agent: prevent very old servers re-joining a cluster with stale data (#…

…17171) * agent: configure server lastseen timestamp Signed-off-by: Dan Bond <danbond@protonmail.com> * use correct config Signed-off-by: Dan Bond <danbond@protonmail.com> * add comments Signed-off-by: Dan Bond <danbond@protonmail.com> * use default age in test golden data Signed-off-by: Dan Bond <danbond@protonmail.com> * add changelog Signed-off-by: Dan Bond <danbond@protonmail.com> * fix runtime test Signed-off-by: Dan Bond <danbond@protonmail.com> * agent: add server_metadata Signed-off-by: Dan Bond <danbond@protonmail.com> * update comments Signed-off-by: Dan Bond <danbond@protonmail.com> * correctly check if metadata file does not exist Signed-off-by: Dan Bond <danbond@protonmail.com> * follow instructions for adding new config Signed-off-by: Dan Bond <danbond@protonmail.com> * add comments Signed-off-by: Dan Bond <danbond@protonmail.com> * update comments Signed-off-by: Dan Bond <danbond@protonmail.com> * Update agent/agent.go Co-authored-by: Dan Upton <daniel@floppy.co> * agent/config: add validation for duration with min Signed-off-by: Dan Bond <danbond@protonmail.com> * docs: add new server_rejoin_age_max config definition Signed-off-by: Dan Bond <danbond@protonmail.com> * agent: add unit test for checking server last seen Signed-off-by: Dan Bond <danbond@protonmail.com> * agent: log continually for 60s before erroring Signed-off-by: Dan Bond <danbond@protonmail.com> * pr comments Signed-off-by: Dan Bond <danbond@protonmail.com> * remove unneeded todo * agent: fix error message Signed-off-by: Dan Bond <danbond@protonmail.com> --------- Signed-off-by: Dan Bond <danbond@protonmail.com> Co-authored-by: Dan Upton <daniel@floppy.co>
hashicorp · May 15, 2023 · 95f462d · 95f462d
1 parent f334fcc
commit 95f462d
Show file tree

Hide file tree

Showing 16 changed files with 385 additions and 46 deletions.
diff --git a/.changelog/17171.txt b/.changelog/17171.txt
@@ -0,0 +1,3 @@
+```release-note:improvement
+agent: add a configurable maximimum age (default: 7 days) to prevent servers re-joining a cluster with stale data 
+```
diff --git a/agent/agent.go b/agent/agent.go
@@ -7,6 +7,7 @@ import (
 	"context"
 	"crypto/tls"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"io"
 	"net"
@@ -22,8 +23,6 @@ import (
 
 	"github.com/armon/go-metrics"
 	"github.com/armon/go-metrics/prometheus"
-	"github.com/hashicorp/consul/agent/rpcclient"
-	"github.com/hashicorp/consul/agent/rpcclient/configentry"
 	"github.com/hashicorp/go-connlimit"
 	"github.com/hashicorp/go-hclog"
 	"github.com/hashicorp/go-memdb"
@@ -50,12 +49,13 @@ import (
 	grpcDNS "github.com/hashicorp/consul/agent/grpc-external/services/dns"
 	middleware "github.com/hashicorp/consul/agent/grpc-middleware"
 	"github.com/hashicorp/consul/agent/hcp/scada"
-	libscada "github.com/hashicorp/consul/agent/hcp/scada"
 	"github.com/hashicorp/consul/agent/local"
 	"github.com/hashicorp/consul/agent/proxycfg"
 	proxycfgglue "github.com/hashicorp/consul/agent/proxycfg-glue"
 	catalogproxycfg "github.com/hashicorp/consul/agent/proxycfg-sources/catalog"
 	localproxycfg "github.com/hashicorp/consul/agent/proxycfg-sources/local"
+	"github.com/hashicorp/consul/agent/rpcclient"
+	"github.com/hashicorp/consul/agent/rpcclient/configentry"
 	"github.com/hashicorp/consul/agent/rpcclient/health"
 	"github.com/hashicorp/consul/agent/structs"
 	"github.com/hashicorp/consul/agent/systemd"
@@ -575,11 +575,11 @@ func (a *Agent) Start(ctx context.Context) error {
 		return err
 	}
 
