f-79: agent: section times, renvoi de system_powerdown, attentes #79

Signed-off-by: GnomeZworc <nicolas.boufidjeline@g3e.fr>
This commit is contained in:
GnomeZworc 2026-10-25 12:42:10 +01:00
commit 8e89865418
Signed by: nicolas.boufideline
GPG key ID: 4406BBBF8845D632
13 changed files with 602 additions and 80 deletions

View file

@ -91,7 +91,7 @@ paths:
"404":
$ref: "#/components/responses/NotFound"
"409":
description: VPC not deletable — only running or error states can be deleted, and all its subnets must be deleted first
description: VPC not deletable — only running or error states can be deleted, and all its subnets must be deleted or deleting; subnets still deleting are awaited by the asynchronous deletion
content:
application/json:
schema:
@ -288,7 +288,7 @@ paths:
"404":
$ref: "#/components/responses/NotFound"
"409":
description: Subnet not deletable — only running or error states can be deleted, and no VM that is not deleted may still have an interface on it
description: Subnet not deletable — only running or error states can be deleted, and every VM with an interface on it must be deleted or deleting; VMs still deleting are awaited by the asynchronous deletion
content:
application/json:
schema:

View file

@ -19,12 +19,22 @@ worker:
# Maximum number of commands queued before Dispatch blocks
buffer_size: 100
# Timing for commands that wait on resource state transitions
dispatcher:
# How long (in seconds) to wait before giving up
timeout_seconds: 300
# Interval (in seconds) between each state check
poll_seconds: 2
# Timing (in seconds) for commands that wait on resource state transitions
times:
# Graceful VM shutdown before QMP quit
timeout_vms_stop: 300
# Interval between two system_powerdown during the shutdown
next_vms_shutdown: 30
# Subnet deletion: wait for its VMs still deleting
timeout_subnet_stop: 360
# VPC deletion: wait for its subnets still deleting
timeout_vpc_stop: 420
# Subnet creation: wait for its VPC to be running
timeout_subnet_start: 300
# VM creation: wait for the subnets of its interfaces to be running
timeout_vms_start: 300
# Interval between each state check
poll_interval: 2
# Bridge interface used when the requested iface_type is not found in interfaces
default_interface: br-000000

View file

@ -24,10 +24,7 @@ type Config struct {
Count int `mapstructure:"count"`
BufferSize int `mapstructure:"buffer_size"`
} `mapstructure:"worker"`
Dispatcher struct {
TimeoutSeconds int `mapstructure:"timeout_seconds"`
PollSeconds int `mapstructure:"poll_seconds"`
} `mapstructure:"dispatcher"`
Times Times `mapstructure:"times"`
Logger struct {
Level string `mapstructure:"level"`
Debug bool `mapstructure:"debug"`
@ -73,8 +70,13 @@ func LoadConfig(path string) (*Config, error) {
v.SetDefault("prometheus.port", 9090)
v.SetDefault("worker.count", 4)
v.SetDefault("worker.buffer_size", 100)
v.SetDefault("dispatcher.timeout_seconds", 300)
v.SetDefault("dispatcher.poll_seconds", 2)
v.SetDefault("times.timeout_vms_stop", TimeoutVMsStop)
v.SetDefault("times.next_vms_shutdown", NextVMsShutdown)
v.SetDefault("times.timeout_subnet_stop", TimeoutSubnetStop)
v.SetDefault("times.timeout_vpc_stop", TimeoutVPCStop)
v.SetDefault("times.timeout_subnet_start", TimeoutSubnetStart)
v.SetDefault("times.timeout_vms_start", TimeoutVMsStart)
v.SetDefault("times.poll_interval", PollInterval)
v.SetDefault("metadata.run_dir", "/run/two/metadata")
v.SetDefault("dhcp.backend", BackendTwo)
v.SetDefault("qemu.ovmf_code_path", "/usr/share/OVMF/OVMF_CODE.fd")
@ -106,6 +108,9 @@ func LoadConfig(path string) (*Config, error) {
if err := cfg.Network.Validate(); err != nil {
return nil, fmt.Errorf("%s: %w", path, err)
}
if err := cfg.Times.Validate(); err != nil {
return nil, fmt.Errorf("%s: %w", path, err)
}
return &cfg, nil
}

