
name string
deploymentStatus string
}{
{
name: "paused deployment",
deploymentStatus: structs.DeploymentStatusPaused,
},
{
name: "failed deployment",
deploymentStatus: structs.DeploymentStatusFailed,
},
}
for _, c := range cases {
t.Run(c.name, func(t *testing.T) {
// Create a deployment that is paused and has placed some canaries
d := structs.NewDeployment(job)
d.Status = c.deploymentStatus
d.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
Promoted: false,
DesiredTotal: 10,
PlacedAllocs: 1,
}
// Create 9 allocations for the original job
var allocs []*structs.Allocation
for i := 1; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
// Create one for the new job
newAlloc := mock.Alloc()
newAlloc.Job = job
newAlloc.JobID = job.ID
newAlloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
newAlloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, 0)
newAlloc.TaskGroup = job.TaskGroups[0].Name
newAlloc.DeploymentID = d.ID
allocs = append(allocs, newAlloc)
mockUpdateFn := allocUpdateFnMock(map[string]allocUpdateType{newAlloc.ID: allocUpdateFnIgnore}, allocUpdateFnDestructive)
reconciler := NewAllocReconciler(testLogger(), mockUpdateFn, false, job.ID, job, d, allocs, nil, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 0,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Ignore: 10,
},
},
})
})
}
}
// Tests the reconciler handles migrations correctly when a deployment is paused
// or failed
func TestReconciler_PausedOrFailedDeployment_Migrations(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = noCanaryUpdate
cases := []struct {
name string
deploymentStatus string
place int
stop int
ignoreAnnotation uint64
migrateAnnotation uint64
stopAnnotation uint64
}{
{
name: "paused deployment",
deploymentStatus: structs.DeploymentStatusPaused,
place: 0,
stop: 3,
ignoreAnnotation: 5,
stopAnnotation: 3,
2017-06-02 23:11:29 +00:00
},
{
name: "failed deployment",
deploymentStatus: structs.DeploymentStatusFailed,
place: 0,
stop: 3,
ignoreAnnotation: 5,
migrateAnnotation: 0,
stopAnnotation: 3,
},
}
for _, c := range cases {
t.Run(c.name, func(t *testing.T) {
// Create a deployment that is paused and has placed some canaries
d := structs.NewDeployment(job)
d.Status = c.deploymentStatus
d.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
Promoted: false,
DesiredTotal: 10,
PlacedAllocs: 8,
}
// Create 8 allocations in the deployment
var allocs []*structs.Allocation
for i := 0; i < 8; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
alloc.DeploymentID = d.ID
allocs = append(allocs, alloc)
}
// Build a map of tainted nodes
tainted := make(map[string]*structs.Node, 3)
for i := 0; i < 3; i++ {
n := mock.Node()
n.ID = allocs[i].NodeID
allocs[i].DesiredTransition.Migrate = helper.BoolToPtr(true)
2017-06-02 23:11:29 +00:00
n.Drain = true
tainted[n.ID] = n
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, d, allocs, tainted, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: c.place,
inplace: 0,
stop: c.stop,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Migrate: c.migrateAnnotation,
Ignore: c.ignoreAnnotation,
Stop: c.stopAnnotation,
},
},
})
})
}
}
// Tests the reconciler handles migrating a canary correctly on a draining node
func TestReconciler_DrainNode_Canary(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = canaryUpdate
// Create a deployment that is paused and has placed some canaries
d := structs.NewDeployment(job)
2017-07-05 19:50:40 +00:00
s := &structs.DeploymentState{
2017-06-02 23:11:29 +00:00
Promoted: false,
DesiredTotal: 10,
DesiredCanaries: 2,
PlacedAllocs: 2,
}
2017-07-05 19:50:40 +00:00
d.TaskGroups[job.TaskGroups[0].Name] = s
2017-06-02 23:11:29 +00:00
// Create 10 allocations from the old job
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
// Create two canaries for the new job
handled := make(map[string]allocUpdateType)
for i := 0; i < 2; i++ {
// Create one canary
canary := mock.Alloc()
canary.Job = job
canary.JobID = job.ID
canary.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
canary.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
canary.TaskGroup = job.TaskGroups[0].Name
canary.DeploymentID = d.ID
2017-07-05 19:50:40 +00:00
s.PlacedCanaries = append(s.PlacedCanaries, canary.ID)
2017-06-02 23:11:29 +00:00
allocs = append(allocs, canary)
handled[canary.ID] = allocUpdateFnIgnore
}
// Build a map of tainted nodes that contains the last canary
tainted := make(map[string]*structs.Node, 1)
n := mock.Node()
n.ID = allocs[11].NodeID
allocs[11].DesiredTransition.Migrate = helper.BoolToPtr(true)
2017-06-02 23:11:29 +00:00
n.Drain = true
tainted[n.ID] = n
mockUpdateFn := allocUpdateFnMock(handled, allocUpdateFnDestructive)
reconciler := NewAllocReconciler(testLogger(), mockUpdateFn, false, job.ID, job, d, allocs, tainted, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 1,
inplace: 0,
stop: 1,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Canary: 1,
Ignore: 11,
},
},
})
assertNamesHaveIndexes(t, intRange(1, 1), stopResultsToNames(r.stop))
assertNamesHaveIndexes(t, intRange(1, 1), placeResultsToNames(r.place))
}
// Tests the reconciler handles migrating a canary correctly on a lost node
