talos/internal/integration/provision/upgrade.go
Andrew Rynhard 69fa63a7b2 refactor: perform upgrade upon reboot
This PR introduces a new strategy for upgrades. Instead of attempting to
zap the partition table, create a new one, and then format the
partitions, this change will only update the `vmlinuz`, and
`initramfs.xz` being used to boot. It introduces an A/B style upgrade
process, which will allow for easy rollbacks. One deviation from our
original intention with upgrades is that this change does not completely
reset a node. It falls just short of that and does not reset the
partition table. This forces us to keep the current partition scheme in
mind as we make changes in the future, because an upgrade assumes a
specific partition scheme. We can improve upgrades further in the
future, but this will at least make them more dependable. Finally, one
more feature in this PR is the ability to keep state. This enables
single node clusters to upgrade since we keep the etcd data around.

Signed-off-by: Andrew Rynhard <andrew@andrewrynhard.com>
2020-03-20 17:32:18 -07:00

415 lines
12 KiB
Go

// This Source Code Form is subject to the terms of the Mozilla Public
// License, v. 2.0. If a copy of the MPL was not distributed with this
// file, You can obtain one at http://mozilla.org/MPL/2.0/.
// +build integration_provision
package provision
import (
"context"
"crypto/sha256"
"fmt"
"io/ioutil"
"net"
"os"
"path/filepath"
"regexp"
"strings"
"time"
"github.com/stretchr/testify/suite"
machineapi "github.com/talos-systems/talos/api/machine"
talosclient "github.com/talos-systems/talos/cmd/osctl/pkg/client"
"github.com/talos-systems/talos/cmd/osctl/pkg/mgmt/helpers"
"github.com/talos-systems/talos/internal/integration/base"
"github.com/talos-systems/talos/internal/pkg/provision"
"github.com/talos-systems/talos/internal/pkg/provision/access"
"github.com/talos-systems/talos/internal/pkg/provision/check"
"github.com/talos-systems/talos/internal/pkg/provision/providers/firecracker"
"github.com/talos-systems/talos/internal/pkg/runtime"
"github.com/talos-systems/talos/pkg/config"
"github.com/talos-systems/talos/pkg/config/machine"
"github.com/talos-systems/talos/pkg/config/types/v1alpha1"
"github.com/talos-systems/talos/pkg/config/types/v1alpha1/generate"
"github.com/talos-systems/talos/pkg/constants"
talosnet "github.com/talos-systems/talos/pkg/net"
"github.com/talos-systems/talos/pkg/retry"
)
type upgradeSpec struct {
ShortName string
SourceKernelPath string
SourceInitramfsPath string
SourceInstallerImage string
SourceVersion string
TargetInstallerImage string
TargetVersion string
MasterNodes int
WorkerNodes int
}
const (
talos03Version = "v0.3.2-1-g71ac6696"
talos04Version = "v0.4.0-alpha.5-19-g8913d9df"
)
var (
defaultNameservers = []net.IP{net.ParseIP("8.8.8.8"), net.ParseIP("1.1.1.1")}
defaultCNIBinPath = []string{"/opt/cni/bin"}
)
const (
defaultCNIConfDir = "/etc/cni/conf.d"
defaultCNICacheDir = "/var/lib/cni"
)
func trimVersion(version string) string {
// remove anything extra after semantic version core, `v0.3.2-1-abcd` -> `v0.3.2`
return regexp.MustCompile(`(-\d+-g[0-9a-f]+)$`).ReplaceAllString(version, "")
}
// upgradeZeroThreeToZeroFour upgrades Talos 0.3.x to Talos 0.4.x.