View file

@ -0,0 +1,64 @@
package configuration
import (
"fmt"
"time"
)
const (
TimeoutVMsStop = 300
NextVMsShutdown = 30
TimeoutSubnetStop = 360
TimeoutVPCStop = 420
TimeoutSubnetStart = 300
TimeoutVMsStart = 300
PollInterval = 2
)
type Times struct {
TimeoutVMsStop int `mapstructure:"timeout_vms_stop"`
NextVMsShutdown int `mapstructure:"next_vms_shutdown"`
TimeoutSubnetStop int `mapstructure:"timeout_subnet_stop"`
TimeoutVPCStop int `mapstructure:"timeout_vpc_stop"`
TimeoutSubnetStart int `mapstructure:"timeout_subnet_start"`
TimeoutVMsStart int `mapstructure:"timeout_vms_start"`
PollInterval int `mapstructure:"poll_interval"`
}
func seconds(value, fallback int) time.Duration {
if value == 0 {
value = fallback
}
return time.Duration(value) * time.Second
}
func (t Times) VMsStop() time.Duration { return seconds(t.TimeoutVMsStop, TimeoutVMsStop) }
func (t Times) VMsShutdown() time.Duration { return seconds(t.NextVMsShutdown, NextVMsShutdown) }
func (t Times) SubnetStop() time.Duration { return seconds(t.TimeoutSubnetStop, TimeoutSubnetStop) }
func (t Times) VPCStop() time.Duration { return seconds(t.TimeoutVPCStop, TimeoutVPCStop) }
func (t Times) SubnetStart() time.Duration { return seconds(t.TimeoutSubnetStart, TimeoutSubnetStart) }
func (t Times) VMsStart() time.Duration { return seconds(t.TimeoutVMsStart, TimeoutVMsStart) }
func (t Times) Poll() time.Duration { return seconds(t.PollInterval, PollInterval) }
func (t Times) Validate() error {
for _, f := range []struct {
key string
value int
}{
{"timeout_vms_stop", t.TimeoutVMsStop},
{"next_vms_shutdown", t.NextVMsShutdown},
{"timeout_subnet_stop", t.TimeoutSubnetStop},
{"timeout_vpc_stop", t.TimeoutVPCStop},
{"timeout_subnet_start", t.TimeoutSubnetStart},
{"timeout_vms_start", t.TimeoutVMsStart},
{"poll_interval", t.PollInterval},
} {
if f.value <= 0 {
return fmt.Errorf("times.%s %d must be a positive number of seconds", f.key, f.value)
}
}
if t.NextVMsShutdown >= t.TimeoutVMsStop {
return fmt.Errorf("times.next_vms_shutdown %d must be below times.timeout_vms_stop %d", t.NextVMsShutdown, t.TimeoutVMsStop)
}
return nil
}