func TestReconciler_LostNode_Canary(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = canaryUpdate
// Create a deployment that is paused and has placed some canaries
d := structs.NewDeployment(job)
2017-07-05 19:50:40 +00:00
s := &structs.DeploymentState{
2017-06-02 23:11:29 +00:00
Promoted: false,
DesiredTotal: 10,
DesiredCanaries: 2,
PlacedAllocs: 2,
}
2017-07-05 19:50:40 +00:00
d.TaskGroups[job.TaskGroups[0].Name] = s
2017-06-02 23:11:29 +00:00
// Create 10 allocations from the old job
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
// Create two canaries for the new job
handled := make(map[string]allocUpdateType)
for i := 0; i < 2; i++ {
// Create one canary
canary := mock.Alloc()
canary.Job = job
canary.JobID = job.ID
canary.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
canary.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
canary.TaskGroup = job.TaskGroups[0].Name
2017-07-05 19:50:40 +00:00
s.PlacedCanaries = append(s.PlacedCanaries, canary.ID)
2017-06-02 23:11:29 +00:00
canary.DeploymentID = d.ID
allocs = append(allocs, canary)
handled[canary.ID] = allocUpdateFnIgnore
}
// Build a map of tainted nodes that contains the last canary
tainted := make(map[string]*structs.Node, 1)
n := mock.Node()
n.ID = allocs[11].NodeID
n.Status = structs.NodeStatusDown
tainted[n.ID] = n
mockUpdateFn := allocUpdateFnMock(handled, allocUpdateFnDestructive)
reconciler := NewAllocReconciler(testLogger(), mockUpdateFn, false, job.ID, job, d, allocs, tainted, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 1,
inplace: 0,
stop: 1,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Canary: 1,
Ignore: 11,
},
},
})
assertNamesHaveIndexes(t, intRange(1, 1), stopResultsToNames(r.stop))
assertNamesHaveIndexes(t, intRange(1, 1), placeResultsToNames(r.place))
}
// Tests the reconciler handles stopping canaries from older deployments
func TestReconciler_StopOldCanaries(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = canaryUpdate
// Create an old deployment that has placed some canaries
d := structs.NewDeployment(job)
2017-07-05 19:50:40 +00:00
s := &structs.DeploymentState{
2017-06-02 23:11:29 +00:00
Promoted: false,
DesiredTotal: 10,
DesiredCanaries: 2,
PlacedAllocs: 2,
}
2017-07-05 19:50:40 +00:00
d.TaskGroups[job.TaskGroups[0].Name] = s
2017-06-02 23:11:29 +00:00
// Update the job
job.Version += 10
2017-06-02 23:11:29 +00:00
// Create 10 allocations from the old job
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
// Create canaries
for i := 0; i < 2; i++ {
// Create one canary
canary := mock.Alloc()
canary.Job = job
canary.JobID = job.ID
canary.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
canary.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
canary.TaskGroup = job.TaskGroups[0].Name
2017-07-05 19:50:40 +00:00
s.PlacedCanaries = append(s.PlacedCanaries, canary.ID)
2017-06-02 23:11:29 +00:00
canary.DeploymentID = d.ID
allocs = append(allocs, canary)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnDestructive, false, job.ID, job, d, allocs, nil, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
newD := structs.NewDeployment(job)
newD.StatusDescription = structs.DeploymentStatusDescriptionRunningNeedsPromotion
2017-06-02 23:11:29 +00:00
newD.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
DesiredCanaries: 2,
DesiredTotal: 10,
}
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: newD,
deploymentUpdates: []*structs.DeploymentStatusUpdate{
{
DeploymentID: d.ID,
Status: structs.DeploymentStatusCancelled,
StatusDescription: structs.DeploymentStatusDescriptionNewerJob,
},
},
place: 2,
inplace: 0,
stop: 2,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Canary: 2,
Stop: 2,
Ignore: 10,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 1), stopResultsToNames(r.stop))
assertNamesHaveIndexes(t, intRange(0, 1), placeResultsToNames(r.place))
}
// Tests the reconciler creates new canaries when the job changes
func TestReconciler_NewCanaries(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = canaryUpdate
// Create 10 allocations from the old job
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnDestructive, false, job.ID, job, nil, allocs, nil, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
newD := structs.NewDeployment(job)
newD.StatusDescription = structs.DeploymentStatusDescriptionRunningNeedsPromotion
2017-06-02 23:11:29 +00:00
newD.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
DesiredCanaries: 2,
DesiredTotal: 10,
}
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: newD,
deploymentUpdates: nil,
place: 2,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Canary: 2,
Ignore: 10,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 1), placeResultsToNames(r.place))
}
// Tests the reconciler creates new canaries when the job changes for multiple
// task groups
func TestReconciler_NewCanaries_MultiTG(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = canaryUpdate
job.TaskGroups = append(job.TaskGroups, job.TaskGroups[0].Copy())
job.TaskGroups[0].Name = "tg2"
// Create 10 allocations from the old job for each tg
