Upgrade to etcd 3.3.17

This commit is contained in:
Joe Betz
2019-10-11 13:28:41 -07:00
parent efb5e676e1
commit 41e03d26d5
22 changed files with 190 additions and 101 deletions

View File

@@ -16,7 +16,9 @@
package endpoint
import (
"context"
"fmt"
"net"
"net/url"
"strings"
"sync"
@@ -228,13 +230,18 @@ func ParseTarget(target string) (string, string, error) {
return parts[0], parts[1], nil
}
// ParseHostPort splits a "<host>:<port>" string into the host and port parts.
// The port part is optional.
func ParseHostPort(hostPort string) (host string, port string) {
parts := strings.SplitN(hostPort, ":", 2)
host = parts[0]
if len(parts) > 1 {
port = parts[1]
// Dialer dials a endpoint using net.Dialer.
// Context cancelation and timeout are supported.
func Dialer(ctx context.Context, dialEp string) (net.Conn, error) {
proto, host, _ := ParseEndpoint(dialEp)
select {
case <-ctx.Done():
return nil, ctx.Err()
default:
}
return host, port
dialer := &net.Dialer{}
if deadline, ok := ctx.Deadline(); ok {
dialer.Deadline = deadline
}
return dialer.DialContext(ctx, proto, host)
}

View File

@@ -25,13 +25,13 @@ import (
"sync"
"time"
"github.com/google/uuid"
"github.com/coreos/etcd/clientv3/balancer"
"github.com/coreos/etcd/clientv3/balancer/picker"
"github.com/coreos/etcd/clientv3/balancer/resolver/endpoint"
"github.com/coreos/etcd/clientv3/credentials"
"github.com/coreos/etcd/etcdserver/api/v3rpc/rpctypes"
"github.com/coreos/etcd/pkg/logutil"
"github.com/google/uuid"
"go.uber.org/zap"
"google.golang.org/grpc"
"google.golang.org/grpc/codes"
@@ -226,24 +226,17 @@ func (c *Client) dialSetupOpts(creds grpccredentials.TransportCredentials, dopts
}
opts = append(opts, dopts...)
// Provide a net dialer that supports cancelation and timeout.
f := func(dialEp string, t time.Duration) (net.Conn, error) {
proto, host, _ := endpoint.ParseEndpoint(dialEp)
select {
case <-c.ctx.Done():
return nil, c.ctx.Err()
default:
}
dialer := &net.Dialer{Timeout: t}
return dialer.DialContext(c.ctx, proto, host)
}
opts = append(opts, grpc.WithDialer(f))
dialer := endpoint.Dialer
if creds != nil {
opts = append(opts, grpc.WithTransportCredentials(creds))
// gRPC load balancer workaround. See credentials.transportCredential for details.
if credsDialer, ok := creds.(TransportCredentialsWithDialer); ok {
dialer = credsDialer.Dialer
}
} else {
opts = append(opts, grpc.WithInsecure())
}
opts = append(opts, grpc.WithContextDialer(dialer))
// Interceptor retry and backoff.
// TODO: Replace all of clientv3/retry.go with interceptor based retry, or with
@@ -262,7 +255,10 @@ func (c *Client) dialSetupOpts(creds grpccredentials.TransportCredentials, dopts
// Dial connects to a single endpoint using the client's config.
func (c *Client) Dial(ep string) (*grpc.ClientConn, error) {
creds := c.directDialCreds(ep)
creds, err := c.directDialCreds(ep)
if err != nil {
return nil, err
}
// Use the grpc passthrough resolver to directly dial a single endpoint.
// This resolver passes through the 'unix' and 'unixs' endpoints schemes used
// by etcd without modification, allowing us to directly dial endpoints and
@@ -365,8 +361,8 @@ func (c *Client) dial(target string, creds grpccredentials.TransportCredentials,
return conn, nil
}
func (c *Client) directDialCreds(ep string) grpccredentials.TransportCredentials {
_, hostPort, scheme := endpoint.ParseEndpoint(ep)
func (c *Client) directDialCreds(ep string) (grpccredentials.TransportCredentials, error) {
_, host, scheme := endpoint.ParseEndpoint(ep)
creds := c.creds
if len(scheme) != 0 {
creds = c.processCreds(scheme)
@@ -375,12 +371,17 @@ func (c *Client) directDialCreds(ep string) grpccredentials.TransportCredentials
// Set the server name must to the endpoint hostname without port since grpc
// otherwise attempts to check if x509 cert is valid for the full endpoint
// including the scheme and port, which fails.
host, _ := endpoint.ParseHostPort(hostPort)
clone.OverrideServerName(host)
overrideServerName, _, err := net.SplitHostPort(host)
if err != nil {
// Either the host didn't have a port or the host could not be parsed. Either way, continue with the
// original host string.
overrideServerName = host
}
clone.OverrideServerName(overrideServerName)
creds = clone
}
}
return creds
return creds, nil
}
func (c *Client) dialWithBalancerCreds(ep string) grpccredentials.TransportCredentials {
@@ -659,3 +660,9 @@ func IsConnCanceled(err error) bool {
// <= gRPC v1.7.x returns 'errors.New("grpc: the client connection is closing")'
return strings.Contains(err.Error(), "grpc: the client connection is closing")
}
// TransportCredentialsWithDialer is for a gRPC load balancer workaround. See credentials.transportCredential for details.
type TransportCredentialsWithDialer interface {
grpccredentials.TransportCredentials
Dialer(ctx context.Context, dialEp string) (net.Conn, error)
}