View file

@ -0,0 +1,95 @@
package configuration
import (
"strings"
"testing"
"time"
)
func TestLoadConfig_TimesDefauts(t *testing.T) {
cfg, err := LoadConfig(writeYAML(t, ""))
if err != nil {
t.Fatalf("LoadConfig a échoué : %v", err)
}
want := Times{
TimeoutVMsStop: 300,
NextVMsShutdown: 30,
TimeoutSubnetStop: 360,
TimeoutVPCStop: 420,
TimeoutSubnetStart: 300,
TimeoutVMsStart: 300,
PollInterval: 2,
}
if cfg.Times != want {
t.Errorf("times par défaut : %+v, attendu %+v", cfg.Times, want)
}
}
func TestLoadConfig_TimesValeursExplicites(t *testing.T) {
cfg, err := LoadConfig(writeYAML(t, "times:\n timeout_vms_stop: 600\n next_vms_shutdown: 10\n timeout_subnet_stop: 700\n timeout_vpc_stop: 800\n timeout_subnet_start: 50\n timeout_vms_start: 60\n poll_interval: 1\n"))
if err != nil {
t.Fatalf("LoadConfig a échoué : %v", err)
}
want := Times{600, 10, 700, 800, 50, 60, 1}
if cfg.Times != want {
t.Errorf("times : %+v, attendu %+v", cfg.Times, want)
}
}
func TestLoadConfig_AncienneSectionDispatcherIgnoree(t *testing.T) {
cfg, err := LoadConfig(writeYAML(t, "dispatcher:\n timeout_seconds: 900\n poll_seconds: 7\n"))
if err != nil {
t.Fatalf("une section dispatcher restante ne doit pas empêcher le démarrage : %v", err)
}
if cfg.Times.TimeoutVMsStop != 300 || cfg.Times.PollInterval != 2 {
t.Errorf("dispatcher.* ne doit rien changer : %+v", cfg.Times)
}
}
func TestLoadConfig_TimesInvalidesRefuses(t *testing.T) {
for _, c := range []struct{ yaml, msg string }{
{"times:\n timeout_vms_stop: 0\n", "times.timeout_vms_stop 0 must be a positive number of seconds"},
{"times:\n next_vms_shutdown: -1\n", "times.next_vms_shutdown -1 must be a positive number of seconds"},
{"times:\n timeout_subnet_stop: 0\n", "times.timeout_subnet_stop 0"},
{"times:\n timeout_vpc_stop: 0\n", "times.timeout_vpc_stop 0"},
{"times:\n timeout_subnet_start: 0\n", "times.timeout_subnet_start 0"},
{"times:\n timeout_vms_start: 0\n", "times.timeout_vms_start 0"},
{"times:\n poll_interval: 0\n", "times.poll_interval 0"},
{"times:\n next_vms_shutdown: 300\n", "times.next_vms_shutdown 300 must be below times.timeout_vms_stop 300"},
} {
if _, err := LoadConfig(writeYAML(t, c.yaml)); err == nil || !strings.Contains(err.Error(), c.msg) {
t.Errorf("%q : erreur %v, attendu %q", c.yaml, err, c.msg)
}
}
if _, err := LoadConfig(writeYAML(t, "times:\n next_vms_shutdown: 299\n")); err != nil {
t.Errorf("next_vms_shutdown 299 sous timeout_vms_stop 300 : %v", err)
}
}
func TestTimesDurees(t *testing.T) {
var zero Times
cases := []struct {
name string
got, want time.Duration
}{
{"VMsStop", zero.VMsStop(), 300 * time.Second},
{"VMsShutdown", zero.VMsShutdown(), 30 * time.Second},
{"SubnetStop", zero.SubnetStop(), 360 * time.Second},
{"VPCStop", zero.VPCStop(), 420 * time.Second},
{"SubnetStart", zero.SubnetStart(), 300 * time.Second},
{"VMsStart", zero.VMsStart(), 300 * time.Second},
{"Poll", zero.Poll(), 2 * time.Second},
}
for _, c := range cases {
if c.got != c.want {
t.Errorf("Times{}.%s() = %v, attendu %v", c.name, c.got, c.want)
}
}
set := Times{11, 12, 13, 14, 15, 16, 17}
got := []time.Duration{set.VMsStop(), set.VMsShutdown(), set.SubnetStop(), set.VPCStop(), set.SubnetStart(), set.VMsStart(), set.Poll()}
for i, d := range got {
if want := time.Duration(11+i) * time.Second; d != want {
t.Errorf("durée %d = %v, attendu %v", i, d, want)
}
}
}