var allocs []*structs.Allocation
for j := 0; j < 2; j++ {
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[j].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[j].Name
allocs = append(allocs, alloc)
}
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnDestructive, false, job.ID, job, nil, allocs, nil, "")
r := reconciler.Compute()
newD := structs.NewDeployment(job)
newD.StatusDescription = structs.DeploymentStatusDescriptionRunningNeedsPromotion
state := &structs.DeploymentState{
DesiredCanaries: 2,
DesiredTotal: 10,
}
newD.TaskGroups[job.TaskGroups[0].Name] = state
newD.TaskGroups[job.TaskGroups[1].Name] = state.Copy()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: newD,
deploymentUpdates: nil,
place: 4,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Canary: 2,
Ignore: 10,
},
job.TaskGroups[1].Name: {
Canary: 2,
Ignore: 10,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 1, 0, 1), placeResultsToNames(r.place))
}
2017-06-02 23:11:29 +00:00
// Tests the reconciler creates new canaries when the job changes and scales up
func TestReconciler_NewCanaries_ScaleUp(t *testing.T) {
// Scale the job up to 15
job := mock.Job()
job.TaskGroups[0].Update = canaryUpdate
job.TaskGroups[0].Count = 15
// Create 10 allocations from the old job
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnDestructive, false, job.ID, job, nil, allocs, nil, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
newD := structs.NewDeployment(job)
newD.StatusDescription = structs.DeploymentStatusDescriptionRunningNeedsPromotion
2017-06-02 23:11:29 +00:00
newD.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
DesiredCanaries: 2,
DesiredTotal: 15,
}
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: newD,
deploymentUpdates: nil,
place: 2,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Canary: 2,
Ignore: 10,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 1), placeResultsToNames(r.place))
}
// Tests the reconciler creates new canaries when the job changes and scales
// down
func TestReconciler_NewCanaries_ScaleDown(t *testing.T) {
// Scale the job down to 5
job := mock.Job()
job.TaskGroups[0].Update = canaryUpdate
job.TaskGroups[0].Count = 5
// Create 10 allocations from the old job
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnDestructive, false, job.ID, job, nil, allocs, nil, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
newD := structs.NewDeployment(job)
newD.StatusDescription = structs.DeploymentStatusDescriptionRunningNeedsPromotion
2017-06-02 23:11:29 +00:00
newD.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
DesiredCanaries: 2,
DesiredTotal: 5,
}
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: newD,
deploymentUpdates: nil,
place: 2,
inplace: 0,
stop: 5,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Canary: 2,
Stop: 5,
Ignore: 5,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 1), placeResultsToNames(r.place))
assertNamesHaveIndexes(t, intRange(5, 9), stopResultsToNames(r.stop))
}
// Tests the reconciler handles filling the names of partially placed canaries
func TestReconciler_NewCanaries_FillNames(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = &structs.UpdateStrategy{
Canary: 4,
MaxParallel: 2,
HealthCheck: structs.UpdateStrategyHealthCheck_Checks,
MinHealthyTime: 10 * time.Second,
HealthyDeadline: 10 * time.Minute,
}
// Create an existing deployment that has placed some canaries
d := structs.NewDeployment(job)
2017-07-05 19:50:40 +00:00
s := &structs.DeploymentState{
2017-06-02 23:11:29 +00:00
Promoted: false,
DesiredTotal: 10,
DesiredCanaries: 4,
PlacedAllocs: 2,
}
2017-07-05 19:50:40 +00:00
d.TaskGroups[job.TaskGroups[0].Name] = s
2017-06-02 23:11:29 +00:00
// Create 10 allocations from the old job
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
// Create canaries but pick names at the ends
for i := 0; i < 4; i += 3 {
// Create one canary
canary := mock.Alloc()
canary.Job = job
canary.JobID = job.ID
canary.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
canary.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
canary.TaskGroup = job.TaskGroups[0].Name
2017-07-05 19:50:40 +00:00
s.PlacedCanaries = append(s.PlacedCanaries, canary.ID)
2017-06-02 23:11:29 +00:00
canary.DeploymentID = d.ID
allocs = append(allocs, canary)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnDestructive, false, job.ID, job, d, allocs, nil, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 2,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Canary: 2,
Ignore: 12,
},
},
})
assertNamesHaveIndexes(t, intRange(1, 2), placeResultsToNames(r.place))
}
// Tests the reconciler handles canary promotion by unblocking max_parallel
func TestReconciler_PromoteCanaries_Unblock(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = canaryUpdate
// Create an existing deployment that has placed some canaries and mark them
// promoted
d := structs.NewDeployment(job)
2017-07-05 19:50:40 +00:00
s := &structs.DeploymentState{
2017-06-02 23:11:29 +00:00
Promoted: true,
DesiredTotal: 10,
DesiredCanaries: 2,
PlacedAllocs: 2,
}
2017-07-05 19:50:40 +00:00
d.TaskGroups[job.TaskGroups[0].Name] = s