func upgradeZeroThreeToZeroFour() upgradeSpec {
return upgradeSpec{
ShortName: fmt.Sprintf("%s-%s", talos03Version, talos04Version),
SourceKernelPath: helpers.ArtifactPath(filepath.Join(trimVersion(talos03Version), constants.KernelUncompressedAsset)),
SourceInitramfsPath: helpers.ArtifactPath(filepath.Join(trimVersion(talos03Version), constants.InitramfsAsset)),
SourceInstallerImage: fmt.Sprintf("%s:%s", constants.DefaultInstallerImageRepository, talos03Version),
SourceVersion: talos03Version,
TargetInstallerImage: fmt.Sprintf("%s:%s", constants.DefaultInstallerImageRepository, talos04Version),
TargetVersion: talos04Version,
MasterNodes: DefaultSettings.MasterNodes,
WorkerNodes: DefaultSettings.WorkerNodes,
}
}
// upgradeZeroFourToCurrent upgrade Talos 0.4.x to the current version of Talos.
func upgradeZeroFourToCurrent() upgradeSpec {
return upgradeSpec{
ShortName: fmt.Sprintf("%s-%s", talos04Version, DefaultSettings.CurrentVersion),
SourceKernelPath: helpers.ArtifactPath(filepath.Join(trimVersion(talos04Version), constants.KernelUncompressedAsset)),
SourceInitramfsPath: helpers.ArtifactPath(filepath.Join(trimVersion(talos04Version), constants.InitramfsAsset)),
SourceInstallerImage: fmt.Sprintf("%s:%s", constants.DefaultInstallerImageRepository, talos04Version),
SourceVersion: talos04Version,
TargetInstallerImage: fmt.Sprintf("%s/%s:%s", DefaultSettings.TargetInstallImageRegistry, constants.DefaultInstallerImageName, DefaultSettings.CurrentVersion),
TargetVersion: DefaultSettings.CurrentVersion,
MasterNodes: DefaultSettings.MasterNodes,
WorkerNodes: DefaultSettings.WorkerNodes,
}
}
type UpgradeSuite struct {
suite.Suite
base.TalosSuite
specGen func() upgradeSpec
spec upgradeSpec
provisioner provision.Provisioner
configBundle *v1alpha1.ConfigBundle
clusterAccess provision.ClusterAccess
ctx context.Context
ctxCancel context.CancelFunc
stateDir string
}
// SetupSuite ...
func (suite *UpgradeSuite) SetupSuite() {
// call generate late in the flow, as it needs to pick up settings overridden by test runner
suite.spec = suite.specGen()
suite.T().Logf("upgrade spec = %v", suite.spec)
// timeout for the whole test
suite.ctx, suite.ctxCancel = context.WithTimeout(context.Background(), 30*time.Minute)
var err error
suite.provisioner, err = firecracker.NewProvisioner(suite.ctx)
suite.Require().NoError(err)
}
// TearDownSuite ...
func (suite *UpgradeSuite) TearDownSuite() {
if suite.T().Failed() && suite.Cluster != nil {
// for failed tests, produce crash dump for easier debugging,
// as cluster is going to be torn down below
suite.provisioner.CrashDump(suite.ctx, suite.Cluster, os.Stderr)
}
if suite.clusterAccess != nil {
suite.Assert().NoError(suite.clusterAccess.Close())
}
if suite.Cluster != nil {
suite.Assert().NoError(suite.provisioner.Destroy(suite.ctx, suite.Cluster))
}
suite.ctxCancel()
if suite.stateDir != "" {
suite.Assert().NoError(os.RemoveAll(suite.stateDir))
}
if suite.provisioner != nil {
suite.Assert().NoError(suite.provisioner.Close())
}
}
// setupCluster provisions source clusters and waits for health
func (suite *UpgradeSuite) setupCluster() {
shortNameHash := sha256.Sum256([]byte(suite.spec.ShortName))
clusterName := fmt.Sprintf("upgrade.%x", shortNameHash[:8])
_, cidr, err := net.ParseCIDR(DefaultSettings.CIDR)
suite.Require().NoError(err)
var gatewayIP net.IP
gatewayIP, err = talosnet.NthIPInNetwork(cidr, 1)
suite.Require().NoError(err)
ips := make([]net.IP, suite.spec.MasterNodes+suite.spec.WorkerNodes)
for i := range ips {