View file

@ -1,6 +1,7 @@
package dispatcher
import (
"errors"
"fmt"
"sort"
"strconv"
@ -93,23 +94,21 @@ func (c CreateSubnetCommand) Prepare(db *badger.DB, cfg *configuration.Config) e
}
func (c CreateSubnetCommand) Execute(db *badger.DB, cfg *configuration.Config) error {
timeout := time.After(time.Duration(cfg.Dispatcher.TimeoutSeconds) * time.Second)
for {
err := waitUntil(cfg.Times.SubnetStart(), cfg.Times.Poll(), func() (bool, error) {
vpcState, err := state.Get(db, "vpc/"+c.VPC)
if err != nil {
return fmt.Errorf("vpc %q not found while waiting", c.VPC)
return false, fmt.Errorf("vpc %q not found while waiting", c.VPC)
}
if vpcState == state.Running {
break
}
if vpcState != state.Creating {
return fmt.Errorf("vpc %q is %s, cannot create subnet %q", c.VPC, vpcState, c.Name)
}
select {
case <-timeout:
return fmt.Errorf("timed out waiting for vpc %q to be running", c.VPC)
case <-time.After(time.Duration(cfg.Dispatcher.PollSeconds) * time.Second):
if vpcState != state.Running && vpcState != state.Creating {
return false, fmt.Errorf("vpc %q is %s, cannot create subnet %q", c.VPC, vpcState, c.Name)
}
return vpcState == state.Running, nil
})
if errors.Is(err, errWaitTimeout) {
return fmt.Errorf("timed out waiting for vpc %q to be running", c.VPC)
}
if err != nil {
return err
}
return subnet.CreateSubnet(db, c.Name, cfg)
}
@ -132,12 +131,41 @@ func (c DeleteSubnetCommand) Prepare(db *badger.DB, _ *configuration.Config) err
if err != nil {
return err
}
if len(vms) > 0 {
return fmt.Errorf("subnet %q is still used by vms %s", c.Name, strings.Join(vms, ", "))
var blocking []string
for _, name := range vms {
if s, err := state.Get(db, "vm/"+name); err != nil || s != state.Deleting {
blocking = append(blocking, name)
}
}
if len(blocking) > 0 {
return fmt.Errorf("subnet %q is still used by vms %s", c.Name, strings.Join(blocking, ", "))
}
return state.Set(db, c.Key(), state.Deleting)
}
func waitVMsDeleted(db *badger.DB, subnetName string, timeout, poll time.Duration) error {
err := waitUntil(timeout, poll, func() (bool, error) {
vms, err := vmsUsingSubnet(db, subnetName)
if err != nil {
return false, err
}
for _, name := range vms {
s, err := state.Get(db, "vm/"+name)
if err != nil {
continue
}
if s != state.Deleting {
return false, fmt.Errorf("vm %q is %s, subnet %q cannot be deleted", name, s, subnetName)
}
}
return len(vms) == 0, nil
})
if errors.Is(err, errWaitTimeout) {
return fmt.Errorf("timed out waiting for the vms of subnet %q to be deleted", subnetName)
}
return err
}
func vmsUsingSubnet(db *badger.DB, subnetName string) ([]string, error) {
entries, err := kv.ListByPrefix(db, "vm/")
if err != nil {
@ -159,6 +187,9 @@ func vmsUsingSubnet(db *badger.DB, subnetName string) ([]string, error) {
}
func (c DeleteSubnetCommand) Execute(db *badger.DB, cfg *configuration.Config) error {
if err := waitVMsDeleted(db, c.Name, cfg.Times.SubnetStop(), cfg.Times.Poll()); err != nil {
return err
}
if err := subnet.DeleteSubnet(db, c.Name, cfg); err != nil {
return err
}

View file

@ -418,7 +418,7 @@ func TestDeleteSubnetCommand_Prepare_RefusedWhileAVMUsesIt(t *testing.T) {
kv.AddInDB(db, "vm/i-c/state", "running")
kv.AddInDB(db, "vm/i-c/nic/0/subnet", "sn-a")
err := DeleteSubnetCommand{Name: "sn-a"}.Prepare(db, nil)
if err == nil || !strings.Contains(err.Error(), `subnet "sn-a" is still used by vms i-a, i-b, i-c`) {
if err == nil || !strings.Contains(err.Error(), `subnet "sn-a" is still used by vms i-a, i-c`) {
t.Fatalf("erreur %v", err)
}
if s, _ := kv.GetFromDB(db, "subnet/sn-a/state"); s != "running" {

View file

@ -1,6 +1,7 @@
package dispatcher
import (
"errors"
"fmt"
"math/rand"
"strconv"
@ -151,24 +152,23 @@ func allocateMetadataPort(db *badger.DB) (int, error) {
}
func (c StartVMCommand) Execute(db *badger.DB, cfg *configuration.Config) error {
timeout := time.After(time.Duration(cfg.Dispatcher.TimeoutSeconds) * time.Second)
timeout := time.Now().Add(cfg.Times.VMsStart())
for _, n := range c.NICs {
for {
err := waitUntil(time.Until(timeout), cfg.Times.Poll(), func() (bool, error) {
subnetState, err := state.Get(db, "subnet/"+n.Subnet)
if err != nil {
return fmt.Errorf("subnet %q not found while waiting", n.Subnet)
return false, fmt.Errorf("subnet %q not found while waiting", n.Subnet)
}
if subnetState == state.Running {
break
}
if subnetState != state.Creating {
return fmt.Errorf("subnet %q is %s, cannot start vm %q", n.Subnet, subnetState, c.Name)
}
select {
case <-timeout:
return fmt.Errorf("timed out waiting for subnet %q to be running", n.Subnet)
case <-time.After(time.Duration(cfg.Dispatcher.PollSeconds) * time.Second):
if subnetState != state.Running && subnetState != state.Creating {
return false, fmt.Errorf("subnet %q is %s, cannot start vm %q", n.Subnet, subnetState, c.Name)
}
return subnetState == state.Running, nil
})
if errors.Is(err, errWaitTimeout) {
return fmt.Errorf("timed out waiting for subnet %q to be running", n.Subnet)
}
if err != nil {
return err
}
}
return vm.StartVM(db, c.Name, cfg)