2017-06-02 23:11:29 +00:00
// Create 10 allocations from the old job
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
// Create the canaries
handled := make(map[string]allocUpdateType)
for i := 0; i < 2; i++ {
// Create one canary
canary := mock.Alloc()
canary.Job = job
canary.JobID = job.ID
canary.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
canary.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
canary.TaskGroup = job.TaskGroups[0].Name
2017-07-05 19:50:40 +00:00
s.PlacedCanaries = append(s.PlacedCanaries, canary.ID)
2017-06-02 23:11:29 +00:00
canary.DeploymentID = d.ID
canary.DeploymentStatus = &structs.AllocDeploymentStatus{
Healthy: helper.BoolToPtr(true),
}
allocs = append(allocs, canary)
handled[canary.ID] = allocUpdateFnIgnore
}
mockUpdateFn := allocUpdateFnMock(handled, allocUpdateFnDestructive)
reconciler := NewAllocReconciler(testLogger(), mockUpdateFn, false, job.ID, job, d, allocs, nil, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
2017-07-15 23:31:33 +00:00
destructive: 2,
stop: 2,
2017-06-02 23:11:29 +00:00
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Stop: 2,
DestructiveUpdate: 2,
Ignore: 8,
},
},
})
2017-07-06 16:55:39 +00:00
assertNoCanariesStopped(t, d, r.stop)
2017-07-15 23:31:33 +00:00
assertNamesHaveIndexes(t, intRange(2, 3), destructiveResultsToNames(r.destructiveUpdate))
assertNamesHaveIndexes(t, intRange(0, 1), stopResultsToNames(r.stop))
2017-06-02 23:11:29 +00:00
}
// Tests the reconciler handles canary promotion when the canary count equals
// the total correctly
func TestReconciler_PromoteCanaries_CanariesEqualCount(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = canaryUpdate
job.TaskGroups[0].Count = 2
// Create an existing deployment that has placed some canaries and mark them
// promoted
d := structs.NewDeployment(job)
2017-07-05 19:50:40 +00:00
s := &structs.DeploymentState{
2017-06-02 23:11:29 +00:00
Promoted: true,
DesiredTotal: 2,
DesiredCanaries: 2,
PlacedAllocs: 2,
}
2017-07-05 19:50:40 +00:00
d.TaskGroups[job.TaskGroups[0].Name] = s
2017-06-02 23:11:29 +00:00
// Create 2 allocations from the old job
var allocs []*structs.Allocation
for i := 0; i < 2; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
// Create the canaries
handled := make(map[string]allocUpdateType)
for i := 0; i < 2; i++ {
// Create one canary
canary := mock.Alloc()
canary.Job = job
canary.JobID = job.ID
canary.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
canary.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
canary.TaskGroup = job.TaskGroups[0].Name
2017-07-05 19:50:40 +00:00
s.PlacedCanaries = append(s.PlacedCanaries, canary.ID)
2017-06-02 23:11:29 +00:00
canary.DeploymentID = d.ID
canary.DeploymentStatus = &structs.AllocDeploymentStatus{
Healthy: helper.BoolToPtr(true),
}
allocs = append(allocs, canary)
handled[canary.ID] = allocUpdateFnIgnore
}
mockUpdateFn := allocUpdateFnMock(handled, allocUpdateFnDestructive)
reconciler := NewAllocReconciler(testLogger(), mockUpdateFn, false, job.ID, job, d, allocs, nil, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
2017-07-06 15:39:16 +00:00
updates := []*structs.DeploymentStatusUpdate{
{
DeploymentID: d.ID,
Status: structs.DeploymentStatusSuccessful,
StatusDescription: structs.DeploymentStatusDescriptionSuccessful,
},
}
2017-06-02 23:11:29 +00:00
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
2017-07-06 15:39:16 +00:00
deploymentUpdates: updates,
2017-06-02 23:11:29 +00:00
place: 0,
inplace: 0,
stop: 2,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Stop: 2,
Ignore: 2,
},
},
})
2017-07-06 16:55:39 +00:00
assertNoCanariesStopped(t, d, r.stop)
2017-06-02 23:11:29 +00:00
assertNamesHaveIndexes(t, intRange(0, 1), stopResultsToNames(r.stop))
}
// Tests the reconciler checks the health of placed allocs to determine the
// limit
func TestReconciler_DeploymentLimit_HealthAccounting(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = noCanaryUpdate
cases := []struct {
healthy int
}{
{
healthy: 0,
},
{
healthy: 1,
},
{
healthy: 2,
},
{
healthy: 3,
},
{
healthy: 4,
},
}
for _, c := range cases {
t.Run(fmt.Sprintf("%d healthy", c.healthy), func(t *testing.T) {
// Create an existing deployment that has placed some canaries and mark them
// promoted
d := structs.NewDeployment(job)
d.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
Promoted: true,
DesiredTotal: 10,
PlacedAllocs: 4,
}
// Create 6 allocations from the old job
var allocs []*structs.Allocation
for i := 4; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
// Create the new allocs
handled := make(map[string]allocUpdateType)
for i := 0; i < 4; i++ {
new := mock.Alloc()
new.Job = job
new.JobID = job.ID
new.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
new.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
new.TaskGroup = job.TaskGroups[0].Name
new.DeploymentID = d.ID
if i < c.healthy {
new.DeploymentStatus = &structs.AllocDeploymentStatus{
Healthy: helper.BoolToPtr(true),
}
}
allocs = append(allocs, new)
handled[new.ID] = allocUpdateFnIgnore
}
mockUpdateFn := allocUpdateFnMock(handled, allocUpdateFnDestructive)