ips[i], err = talosnet.NthIPInNetwork(cidr, i+2)
suite.Require().NoError(err)
}
suite.stateDir, err = ioutil.TempDir("", "talos-integration")
suite.Require().NoError(err)
suite.T().Logf("initalizing provisioner with cluster name %q, state directory %q", clusterName, suite.stateDir)
request := provision.ClusterRequest{
Name: clusterName,
Network: provision.NetworkRequest{
Name: clusterName,
CIDR: *cidr,
GatewayAddr: gatewayIP,
MTU: DefaultSettings.MTU,
Nameservers: defaultNameservers,
CNI: provision.CNIConfig{
BinPath: defaultCNIBinPath,
ConfDir: defaultCNIConfDir,
CacheDir: defaultCNICacheDir,
},
LoadBalancer: provision.LoadBalancerConfig{
// as we're upgrading from older versions of Talos,
// disable LB to all apid nodes
LimitApidOnlyInitNode: true,
},
},
KernelPath: suite.spec.SourceKernelPath,
InitramfsPath: suite.spec.SourceInitramfsPath,
SelfExecutable: suite.OsctlPath,
StateDirectory: suite.stateDir,
}
defaultInternalLB, defaultExternalLB := suite.provisioner.GetLoadBalancers(request.Network)
genOptions := suite.provisioner.GenOptions(request.Network)
for _, registryMirror := range DefaultSettings.RegistryMirrors {
parts := strings.SplitN(registryMirror, "=", 2)
suite.Require().Len(parts, 2)
genOptions = append(genOptions, generate.WithRegistryMirror(parts[0], parts[1]))
}
suite.configBundle, err = config.NewConfigBundle(config.WithInputOptions(
&config.InputOptions{
ClusterName: clusterName,
Endpoint: fmt.Sprintf("https://%s:6443", defaultInternalLB),
KubeVersion: constants.DefaultKubernetesVersion, // TODO: should be upgradeable
GenOptions: append(
genOptions,
generate.WithEndpointList([]string{defaultExternalLB}),
generate.WithInstallImage(suite.spec.SourceInstallerImage),
),
}))
suite.Require().NoError(err)
for i := 0; i < suite.spec.MasterNodes; i++ {
var cfg runtime.Configurator
if i == 0 {
cfg = suite.configBundle.Init()
} else {
cfg = suite.configBundle.ControlPlane()
}
request.Nodes = append(request.Nodes,
provision.NodeRequest{
Name: fmt.Sprintf("master-%d", i+1),
IP: ips[i],
Memory: DefaultSettings.MemMB * 1024 * 1024,
NanoCPUs: DefaultSettings.CPUs * 1000 * 1000 * 1000,
DiskSize: DefaultSettings.DiskGB * 1024 * 1024 * 1024,
Config: cfg,
})
}
for i := 1; i <= suite.spec.WorkerNodes; i++ {
request.Nodes = append(request.Nodes,
provision.NodeRequest{
Name: fmt.Sprintf("worker-%d", i),
IP: ips[suite.spec.MasterNodes+i-1],
Memory: DefaultSettings.MemMB * 1024 * 1024,
NanoCPUs: DefaultSettings.CPUs * 1000 * 1000 * 1000,
DiskSize: DefaultSettings.DiskGB * 1024 * 1024 * 1024,
Config: suite.configBundle.Join(),
})
}
suite.Cluster, err = suite.provisioner.Create(suite.ctx, request, provision.WithBootladerEmulation(), provision.WithTalosConfig(suite.configBundle.TalosConfig()))
suite.Require().NoError(err)
suite.clusterAccess = access.NewAdapter(suite.Cluster, provision.WithTalosConfig(suite.configBundle.TalosConfig()))
suite.waitForClusterHealth()
}
// waitForClusterHealth asserts cluster health after any change
func (suite *UpgradeSuite) waitForClusterHealth() {
checkCtx, checkCtxCancel := context.WithTimeout(suite.ctx, 10*time.Minute)
defer checkCtxCancel()
suite.Require().NoError(check.Wait(checkCtx, suite.clusterAccess, check.DefaultClusterChecks(), check.StderrReporter()))
}
func (suite *UpgradeSuite) assertSameVersionCluster(client *talosclient.Client, expectedVersion string) {
nodes := make([]string, len(suite.Cluster.Info().Nodes))
for i, node := range suite.Cluster.Info().Nodes {
nodes[i] = node.PrivateIP.String()
}
ctx := talosclient.WithNodes(suite.ctx, nodes...)