View File

@@ -7,6 +7,7 @@ go_library(
importpath = "github.com/coreos/etcd/clientv3/credentials",
visibility = ["//visibility:public"],
deps = [
"//vendor/github.com/coreos/etcd/clientv3/balancer/resolver/endpoint:go_default_library",
"//vendor/github.com/coreos/etcd/etcdserver/api/v3rpc/rpctypes:go_default_library",
"//vendor/google.golang.org/grpc/credentials:go_default_library",
],

View File

@@ -22,6 +22,7 @@ import (
"net"
"sync"
"github.com/coreos/etcd/clientv3/balancer/resolver/endpoint"
"github.com/coreos/etcd/etcdserver/api/v3rpc/rpctypes"
grpccredentials "google.golang.org/grpc/credentials"
)
@@ -65,38 +66,37 @@ func (b *bundle) NewWithMode(mode string) (grpccredentials.Bundle, error) {
}
// transportCredential implements "grpccredentials.TransportCredentials" interface.
// transportCredential wraps TransportCredentials to track which
// addresses are dialed for which endpoints, and then sets the authority when checking the endpoint's cert to the
// hostname or IP of the dialed endpoint.
// This is a workaround of a gRPC load balancer issue. gRPC uses the dialed target's service name as the authority when
// checking all endpoint certs, which does not work for etcd servers using their hostname or IP as the Subject Alternative Name
// in their TLS certs.
// To enable, include both WithTransportCredentials(creds) and WithContextDialer(creds.Dialer)
// when dialing.
type transportCredential struct {
gtc grpccredentials.TransportCredentials
mu sync.Mutex
// addrToEndpoint maps from the connection addresses that are dialed to the hostname or IP of the
// endpoint provided to the dialer when dialing
addrToEndpoint map[string]string
}
func newTransportCredential(cfg *tls.Config) *transportCredential {
return &transportCredential{
gtc: grpccredentials.NewTLS(cfg),
gtc: grpccredentials.NewTLS(cfg),
addrToEndpoint: map[string]string{},
}
}
func (tc *transportCredential) ClientHandshake(ctx context.Context, authority string, rawConn net.Conn) (net.Conn, grpccredentials.AuthInfo, error) {
// Only overwrite when authority is an IP address!
// Let's say, a server runs SRV records on "etcd.local" that resolves
// to "m1.etcd.local", and its SAN field also includes "m1.etcd.local".
// But what if SAN does not include its resolved IP address (e.g. 127.0.0.1)?
// Then, the server should only authenticate using its DNS hostname "m1.etcd.local",
// instead of overwriting it with its IP address.
// And we do not overwrite "localhost" either. Only overwrite IP addresses!
if isIP(authority) {
target := rawConn.RemoteAddr().String()
if authority != target {
// When user dials with "grpc.WithDialer", "grpc.DialContext" "cc.parsedTarget"
// update only happens once. This is problematic, because when TLS is enabled,
// retries happen through "grpc.WithDialer" with static "cc.parsedTarget" from
// the initial dial call.
// If the server authenticates by IP addresses, we want to set a new endpoint as
// a new authority. Otherwise
// "transport: authentication handshake failed: x509: certificate is valid for 127.0.0.1, 192.168.121.180, not 192.168.223.156"
// when the new dial target is "192.168.121.180" whose certificate host name is also "192.168.121.180"
// but client tries to authenticate with previously set "cc.parsedTarget" field "192.168.223.156"
authority = target
}
// Set the authority when checking the endpoint's cert to the hostname or IP of the dialed endpoint
tc.mu.Lock()
dialEp, ok := tc.addrToEndpoint[rawConn.RemoteAddr().String()]
tc.mu.Unlock()
if ok {
_, host, _ := endpoint.ParseEndpoint(dialEp)
authority = host
}
return tc.gtc.ClientHandshake(ctx, authority, rawConn)
}
@@ -115,8 +115,15 @@ func (tc *transportCredential) Info() grpccredentials.ProtocolInfo {
}
func (tc *transportCredential) Clone() grpccredentials.TransportCredentials {
copy := map[string]string{}
tc.mu.Lock()
for k, v := range tc.addrToEndpoint {
copy[k] = v
}
tc.mu.Unlock()
return &transportCredential{
gtc: tc.gtc.Clone(),
gtc: tc.gtc.Clone(),
addrToEndpoint: copy,
}
}
@@ -124,6 +131,17 @@ func (tc *transportCredential) OverrideServerName(serverNameOverride string) err
return tc.gtc.OverrideServerName(serverNameOverride)
}
func (tc *transportCredential) Dialer(ctx context.Context, dialEp string) (net.Conn, error) {
// Keep track of which addresses are dialed for which endpoints
conn, err := endpoint.Dialer(ctx, dialEp)
if conn != nil {
tc.mu.Lock()
tc.addrToEndpoint[conn.RemoteAddr().String()] = dialEp
tc.mu.Unlock()
}
return conn, err
}
// perRPCCredential implements "grpccredentials.PerRPCCredentials" interface.
type perRPCCredential struct {
authToken string