reconciler := NewAllocReconciler(testLogger(), mockUpdateFn, false, job.ID, job, d, allocs, nil, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
2017-07-15 23:31:33 +00:00
destructive: c.healthy,
2017-06-02 23:11:29 +00:00
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
DestructiveUpdate: uint64(c.healthy),
Ignore: uint64(10 - c.healthy),
},
},
})
if c.healthy != 0 {
2017-07-15 23:31:33 +00:00
assertNamesHaveIndexes(t, intRange(4, 3+c.healthy), destructiveResultsToNames(r.destructiveUpdate))
2017-06-02 23:11:29 +00:00
}
})
}
}
// Tests the reconciler handles an alloc on a tainted node during a rolling
// update
func TestReconciler_TaintedNode_RollingUpgrade(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = noCanaryUpdate
// Create an existing deployment that has some placed allocs
d := structs.NewDeployment(job)
d.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
Promoted: true,
DesiredTotal: 10,
2017-07-15 23:31:33 +00:00
PlacedAllocs: 7,
2017-06-02 23:11:29 +00:00
}
// Create 2 allocations from the old job
2017-06-02 23:11:29 +00:00
var allocs []*structs.Allocation
for i := 8; i < 10; i++ {
2017-06-02 23:11:29 +00:00
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
// Create the healthy replacements
handled := make(map[string]allocUpdateType)
for i := 0; i < 8; i++ {
2017-06-02 23:11:29 +00:00
new := mock.Alloc()
new.Job = job
new.JobID = job.ID
new.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
new.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
new.TaskGroup = job.TaskGroups[0].Name
new.DeploymentID = d.ID
new.DeploymentStatus = &structs.AllocDeploymentStatus{
Healthy: helper.BoolToPtr(true),
}
allocs = append(allocs, new)
handled[new.ID] = allocUpdateFnIgnore
}
// Build a map of tainted nodes
tainted := make(map[string]*structs.Node, 3)
for i := 0; i < 3; i++ {
2017-06-02 23:11:29 +00:00
n := mock.Node()
n.ID = allocs[2+i].NodeID
2017-06-02 23:11:29 +00:00
if i == 0 {
n.Status = structs.NodeStatusDown
} else {
n.Drain = true
allocs[2+i].DesiredTransition.Migrate = helper.BoolToPtr(true)
2017-06-02 23:11:29 +00:00
}
tainted[n.ID] = n
}
mockUpdateFn := allocUpdateFnMock(handled, allocUpdateFnDestructive)
reconciler := NewAllocReconciler(testLogger(), mockUpdateFn, false, job.ID, job, d, allocs, tainted, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
2018-03-12 20:44:33 +00:00
place: 3,
destructive: 2,
2018-03-12 20:44:33 +00:00
stop: 3,
2017-06-02 23:11:29 +00:00
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 1, // Place the lost
Stop: 1, // Stop the lost
2018-03-12 20:44:33 +00:00
Migrate: 2, // Migrate the tainted
DestructiveUpdate: 2,
2018-03-12 20:44:33 +00:00
Ignore: 5,
2017-06-02 23:11:29 +00:00
},
},
})
assertNamesHaveIndexes(t, intRange(8, 9), destructiveResultsToNames(r.destructiveUpdate))
2018-03-12 20:44:33 +00:00
assertNamesHaveIndexes(t, intRange(0, 2), placeResultsToNames(r.place))
assertNamesHaveIndexes(t, intRange(0, 2), stopResultsToNames(r.stop))
2017-06-02 23:11:29 +00:00
}
// Tests the reconciler handles a failed deployment and only replaces lost
// deployments
func TestReconciler_FailedDeployment_PlacementLost(t *testing.T) {
2017-06-02 23:11:29 +00:00
job := mock.Job()
job.TaskGroups[0].Update = noCanaryUpdate
// Create an existing failed deployment that has some placed allocs
d := structs.NewDeployment(job)
d.Status = structs.DeploymentStatusFailed
d.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
Promoted: true,
DesiredTotal: 10,
PlacedAllocs: 4,
}
// Create 6 allocations from the old job
var allocs []*structs.Allocation
for i := 4; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
// Create the healthy replacements
handled := make(map[string]allocUpdateType)
for i := 0; i < 4; i++ {
new := mock.Alloc()
new.Job = job
new.JobID = job.ID
new.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
new.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
new.TaskGroup = job.TaskGroups[0].Name
new.DeploymentID = d.ID
new.DeploymentStatus = &structs.AllocDeploymentStatus{
Healthy: helper.BoolToPtr(true),
}
allocs = append(allocs, new)
handled[new.ID] = allocUpdateFnIgnore
}
// Build a map of tainted nodes
tainted := make(map[string]*structs.Node, 2)
for i := 0; i < 2; i++ {
n := mock.Node()
n.ID = allocs[6+i].NodeID
if i == 0 {
n.Status = structs.NodeStatusDown
} else {
n.Drain = true
allocs[6+i].DesiredTransition.Migrate = helper.BoolToPtr(true)
2017-06-02 23:11:29 +00:00
}
tainted[n.ID] = n
}
mockUpdateFn := allocUpdateFnMock(handled, allocUpdateFnDestructive)
reconciler := NewAllocReconciler(testLogger(), mockUpdateFn, false, job.ID, job, d, allocs, tainted, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 1, // Only replace the lost node
2017-06-02 23:11:29 +00:00
inplace: 0,
stop: 2,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 1,
2017-06-02 23:11:29 +00:00
Stop: 2,
Ignore: 8,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 0), placeResultsToNames(r.place))
2017-06-02 23:11:29 +00:00
assertNamesHaveIndexes(t, intRange(0, 1), stopResultsToNames(r.stop))
}
// Tests the reconciler handles a run after a deployment is complete
// successfully.