View file

@ -1,6 +1,7 @@
package dispatcher
import (
"errors"
"fmt"
"net"
"strconv"
@ -129,31 +130,8 @@ func (c DeleteVPCCommand) Prepare(db *badger.DB, _ *configuration.Config) error
}
func (c DeleteVPCCommand) Execute(db *badger.DB, cfg *configuration.Config) error {
timeout := time.After(time.Duration(cfg.Dispatcher.TimeoutSeconds) * time.Second)
for {
entries, err := kv.ListByPrefix(db, "subnet/")
if err != nil {
return fmt.Errorf("failed to list subnets: %w", err)
}
pending := false
for key, value := range entries {
if strings.HasSuffix(key, "/vpc") && value == c.Name {
subnetName := strings.Split(key, "/")[1]
s, _ := state.Get(db, "subnet/"+subnetName)
if s == state.Deleting {
pending = true
break
}
}
}
if !pending {
break
}
select {
case <-timeout:
return fmt.Errorf("timed out waiting for subnets of vpc %q to be deleted", c.Name)
case <-time.After(time.Duration(cfg.Dispatcher.PollSeconds) * time.Second):
}
if err := waitSubnetsDeleted(db, c.Name, cfg.Times.VPCStop(), cfg.Times.Poll()); err != nil {
return err
}
if err := vpc.DeleteVPC(db, c.Name); err != nil {
return err
@ -167,3 +145,35 @@ func (c DeleteVPCCommand) Execute(db *badger.DB, cfg *configuration.Config) erro
}
return nil
}
func waitSubnetsDeleted(db *badger.DB, vpcName string, timeout, poll time.Duration) error {
err := waitUntil(timeout, poll, func() (bool, error) {
entries, err := kv.ListByPrefix(db, "subnet/")
if err != nil {
return false, fmt.Errorf("failed to list subnets: %w", err)
}
pending := false
for key, value := range entries {
if !strings.HasSuffix(key, "/vpc") || value != vpcName {
continue
}
subnetName := strings.Split(key, "/")[1]
s, err := state.Get(db, "subnet/"+subnetName)
if err != nil {
continue
}
switch s {
case state.Deleted:
case state.Deleting:
pending = true
default:
return false, fmt.Errorf("subnet %q is %s, vpc %q cannot be deleted", subnetName, s, vpcName)
}
}
return !pending, nil
})
if errors.Is(err, errWaitTimeout) {
return fmt.Errorf("timed out waiting for subnets of vpc %q to be deleted", vpcName)
}
return err
}

View file

@ -0,0 +1,26 @@
package dispatcher
import (
"errors"
"time"
)
var errWaitTimeout = errors.New("timed out")
func waitUntil(timeout, poll time.Duration, done func() (bool, error)) error {
deadline := time.After(timeout)
for {
ok, err := done()
if err != nil {
return err
}
if ok {
return nil
}
select {
case <-deadline:
return errWaitTimeout
case <-time.After(poll):
}
}
}