View File

@@ -273,17 +273,17 @@ func NewConfig() *Config {
TickMs: 100,
ElectionMs: 1000,
InitialElectionTickAdvance: true,
LPUrls: []url.URL{*lpurl},
LCUrls: []url.URL{*lcurl},
APUrls: []url.URL{*apurl},
ACUrls: []url.URL{*acurl},
ClusterState: ClusterStateFlagNew,
InitialClusterToken: "etcd-cluster",
StrictReconfigCheck: DefaultStrictReconfigCheck,
LogOutput: DefaultLogOutput,
Metrics: "basic",
EnableV2: DefaultEnableV2,
AuthToken: "simple",
LPUrls: []url.URL{*lpurl},
LCUrls: []url.URL{*lcurl},
APUrls: []url.URL{*apurl},
ACUrls: []url.URL{*acurl},
ClusterState: ClusterStateFlagNew,
InitialClusterToken: "etcd-cluster",
StrictReconfigCheck: DefaultStrictReconfigCheck,
LogOutput: DefaultLogOutput,
Metrics: "basic",
EnableV2: DefaultEnableV2,
AuthToken: "simple",
}
cfg.InitialCluster = cfg.InitialClusterFromName(cfg.Name)
return cfg

View File

@@ -73,11 +73,11 @@ func handleV2(mux *http.ServeMux, server etcdserver.ServerV2, timeout time.Durat
}
mh := &membersHandler{
sec: sec,
server: server,
cluster: server.Cluster(),
timeout: timeout,
clock: clockwork.NewRealClock(),
sec: sec,
server: server,
cluster: server.Cluster(),
timeout: timeout,
clock: clockwork.NewRealClock(),
clientCertAuthEnabled: server.ClientCertAuthEnabled(),
}

View File

@@ -292,7 +292,7 @@ func (b *backend) Defrag() error {
func (b *backend) defrag() error {
now := time.Now()
// TODO: make this non-blocking?
// lock batchTx to ensure nobody is using previous tx, and then
// close previous ongoing tx.

View File

@@ -309,14 +309,7 @@ func (s *store) Restore(b backend.Backend) error {
}
func (s *store) restore() error {
b := s.b
reportDbTotalSizeInBytesMu.Lock()
reportDbTotalSizeInBytes = func() float64 { return float64(b.Size()) }
reportDbTotalSizeInBytesMu.Unlock()
reportDbTotalSizeInUseInBytesMu.Lock()
reportDbTotalSizeInUseInBytes = func() float64 { return float64(b.SizeInUse()) }
reportDbTotalSizeInUseInBytesMu.Unlock()
s.setupMetricsReporter()
min, max := newRevBytes(), newRevBytes()
revToBytes(revision{main: 1}, min)
@@ -496,6 +489,30 @@ func (s *store) ConsistentIndex() uint64 {
return v
}
func (s *store) setupMetricsReporter() {
b := s.b
reportDbTotalSizeInBytesMu.Lock()
reportDbTotalSizeInBytes = func() float64 { return float64(b.Size()) }
reportDbTotalSizeInBytesMu.Unlock()
reportDbTotalSizeInUseInBytesMu.Lock()
reportDbTotalSizeInUseInBytes = func() float64 { return float64(b.SizeInUse()) }
reportDbTotalSizeInUseInBytesMu.Unlock()
reportCurrentRevMu.Lock()
reportCurrentRev = func() float64 {
s.revMu.RLock()
defer s.revMu.RUnlock()
return float64(s.currentRev)
}
reportCurrentRevMu.Unlock()
reportCompactRevMu.Lock()
reportCompactRev = func() float64 {
s.revMu.RLock()
defer s.revMu.RUnlock()
return float64(s.compactMainRev)
}
reportCompactRevMu.Unlock()
}
// appendMarkTombstone appends tombstone mark to normal revision bytes.
func appendMarkTombstone(b []byte) []byte {
if len(b) != revBytesLen {