func TestReconciler_CompleteDeployment(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = canaryUpdate
2017-07-05 19:50:40 +00:00
d := structs.NewDeployment(job)
d.Status = structs.DeploymentStatusSuccessful
d.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
Promoted: true,
DesiredTotal: 10,
DesiredCanaries: 2,
PlacedAllocs: 10,
HealthyAllocs: 10,
}
2017-06-02 23:11:29 +00:00
// Create allocations from the old job
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-06-02 23:11:29 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
2017-07-05 19:50:40 +00:00
alloc.DeploymentID = d.ID
alloc.DeploymentStatus = &structs.AllocDeploymentStatus{
Healthy: helper.BoolToPtr(true),
2017-06-02 23:11:29 +00:00
}
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, d, allocs, nil, "")
2017-06-02 23:11:29 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 0,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Ignore: 10,
},
},
})
}
2017-07-05 19:50:40 +00:00
// Test that a failed deployment cancels non-promoted canaries
func TestReconciler_FailedDeployment_CancelCanaries(t *testing.T) {
// Create a job with two task groups
job := mock.Job()
job.TaskGroups[0].Update = canaryUpdate
job.TaskGroups = append(job.TaskGroups, job.TaskGroups[0].Copy())
job.TaskGroups[1].Name = "two"
// Create an existing failed deployment that has promoted one task group
d := structs.NewDeployment(job)
d.Status = structs.DeploymentStatusFailed
s0 := &structs.DeploymentState{
Promoted: true,
DesiredTotal: 10,
DesiredCanaries: 2,
PlacedAllocs: 4,
}
s1 := &structs.DeploymentState{
Promoted: false,
DesiredTotal: 10,
DesiredCanaries: 2,
PlacedAllocs: 2,
}
d.TaskGroups[job.TaskGroups[0].Name] = s0
d.TaskGroups[job.TaskGroups[1].Name] = s1
// Create 6 allocations from the old job
var allocs []*structs.Allocation
handled := make(map[string]allocUpdateType)
for _, group := range []int{0, 1} {
replacements := 4
state := s0
if group == 1 {
replacements = 2
state = s1
}
// Create the healthy replacements
for i := 0; i < replacements; i++ {
new := mock.Alloc()
new.Job = job
new.JobID = job.ID
new.NodeID = uuid.Generate()
2017-07-05 19:50:40 +00:00
new.Name = structs.AllocName(job.ID, job.TaskGroups[group].Name, uint(i))
new.TaskGroup = job.TaskGroups[group].Name
new.DeploymentID = d.ID
new.DeploymentStatus = &structs.AllocDeploymentStatus{
Healthy: helper.BoolToPtr(true),
}
allocs = append(allocs, new)
handled[new.ID] = allocUpdateFnIgnore
// Add the alloc to the canary list
if i < 2 {
state.PlacedCanaries = append(state.PlacedCanaries, new.ID)
}
}
for i := replacements; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-07-05 19:50:40 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[group].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[group].Name
allocs = append(allocs, alloc)
}
}
mockUpdateFn := allocUpdateFnMock(handled, allocUpdateFnDestructive)
reconciler := NewAllocReconciler(testLogger(), mockUpdateFn, false, job.ID, job, d, allocs, nil, "")
2017-07-05 19:50:40 +00:00
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 0,
inplace: 0,
stop: 2,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Ignore: 10,
},
job.TaskGroups[1].Name: {
Stop: 2,
Ignore: 8,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 1), stopResultsToNames(r.stop))
}
2017-07-05 19:55:51 +00:00
2017-07-06 02:46:57 +00:00
// Test that a failed deployment and updated job works
func TestReconciler_FailedDeployment_NewJob(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = noCanaryUpdate
// Create an existing failed deployment that has some placed allocs
d := structs.NewDeployment(job)
d.Status = structs.DeploymentStatusFailed
d.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
Promoted: true,
DesiredTotal: 10,
PlacedAllocs: 4,
}
// Create 6 allocations from the old job
var allocs []*structs.Allocation
for i := 4; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-07-06 02:46:57 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
// Create the healthy replacements
for i := 0; i < 4; i++ {
new := mock.Alloc()
new.Job = job
new.JobID = job.ID
new.NodeID = uuid.Generate()
2017-07-06 02:46:57 +00:00
new.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
new.TaskGroup = job.TaskGroups[0].Name
new.DeploymentID = d.ID
new.DeploymentStatus = &structs.AllocDeploymentStatus{
Healthy: helper.BoolToPtr(true),
}
allocs = append(allocs, new)
}
// Up the job version
jobNew := job.Copy()
jobNew.Version += 100
2017-07-06 02:46:57 +00:00
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnDestructive, false, job.ID, jobNew, d, allocs, nil, "")
2017-07-06 02:46:57 +00:00
r := reconciler.Compute()
dnew := structs.NewDeployment(jobNew)
2017-07-06 02:46:57 +00:00
dnew.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
DesiredTotal: 10,
}
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: dnew,
deploymentUpdates: nil,
2017-07-15 23:31:33 +00:00
destructive: 4,
2017-07-06 02:46:57 +00:00
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
DestructiveUpdate: 4,
Ignore: 6,
},
},
})
2017-07-15 23:31:33 +00:00
assertNamesHaveIndexes(t, intRange(0, 3), destructiveResultsToNames(r.destructiveUpdate))
2017-07-06 02:46:57 +00:00
}
2017-07-06 15:39:16 +00:00
// Tests the reconciler marks a deployment as complete
func TestReconciler_MarkDeploymentComplete(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = noCanaryUpdate
d := structs.NewDeployment(job)
d.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
Promoted: true,
DesiredTotal: 10,
PlacedAllocs: 10,
HealthyAllocs: 10,
}
// Create allocations from the old job
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
2017-07-06 15:39:16 +00:00
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
alloc.DeploymentID = d.ID
alloc.DeploymentStatus = &structs.AllocDeploymentStatus{
Healthy: helper.BoolToPtr(true),
}
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, job, d, allocs, nil, "")
2017-07-06 15:39:16 +00:00
r := reconciler.Compute()
updates := []*structs.DeploymentStatusUpdate{
{
DeploymentID: d.ID,
Status: structs.DeploymentStatusSuccessful,
StatusDescription: structs.DeploymentStatusDescriptionSuccessful,
},
}
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: updates,
place: 0,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Ignore: 10,
},
},
})
}
// Tests the reconciler handles changing a job such that a deployment is created
// while doing a scale up but as the second eval.