View file

@ -0,0 +1,168 @@
package dispatcher
import (
"errors"
"strings"
"testing"
"time"
configuration "git.g3e.fr/syonad/two/internal/config/agent"
"git.g3e.fr/syonad/two/pkg/db/kv"
)
func TestWaitUntil(t *testing.T) {
calls := 0
err := waitUntil(time.Second, time.Millisecond, func() (bool, error) {
calls++
return calls == 3, nil
})
if err != nil || calls != 3 {
t.Errorf("err %v, %d appels, attendu fin au troisième", err, calls)
}
if err := waitUntil(30*time.Millisecond, 5*time.Millisecond, func() (bool, error) { return false, nil }); !errors.Is(err, errWaitTimeout) {
t.Errorf("délai dépassé : %v", err)
}
boom := errors.New("boom")
calls = 0
err = waitUntil(time.Second, time.Millisecond, func() (bool, error) {
calls++
return false, boom
})
if !errors.Is(err, boom) || calls != 1 {
t.Errorf("erreur %v après %d appels, attendu boom au premier", err, calls)
}
}
func TestDeleteSubnetCommand_Prepare_AcceptsVMsBeingDeleted(t *testing.T) {
_, db := newTestDispatcher(t)
kv.AddInDB(db, "subnet/sn-a/state", "running")
kv.AddInDB(db, "vm/i-a/state", "deleting")
kv.AddInDB(db, "vm/i-a/nic/0/subnet", "sn-a")
kv.AddInDB(db, "vm/i-gone/state", "deleted")
kv.AddInDB(db, "vm/i-gone/nic/0/subnet", "sn-a")
if err := (DeleteSubnetCommand{Name: "sn-a"}).Prepare(db, nil); err != nil {
t.Fatalf("VM en deleting : %v", err)
}
if s, _ := kv.GetFromDB(db, "subnet/sn-a/state"); s != "deleting" {
t.Errorf("état %q, attendu deleting", s)
}
}
func TestDeleteSubnetCommand_Prepare_ListsOnlyTheVMsNotBeingDeleted(t *testing.T) {
_, db := newTestDispatcher(t)
kv.AddInDB(db, "subnet/sn-a/state", "running")
kv.AddInDB(db, "vm/i-a/state", "deleting")
kv.AddInDB(db, "vm/i-a/nic/0/subnet", "sn-a")
kv.AddInDB(db, "vm/i-b/state", "running")
kv.AddInDB(db, "vm/i-b/nic/0/subnet", "sn-a")
err := DeleteSubnetCommand{Name: "sn-a"}.Prepare(db, nil)
if err == nil || !strings.Contains(err.Error(), `subnet "sn-a" is still used by vms i-b`) || strings.Contains(err.Error(), "i-a") {
t.Fatalf("erreur %v", err)
}
}
func TestWaitVMsDeleted_ReturnsOnceTheVMsAreGone(t *testing.T) {
_, db := newTestDispatcher(t)
kv.AddInDB(db, "vm/i-a/state", "deleting")
kv.AddInDB(db, "vm/i-a/nic/0/subnet", "sn-a")
kv.AddInDB(db, "vm/i-other/state", "running")
kv.AddInDB(db, "vm/i-other/nic/0/subnet", "sn-b")
go func() {
time.Sleep(30 * time.Millisecond)
kv.DeleteInDB(db, "vm/i-a")
}()
if err := waitVMsDeleted(db, "sn-a", 2*time.Second, 5*time.Millisecond); err != nil {
t.Fatalf("attente : %v", err)
}
}
func TestWaitVMsDeleted_FailsWhenAVMEndsInError(t *testing.T) {
_, db := newTestDispatcher(t)
kv.AddInDB(db, "vm/i-a/state", "deleting")
kv.AddInDB(db, "vm/i-a/nic/0/subnet", "sn-a")
go func() {
time.Sleep(20 * time.Millisecond)
kv.AddInDB(db, "vm/i-a/state", "error")
}()
err := waitVMsDeleted(db, "sn-a", 2*time.Second, 5*time.Millisecond)
if err == nil || !strings.Contains(err.Error(), `vm "i-a" is error, subnet "sn-a" cannot be deleted`) {
t.Fatalf("erreur %v", err)
}
}
func TestWaitVMsDeleted_TimesOut(t *testing.T) {
_, db := newTestDispatcher(t)
kv.AddInDB(db, "vm/i-a/state", "deleting")
kv.AddInDB(db, "vm/i-a/nic/0/subnet", "sn-a")
err := waitVMsDeleted(db, "sn-a", 30*time.Millisecond, 5*time.Millisecond)
if err == nil || err.Error() != `timed out waiting for the vms of subnet "sn-a" to be deleted` {
t.Fatalf("erreur %v", err)
}
}
func TestWaitSubnetsDeleted_ReturnsOnceTheSubnetsAreGone(t *testing.T) {
_, db := newTestDispatcher(t)
kv.AddInDB(db, "subnet/sn-a/state", "deleting")
kv.AddInDB(db, "subnet/sn-a/vpc", "vp-a")
kv.AddInDB(db, "subnet/sn-b/state", "deleted")
kv.AddInDB(db, "subnet/sn-b/vpc", "vp-a")
kv.AddInDB(db, "subnet/sn-x/state", "running")
kv.AddInDB(db, "subnet/sn-x/vpc", "vp-x")
go func() {
time.Sleep(30 * time.Millisecond)
kv.DeleteInDB(db, "subnet/sn-a")
}()
if err := waitSubnetsDeleted(db, "vp-a", 2*time.Second, 5*time.Millisecond); err != nil {
t.Fatalf("attente : %v", err)
}
}
func TestWaitSubnetsDeleted_FailsWhenASubnetEndsInError(t *testing.T) {
_, db := newTestDispatcher(t)
kv.AddInDB(db, "subnet/sn-a/state", "deleting")
kv.AddInDB(db, "subnet/sn-a/vpc", "vp-a")
go func() {
time.Sleep(20 * time.Millisecond)
kv.AddInDB(db, "subnet/sn-a/state", "error")
}()
err := waitSubnetsDeleted(db, "vp-a", 2*time.Second, 5*time.Millisecond)
if err == nil || !strings.Contains(err.Error(), `subnet "sn-a" is error, vpc "vp-a" cannot be deleted`) {
t.Fatalf("erreur %v", err)
}
}
func TestWaitSubnetsDeleted_TimesOut(t *testing.T) {
_, db := newTestDispatcher(t)
kv.AddInDB(db, "subnet/sn-a/state", "deleting")
kv.AddInDB(db, "subnet/sn-a/vpc", "vp-a")
err := waitSubnetsDeleted(db, "vp-a", 30*time.Millisecond, 5*time.Millisecond)
if err == nil || err.Error() != `timed out waiting for subnets of vpc "vp-a" to be deleted` {
t.Fatalf("erreur %v", err)
}
}
func TestDeleteSubnetCommand_Execute_FailsBeforeTouchingTheHostWhenAVMIsInError(t *testing.T) {
_, db := newTestDispatcher(t)
kv.AddInDB(db, "subnet/sn-a/state", "deleting")
kv.AddInDB(db, "vm/i-a/state", "error")
kv.AddInDB(db, "vm/i-a/nic/0/subnet", "sn-a")
cfg := &configuration.Config{Times: configuration.Times{TimeoutSubnetStop: 1, PollInterval: 1}}
err := DeleteSubnetCommand{Name: "sn-a"}.Execute(db, cfg)
if err == nil || err.Error() != `vm "i-a" is error, subnet "sn-a" cannot be deleted` {
t.Fatalf("erreur %v", err)
}
}
func TestDeleteVPCCommand_Execute_FailsBeforeTouchingTheHostWhenASubnetIsInError(t *testing.T) {
_, db := newTestDispatcher(t)
kv.AddInDB(db, "vpc/vp-a/state", "deleting")
kv.AddInDB(db, "subnet/sn-a/state", "error")
kv.AddInDB(db, "subnet/sn-a/vpc", "vp-a")
cfg := &configuration.Config{Times: configuration.Times{TimeoutVPCStop: 1, PollInterval: 1}}
err := DeleteVPCCommand{Name: "vp-a"}.Execute(db, cfg)
if err == nil || err.Error() != `subnet "sn-a" is error, vpc "vp-a" cannot be deleted` {
t.Fatalf("erreur %v", err)
}
}