-	// copy over the existing node id, this cannot be
-	// changed while running anyways but this prevents
-	// breaking some existing behavior. then overwrite
-	// the configuration
+	// Copy over the existing node id. This cannot be
+	// changed while running, but this prevents
+	// breaking some existing behavior.
 	c.NodeID = a.config.NodeID
+	// Overwrite the configuration.
 	a.config = c
 
 	if err := a.tlsConfigurator.Update(a.config.TLS); err != nil {
@@ -625,6 +625,20 @@ func (a *Agent) Start(ctx context.Context) error {
 	if c.ServerMode {
 		serverLogger := a.baseDeps.Logger.NamedIntercept(logging.ConsulServer)
 
+		// Check for a last seen timestamp and exit if deemed stale before attempting to join
+		// Serf/Raft or listen for requests.
+		if err := a.checkServerLastSeen(consul.ReadServerMetadata); err != nil {
+			deadline := time.Now().Add(time.Minute)
+			for time.Now().Before(deadline) {
+				a.logger.Error("startup error", "error", err)
+				time.Sleep(10 * time.Second)
+			}
+			return err
+		}
+
+		// periodically write server metadata to disk.
+		go a.persistServerMetadata()
+
 		incomingRPCLimiter := consul.ConfiguredIncomingRPCLimiter(
 			&lib.StopChannelContext{StopCh: a.shutdownCh},
 			serverLogger,
@@ -661,7 +675,6 @@ func (a *Agent) Start(ctx context.Context) error {
 				return fmt.Errorf("failed to start server cert manager: %w", err)
 			}
 		}
-
 	} else {
 		a.externalGRPCServer = external.NewServer(
 			a.logger.Named("grpc.external"),
@@ -1094,7 +1107,7 @@ func (a *Agent) listenHTTP() ([]apiServer, error) {
 				MaxHeaderBytes: a.config.HTTPMaxHeaderBytes,
 			}
 
-			if libscada.IsCapability(l.Addr()) {
+			if scada.IsCapability(l.Addr()) {
 				// wrap in http2 server handler
 				httpServer.Handler = h2c.NewHandler(srv.handler(a.config.EnableDebug), &http2.Server{})
 			}
@@ -1521,6 +1534,8 @@ func newConsulConfig(runtimeCfg *config.RuntimeConfig, logger hclog.Logger) (*co
 
 	cfg.Reporting.License.Enabled = runtimeCfg.Reporting.License.Enabled
 
+	cfg.ServerRejoinAgeMax = runtimeCfg.ServerRejoinAgeMax
+
 	enterpriseConsulConfig(cfg, runtimeCfg)
 
 	return cfg, nil
@@ -4529,7 +4544,70 @@ func (a *Agent) proxyDataSources() proxycfg.DataSources {
 
 	a.fillEnterpriseProxyDataSources(&sources)
 	return sources
+}
+
+// persistServerMetadata periodically writes a server's metadata to a file
+// in the configured data directory.
+func (a *Agent) persistServerMetadata() {
+	file := filepath.Join(a.config.DataDir, consul.ServerMetadataFile)
+
+	// Create a timer with no initial tick to allow metadata to be written immediately.
+	t := time.NewTimer(0)
+	defer t.Stop()
+
+	for {
+		select {
+		case <-t.C:
+			// Reset the timer to the larger periodic interval.
+			t.Reset(1 * time.Hour)
+
+			f, err := consul.OpenServerMetadata(file)
+			if err != nil {
+				a.logger.Error("failed to open existing server metadata: %w", err)
+				continue
+			}
+
+			if err := consul.WriteServerMetadata(f); err != nil {
+				f.Close()
+				a.logger.Error("failed to write server metadata: %w", err)
+				continue
+			}
+
+			f.Close()
+		case <-a.shutdownCh:
+			return
+		}
+	}
+}
 