func TestReconciler_JobChange_ScaleUp_SecondEval(t *testing.T) {
// Scale the job up to 15
job := mock.Job()
job.TaskGroups[0].Update = noCanaryUpdate
job.TaskGroups[0].Count = 30
// Create a deployment that is paused and has placed some canaries
d := structs.NewDeployment(job)
d.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
Promoted: false,
DesiredTotal: 30,
PlacedAllocs: 20,
}
// Create 10 allocations from the old job
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
// Create 20 from new job
handled := make(map[string]allocUpdateType)
for i := 10; i < 30; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.DeploymentID = d.ID
alloc.NodeID = uuid.Generate()
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
handled[alloc.ID] = allocUpdateFnIgnore
}
mockUpdateFn := allocUpdateFnMock(handled, allocUpdateFnDestructive)
reconciler := NewAllocReconciler(testLogger(), mockUpdateFn, false, job.ID, job, d, allocs, nil, "")
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
// All should be ignored because nothing has been marked as
// healthy.
Ignore: 30,
},
},
})
}
// Tests the reconciler doesn't stop allocations when doing a rolling upgrade
// where the count of the old job allocs is < desired count.
func TestReconciler_RollingUpgrade_MissingAllocs(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = noCanaryUpdate
// Create 7 allocations from the old job
var allocs []*structs.Allocation
for i := 0; i < 7; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnDestructive, false, job.ID, job, nil, allocs, nil, "")
r := reconciler.Compute()
d := structs.NewDeployment(job)
d.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
DesiredTotal: 10,
}
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: d,
deploymentUpdates: nil,
place: 3,
destructive: 1,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 3,
DestructiveUpdate: 1,
Ignore: 6,
},
},
})
assertNamesHaveIndexes(t, intRange(7, 9), placeResultsToNames(r.place))
assertNamesHaveIndexes(t, intRange(0, 0), destructiveResultsToNames(r.destructiveUpdate))
}
// Tests that the reconciler handles rerunning a batch job in the case that the
// allocations are from an older instance of the job.
func TestReconciler_Batch_Rerun(t *testing.T) {
job := mock.Job()
job.Type = structs.JobTypeBatch
job.TaskGroups[0].Update = nil
// Create 10 allocations from the old job and have them be complete
var allocs []*structs.Allocation
for i := 0; i < 10; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
alloc.ClientStatus = structs.AllocClientStatusComplete
alloc.DesiredStatus = structs.AllocDesiredStatusStop
allocs = append(allocs, alloc)
}
// Create a copy of the job that is "new"
job2 := job.Copy()
job2.CreateIndex++
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, true, job2.ID, job2, nil, allocs, nil, "")
r := reconciler.Compute()
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: nil,
place: 10,
destructive: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Place: 10,
DestructiveUpdate: 0,
Ignore: 10,
},
},
})
assertNamesHaveIndexes(t, intRange(0, 9), placeResultsToNames(r.place))
}
2018-01-19 17:58:59 +00:00
// Test that a failed deployment will not result in rescheduling failed allocations
func TestReconciler_FailedDeployment_DontReschedule(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = noCanaryUpdate
tgName := job.TaskGroups[0].Name
now := time.Now()
2018-01-19 17:58:59 +00:00
// Create an existing failed deployment that has some placed allocs
d := structs.NewDeployment(job)
d.Status = structs.DeploymentStatusFailed
d.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
Promoted: true,
DesiredTotal: 5,
PlacedAllocs: 4,
}
// Create 4 allocations and mark two as failed
var allocs []*structs.Allocation
for i := 0; i < 4; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
allocs = append(allocs, alloc)
}
//create some allocations that are reschedulable now
2018-01-19 17:58:59 +00:00
allocs[2].ClientStatus = structs.AllocClientStatusFailed
allocs[2].TaskStates = map[string]*structs.TaskState{tgName: {State: "start",
StartedAt: now.Add(-1 * time.Hour),
FinishedAt: now.Add(-10 * time.Second)}}
2018-01-19 17:58:59 +00:00
allocs[3].ClientStatus = structs.AllocClientStatusFailed
allocs[3].TaskStates = map[string]*structs.TaskState{tgName: {State: "start",
StartedAt: now.Add(-1 * time.Hour),
FinishedAt: now.Add(-10 * time.Second)}}
2018-01-19 17:58:59 +00:00
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnDestructive, false, job.ID, job, d, allocs, nil, "")