View file

@ -1,6 +1,7 @@
package vm
import (
"encoding/json"
"fmt"
"os"
"path/filepath"
@ -47,13 +48,7 @@ func StopVM(db *badger.DB, name string, cfg *configuration.Config) error {
if _, err := os.Stat(socketPath); err == nil {
// socket présent : tenter l'arrêt gracieux
if _, err := qmp.Send(socketPath, []string{`{"execute":"system_powerdown"}`}); err == nil {
waitQMPDead(socketPath,
time.Duration(cfg.Dispatcher.TimeoutSeconds)*time.Second,
time.Duration(cfg.Dispatcher.PollSeconds)*time.Second,
)
}
// connexion QMP échouée : QEMU déjà mort
shutdownQEMU(qmp.Send, socketPath, cfg.Times.VMsStop(), cfg.Times.VMsShutdown(), cfg.Times.Poll())
}
// socket absent ou QEMU déjà arrêté : cleanup direct
@ -99,15 +94,31 @@ func removeDHCPFiles(d vmData, name string, backend dhcpbackend.Backend) error {
return nil
}
func waitQMPDead(socketPath string, timeout, poll time.Duration) {
timer := time.After(timeout)
type qmpSender func(socketPath string, commands []string) ([]json.RawMessage, error)
const (
qmpPowerdown = `{"execute":"system_powerdown"}`
qmpQuit = `{"execute":"quit"}`
)
func shutdownQEMU(send qmpSender, socketPath string, timeout, resend, poll time.Duration) {
if _, err := send(socketPath, []string{qmpPowerdown}); err != nil {
return
}
deadline := time.After(timeout)
again := time.NewTicker(resend)
defer again.Stop()
for {
select {
case <-timer:
qmp.Send(socketPath, []string{`{"execute":"quit"}`})
case <-deadline:
send(socketPath, []string{qmpQuit})
return
case <-again.C:
if _, err := send(socketPath, []string{qmpPowerdown}); err != nil {
return
}
case <-time.After(poll):
if _, err := qmp.Send(socketPath, nil); err != nil {
if _, err := send(socketPath, nil); err != nil {
return
}
}