var v *machineapi.VersionResponse
err := retry.Constant(
time.Minute,
).Retry(func() error {
var e error
v, e = client.Version(ctx)
return retry.ExpectedError(e)
})
suite.Require().NoError(err)
suite.Require().Len(v.Messages, len(nodes))
for _, version := range v.Messages {
suite.Assert().Equal(expectedVersion, version.Version.Tag)
}
}
func (suite *UpgradeSuite) readVersion(client *talosclient.Client, nodeCtx context.Context) (version string, err error) {
var v *machineapi.VersionResponse
v, err = client.Version(nodeCtx)
if err != nil {
return
}
version = v.Messages[0].Version.Tag
return
}
func (suite *UpgradeSuite) upgradeNode(client *talosclient.Client, node provision.NodeInfo) {
suite.T().Logf("upgrading node %s", node.PrivateIP)
nodeCtx := talosclient.WithNodes(suite.ctx, node.PrivateIP.String())
resp, err := client.Upgrade(nodeCtx, suite.spec.TargetInstallerImage, false)
suite.Require().NoError(err)
suite.Require().Equal("Upgrade request received", resp.Messages[0].Ack)
// wait for the version to be equal to target version
suite.Require().NoError(retry.Constant(5 * time.Minute).Retry(func() error {
var version string
version, err = suite.readVersion(client, nodeCtx)
if err != nil {
// API might be unresponsive during upgrade
return retry.ExpectedError(err)
}
if version != suite.spec.TargetVersion {
// upgrade not finished yet
return retry.ExpectedError(fmt.Errorf("node %q version doesn't match expected: expected %q, got %q", node.PrivateIP.String(), suite.spec.TargetVersion, version))
}
return nil
}))
suite.waitForClusterHealth()
}
// TestRolling performs rolling upgrade starting with master nodes.
func (suite *UpgradeSuite) TestRolling() {
suite.setupCluster()
client, err := suite.clusterAccess.Client()
suite.Require().NoError(err)
// verify initial cluster version
suite.assertSameVersionCluster(client, suite.spec.SourceVersion)
// upgrade master nodes
for _, node := range suite.Cluster.Info().Nodes {
if node.Type == machine.TypeInit || node.Type == machine.TypeControlPlane {
suite.upgradeNode(client, node)
}
}
// upgrade worker nodes
for _, node := range suite.Cluster.Info().Nodes {
if node.Type == machine.TypeWorker {
suite.upgradeNode(client, node)
}
}
// verify final cluster version
suite.assertSameVersionCluster(client, suite.spec.TargetVersion)
}
// SuiteName ...
func (suite *UpgradeSuite) SuiteName() string {
if suite.spec.ShortName == "" {
suite.spec = suite.specGen()
}
return fmt.Sprintf("provision.UpgradeSuite.%s", suite.spec.ShortName)
}
func init() {
allSuites = append(allSuites,
&UpgradeSuite{specGen: upgradeZeroThreeToZeroFour},
&UpgradeSuite{specGen: upgradeZeroFourToCurrent},
)
}