2018-01-19 17:58:59 +00:00
r := reconciler.Compute()
// Assert that no rescheduled placements were created
assertResults(t, r, &resultExpectation{
place: 0,
createDeployment: nil,
deploymentUpdates: nil,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Ignore: 2,
},
},
})
}
// Test that a running deployment with failed allocs will not result in rescheduling failed allocations
func TestReconciler_DeploymentWithFailedAllocs_DontReschedule(t *testing.T) {
job := mock.Job()
job.TaskGroups[0].Update = noCanaryUpdate
tgName := job.TaskGroups[0].Name
now := time.Now()
// Mock deployment with failed allocs, but deployment watcher hasn't marked it as failed yet
d := structs.NewDeployment(job)
d.Status = structs.DeploymentStatusRunning
d.TaskGroups[job.TaskGroups[0].Name] = &structs.DeploymentState{
Promoted: false,
DesiredTotal: 5,
PlacedAllocs: 4,
}
// Create 4 allocations and mark two as failed
var allocs []*structs.Allocation
for i := 0; i < 4; i++ {
alloc := mock.Alloc()
alloc.Job = job
alloc.JobID = job.ID
alloc.NodeID = uuid.Generate()
alloc.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
alloc.TaskGroup = job.TaskGroups[0].Name
alloc.DeploymentID = d.ID
allocs = append(allocs, alloc)
}
// Create allocs that are reschedulable now
allocs[2].ClientStatus = structs.AllocClientStatusFailed
allocs[2].TaskStates = map[string]*structs.TaskState{tgName: {State: "start",
StartedAt: now.Add(-1 * time.Hour),
FinishedAt: now.Add(-10 * time.Second)}}
allocs[3].ClientStatus = structs.AllocClientStatusFailed
allocs[3].TaskStates = map[string]*structs.TaskState{tgName: {State: "start",
StartedAt: now.Add(-1 * time.Hour),
FinishedAt: now.Add(-10 * time.Second)}}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnDestructive, false, job.ID, job, d, allocs, nil, "")
r := reconciler.Compute()
// Assert that no rescheduled placements were created
assertResults(t, r, &resultExpectation{
place: 0,
createDeployment: nil,
deploymentUpdates: nil,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Ignore: 2,
},
},
})
}
// Test that a failed deployment cancels non-promoted canaries
func TestReconciler_FailedDeployment_AutoRevert_CancelCanaries(t *testing.T) {
// Create a job
job := mock.Job()
job.TaskGroups[0].Count = 3
job.TaskGroups[0].Update = &structs.UpdateStrategy{
Canary: 3,
MaxParallel: 2,
HealthCheck: structs.UpdateStrategyHealthCheck_Checks,
MinHealthyTime: 10 * time.Second,
HealthyDeadline: 10 * time.Minute,
Stagger: 31 * time.Second,
}
// Create v1 of the job
jobv1 := job.Copy()
jobv1.Version = 1
jobv1.TaskGroups[0].Meta = map[string]string{"version": "1"}
// Create v2 of the job
jobv2 := job.Copy()
jobv2.Version = 2
jobv2.TaskGroups[0].Meta = map[string]string{"version": "2"}
// Create an existing failed deployment that has promoted one task group
d := structs.NewDeployment(jobv2)
state := &structs.DeploymentState{
Promoted: false,
DesiredTotal: 3,
PlacedAllocs: 3,
}
d.TaskGroups[job.TaskGroups[0].Name] = state
// Create the original
var allocs []*structs.Allocation
for i := 0; i < 3; i++ {
new := mock.Alloc()
new.Job = jobv2
new.JobID = job.ID
new.NodeID = uuid.Generate()
new.Name = structs.AllocName(job.ID, job.TaskGroups[0].Name, uint(i))
new.TaskGroup = job.TaskGroups[0].Name
new.DeploymentID = d.ID
new.DeploymentStatus = &structs.AllocDeploymentStatus{
Healthy: helper.BoolToPtr(true),
}
new.ClientStatus = structs.AllocClientStatusRunning
allocs = append(allocs, new)
}
for i := 0; i < 3; i++ {
new := mock.Alloc()
new.Job = jobv1
new.JobID = jobv1.ID
new.NodeID = uuid.Generate()
new.Name = structs.AllocName(jobv1.ID, jobv1.TaskGroups[0].Name, uint(i))
new.TaskGroup = job.TaskGroups[0].Name
new.DeploymentID = uuid.Generate()
new.DeploymentStatus = &structs.AllocDeploymentStatus{
Healthy: helper.BoolToPtr(false),
}
new.DesiredStatus = structs.AllocDesiredStatusStop
new.ClientStatus = structs.AllocClientStatusFailed
allocs = append(allocs, new)
}
reconciler := NewAllocReconciler(testLogger(), allocUpdateFnIgnore, false, job.ID, jobv2, d, allocs, nil, "")
r := reconciler.Compute()
updates := []*structs.DeploymentStatusUpdate{
{
DeploymentID: d.ID,
Status: structs.DeploymentStatusSuccessful,
StatusDescription: structs.DeploymentStatusDescriptionSuccessful,
},
}
// Assert the correct results
assertResults(t, r, &resultExpectation{
createDeployment: nil,
deploymentUpdates: updates,
place: 0,
inplace: 0,
stop: 0,
desiredTGUpdates: map[string]*structs.DesiredUpdates{
job.TaskGroups[0].Name: {
Stop: 0,
InPlaceUpdate: 0,
Ignore: 3,
},
},
})
}