+// checkServerLastSeen is a safety check that only occurs once of startup to prevent old servers
+// with stale data from rejoining an existing cluster.
+//
+// It attempts to read a server's metadata file and check the last seen Unix timestamp against a
+// configurable max age. If the metadata file does not exist, we treat this as an initial startup
+// and return no error.
+//
+// Example: if the server recorded a last seen timestamp of now-7d, and we configure a max age
+// of 3d, then we should prevent the server from rejoining.
+func (a *Agent) checkServerLastSeen(readFn consul.ServerMetadataReadFunc) error {
+	filename := filepath.Join(a.config.DataDir, consul.ServerMetadataFile)
+
+	// Read server metadata file.
+	md, err := readFn(filename)
+	if err != nil {
+		// Return early if it doesn't exist as this likely indicates the server is starting for the first time.
+		if errors.Is(err, os.ErrNotExist) {
+			return nil
+		}
+		return fmt.Errorf("error reading server metadata: %w", err)
+	}
+
+	maxAge := a.config.ServerRejoinAgeMax
+	if md.IsLastSeenStale(maxAge) {
+		return fmt.Errorf("refusing to rejoin cluster because server has been offline for more than the configured server_rejoin_age_max (%s) - consider wiping your data dir", maxAge)
+	}
+
+	return nil
 }
 
 func listenerPortKey(svcID structs.ServiceID, checkID structs.CheckID) string {

diff --git a/agent/agent_test.go b/agent/agent_test.go
@@ -12,6 +12,7 @@ import (
 	"crypto/x509"
 	"encoding/base64"
 	"encoding/json"
+	"errors"
 	"fmt"
 	mathrand "math/rand"
 	"net"
@@ -6204,6 +6205,70 @@ cloud {
 	require.NoError(t, err)
 }
 
+func TestAgent_checkServerLastSeen(t *testing.T) {
+	bd := BaseDeps{
+		Deps: consul.Deps{
+			Logger:       hclog.NewInterceptLogger(nil),
+			Tokens:       new(token.Store),
+			GRPCConnPool: &fakeGRPCConnPool{},
+		},
+		RuntimeConfig: &config.RuntimeConfig{},
+		Cache:         cache.New(cache.Options{}),
+	}
+	agent, err := New(bd)
+	require.NoError(t, err)
+
+	// Test that an ErrNotExist OS error is treated as ok.
+	t.Run("TestReadErrNotExist", func(t *testing.T) {
+		readFn := func(filename string) (*consul.ServerMetadata, error) {
+			return nil, os.ErrNotExist
+		}
+
+		err := agent.checkServerLastSeen(readFn)
+		require.NoError(t, err)
+	})
+
+	// Test that an error reading server metadata is treated as an error.
+	t.Run("TestReadErr", func(t *testing.T) {
+		expected := errors.New("read error")
+		readFn := func(filename string) (*consul.ServerMetadata, error) {
+			return nil, expected
+		}
+
+		err := agent.checkServerLastSeen(readFn)
+		require.ErrorIs(t, err, expected)
+	})
+
+	// Test that a server with a 7d old last seen timestamp is treated as an error.
+	t.Run("TestIsLastSeenStaleErr", func(t *testing.T) {
+		agent.config.ServerRejoinAgeMax = time.Hour
+
+		readFn := func(filename string) (*consul.ServerMetadata, error) {
+			return &consul.ServerMetadata{
+				LastSeenUnix: time.Now().Add(-24 * 7 * time.Hour).Unix(),
+			}, nil
+		}
+
+		err := agent.checkServerLastSeen(readFn)
+		require.Error(t, err)
+		require.ErrorContains(t, err, "refusing to rejoin cluster because server has been offline for more than the configured server_rejoin_age_max")
+	})
+
+	// Test that a server with a 6h old last seen timestamp is not treated as an error.
+	t.Run("TestNoErr", func(t *testing.T) {
+		agent.config.ServerRejoinAgeMax = 24 * 7 * time.Hour
+
+		readFn := func(filename string) (*consul.ServerMetadata, error) {
+			return &consul.ServerMetadata{
+				LastSeenUnix: time.Now().Add(-6 * time.Hour).Unix(),
+			}, nil
+		}
+
+		err := agent.checkServerLastSeen(readFn)
+		require.NoError(t, err)
+	})
+}
+
 func getExpectedCaPoolByFile(t *testing.T) *x509.CertPool {
 	pool := x509.NewCertPool()
 	data, err := os.ReadFile("../test/ca/root.cer")

diff --git a/agent/config/builder.go b/agent/config/builder.go
@@ -28,15 +28,14 @@ import (
 	"github.com/hashicorp/memberlist"
 	"golang.org/x/time/rate"
 