View File

@@ -206,6 +206,38 @@ var (
// highest bucket start of 0.01 sec * 2^14 == 163.84 sec
Buckets: prometheus.ExponentialBuckets(.01, 2, 15),
})
currentRev = prometheus.NewGaugeFunc(prometheus.GaugeOpts{
Namespace: "etcd_debugging",
Subsystem: "mvcc",
Name: "current_revision",
Help: "The current revision of store.",
},
func() float64 {
reportCurrentRevMu.RLock()
defer reportCurrentRevMu.RUnlock()
return reportCurrentRev()
},
)
// overridden by mvcc initialization
reportCurrentRevMu sync.RWMutex
reportCurrentRev = func() float64 { return 0 }
compactRev = prometheus.NewGaugeFunc(prometheus.GaugeOpts{
Namespace: "etcd_debugging",
Subsystem: "mvcc",
Name: "compact_revision",
Help: "The revision of the last compaction in store.",
},
func() float64 {
reportCompactRevMu.RLock()
defer reportCompactRevMu.RUnlock()
return reportCompactRev()
},
)
// overridden by mvcc initialization
reportCompactRevMu sync.RWMutex
reportCompactRev = func() float64 { return 0 }
)
func init() {
@@ -228,6 +260,8 @@ func init() {
prometheus.MustRegister(dbTotalSizeInUse)
prometheus.MustRegister(hashDurations)
prometheus.MustRegister(hashRevDurations)
prometheus.MustRegister(currentRev)
prometheus.MustRegister(compactRev)
}
// ReportEventReceived reports that an event is received.

View File

@@ -70,13 +70,13 @@ func CheckAfterTest(d time.Duration) error {
}
var bad string
badSubstring := map[string]string{
").writeLoop(": "a Transport",
").writeLoop(": "a Transport",
"created by net/http/httptest.(*Server).Start": "an httptest.Server",
"timeoutHandler": "a TimeoutHandler",
"net.(*netFD).connect(": "a timing out dial",
").noteClientGone(": "a closenotifier sender",
").readLoop(": "a Transport",
".grpc": "a gRPC resource",
"timeoutHandler": "a TimeoutHandler",
"net.(*netFD).connect(": "a timing out dial",
").noteClientGone(": "a closenotifier sender",
").readLoop(": "a Transport",
".grpc": "a gRPC resource",
}
var stacks string

View File

@@ -53,6 +53,9 @@ func wrapTLS(addr, scheme string, tlsinfo *TLSInfo, l net.Listener) (net.Listene
if scheme != "https" && scheme != "unixs" {
return l, nil
}
if tlsinfo != nil && tlsinfo.SkipClientSANVerify {
return NewTLSListener(l, tlsinfo)
}
return newTLSListener(l, tlsinfo, checkSAN)
}
@@ -65,6 +68,8 @@ type TLSInfo struct {
CRLFile string
InsecureSkipVerify bool
SkipClientSANVerify bool
// ServerName ensures the cert matches the given host in case of discovery / virtual hosting
ServerName string
@@ -95,7 +100,7 @@ func (info TLSInfo) Empty() bool {
return info.CertFile == "" && info.KeyFile == ""
}
func SelfCert(dirpath string, hosts []string) (info TLSInfo, err error) {
func SelfCert(dirpath string, hosts []string, additionalUsages ...x509.ExtKeyUsage) (info TLSInfo, err error) {
if err = os.MkdirAll(dirpath, 0700); err != nil {
return
}
@@ -124,7 +129,7 @@ func SelfCert(dirpath string, hosts []string) (info TLSInfo, err error) {
NotAfter: time.Now().Add(365 * (24 * time.Hour)),
KeyUsage: x509.KeyUsageKeyEncipherment | x509.KeyUsageDigitalSignature,
ExtKeyUsage: []x509.ExtKeyUsage{x509.ExtKeyUsageServerAuth},
ExtKeyUsage: append([]x509.ExtKeyUsage{x509.ExtKeyUsageServerAuth}, additionalUsages...),
BasicConstraintsValid: true,
}

View File

@@ -26,7 +26,7 @@ import (
var (
// MinClusterVersion is the min cluster version this etcd binary is compatible with.
MinClusterVersion = "3.0.0"
Version = "3.3.15"
Version = "3.3.17"
APIVersion = "unknown"
// Git SHA Value will be set during build