View file

@ -0,0 +1,102 @@
package vm
import (
"encoding/json"
"errors"
"sync"
"testing"
"time"
)
type fakeQEMU struct {
mu sync.Mutex
calls []string
dieAfter int
powerdown int
dead bool
}
func (f *fakeQEMU) send(_ string, commands []string) ([]json.RawMessage, error) {
f.mu.Lock()
defer f.mu.Unlock()
call := "ping"
if len(commands) > 0 {
call = commands[0]
}
f.calls = append(f.calls, call)
if f.dead {
return nil, errors.New("qmp dial: connection refused")
}
switch call {
case qmpPowerdown:
f.powerdown++
if f.dieAfter > 0 && f.powerdown >= f.dieAfter {
f.dead = true
}
case qmpQuit:
f.dead = true
}
return nil, nil
}
func (f *fakeQEMU) count(call string) int {
f.mu.Lock()
defer f.mu.Unlock()
n := 0
for _, c := range f.calls {
if c == call {
n++
}
}
return n
}
func TestShutdownQEMU_ResendsPowerdownUntilTheGuestListens(t *testing.T) {
f := &fakeQEMU{dieAfter: 3}
start := time.Now()
shutdownQEMU(f.send, "qmp.sock", 5*time.Second, 50*time.Millisecond, 10*time.Millisecond)
if d := time.Since(start); d > 2*time.Second {
t.Errorf("arrêt en %v : le renvoi aurait dû l'obtenir bien avant le délai", d)
}
if n := f.count(qmpPowerdown); n != 3 {
t.Errorf("%d system_powerdown envoyés, attendu 3 (le troisième est entendu)", n)
}
if n := f.count(qmpQuit); n != 0 {
t.Errorf("quit envoyé alors que l'invité s'est arrêté : %q", f.calls)
}
}
func TestShutdownQEMU_QuitsAtTheDeadline(t *testing.T) {
f := &fakeQEMU{}
start := time.Now()
shutdownQEMU(f.send, "qmp.sock", 200*time.Millisecond, 60*time.Millisecond, 10*time.Millisecond)
d := time.Since(start)
if d < 200*time.Millisecond || d > 2*time.Second {
t.Errorf("arrêt en %v, attendu au délai de 200ms", d)
}
if n := f.count(qmpPowerdown); n < 3 {
t.Errorf("%d system_powerdown envoyés en 200ms avec un renvoi toutes les 60ms", n)
}
if n := f.count(qmpQuit); n != 1 {
t.Errorf("%d quit envoyés, attendu 1 : %q", n, f.calls)
}
}
func TestShutdownQEMU_StopsWhenTheGuestShutsDownOnItsOwn(t *testing.T) {
f := &fakeQEMU{dieAfter: 1}
shutdownQEMU(f.send, "qmp.sock", 5*time.Second, time.Second, 10*time.Millisecond)
if n := f.count(qmpPowerdown); n != 1 {
t.Errorf("%d system_powerdown envoyés, attendu 1", n)
}
if n := f.count(qmpQuit); n != 0 {
t.Errorf("quit envoyé : %q", f.calls)
}
}
func TestShutdownQEMU_NothingMoreWhenQEMUIsAlreadyDead(t *testing.T) {
f := &fakeQEMU{dead: true}
shutdownQEMU(f.send, "qmp.sock", 5*time.Second, time.Second, 10*time.Millisecond)
if len(f.calls) != 1 {
t.Errorf("appels QMP : %q, attendu un seul essai", f.calls)
}
}