-	hcpconfig "github.com/hashicorp/consul/agent/hcp/config"
-
 	"github.com/hashicorp/consul/agent/cache"
 	"github.com/hashicorp/consul/agent/checks"
 	"github.com/hashicorp/consul/agent/connect/ca"
 	"github.com/hashicorp/consul/agent/consul"
 	"github.com/hashicorp/consul/agent/consul/authmethod/ssoauth"
 	consulrate "github.com/hashicorp/consul/agent/consul/rate"
 	"github.com/hashicorp/consul/agent/dns"
+	hcpconfig "github.com/hashicorp/consul/agent/hcp/config"
 	"github.com/hashicorp/consul/agent/rpc/middleware"
 	"github.com/hashicorp/consul/agent/structs"
 	"github.com/hashicorp/consul/agent/token"
@@ -1090,6 +1089,7 @@ func (b *builder) build() (rt RuntimeConfig, err error) {
 		ServerMode:                        serverMode,
 		ServerName:                        stringVal(c.ServerName),
 		ServerPort:                        serverPort,
+		ServerRejoinAgeMax:                b.durationValWithDefaultMin("server_rejoin_age_max", c.ServerRejoinAgeMax, 24*7*time.Hour, 6*time.Hour),
 		Services:                          services,
 		SessionTTLMin:                     b.durationVal("session_ttl_min", c.SessionTTLMin),
 		SkipLeaveOnInt:                    skipLeaveOnInt,
@@ -1952,6 +1952,16 @@ func (b *builder) durationValWithDefault(name string, v *string, defaultVal time
 	return d
 }
 
+// durationValWithDefaultMin is equivalent to durationValWithDefault, but enforces a minimum duration.
+func (b *builder) durationValWithDefaultMin(name string, v *string, defaultVal, minVal time.Duration) (d time.Duration) {
+	d = b.durationValWithDefault(name, v, defaultVal)
+	if d < minVal {
+		b.err = multierror.Append(b.err, fmt.Errorf("%s: duration '%s' cannot be less than: %s", name, *v, minVal))
+	}
+
+	return d
+}
+
 func (b *builder) durationVal(name string, v *string) (d time.Duration) {
 	return b.durationValWithDefault(name, v, 0)
 }

diff --git a/agent/config/builder_test.go b/agent/config/builder_test.go
@@ -311,6 +311,21 @@ func TestBuilder_DurationVal_InvalidDuration(t *testing.T) {
 	require.Contains(t, b.err.Error(), badDuration2)
 }
 
+func TestBuilder_DurationValWithDefaultMin(t *testing.T) {
+	b := builder{}
+
+	// Attempt to validate that a duration of 10 hours will not error when the min val is 1 hour.
+	dur := "10h0m0s"
+	b.durationValWithDefaultMin("field2", &dur, 24*7*time.Hour, time.Hour)
+	require.NoError(t, b.err)
+
+	// Attempt to validate that a duration of 1 min will error when the min val is 1 hour.
+	dur = "0h1m0s"
+	b.durationValWithDefaultMin("field1", &dur, 24*7*time.Hour, time.Hour)
+	require.Error(t, b.err)
+	require.Contains(t, b.err.Error(), "1 error")
+}
+
 func TestBuilder_ServiceVal_MultiError(t *testing.T) {
 	b := builder{}
 	b.serviceVal(&ServiceDefinition{

diff --git a/agent/config/config.go b/agent/config/config.go
@@ -228,6 +228,7 @@ type Config struct {
 	SerfBindAddrWAN                  *string             `mapstructure:"serf_wan" json:"serf_wan,omitempty"`
 	ServerMode                       *bool               `mapstructure:"server" json:"server,omitempty"`
 	ServerName                       *string             `mapstructure:"server_name" json:"server_name,omitempty"`
+	ServerRejoinAgeMax               *string             `mapstructure:"server_rejoin_age_max" json:"server_rejoin_age_max,omitempty"`
 	Service                          *ServiceDefinition  `mapstructure:"service" json:"-"`
 	Services                         []ServiceDefinition `mapstructure:"services" json:"-"`
 	SessionTTLMin                    *string             `mapstructure:"session_ttl_min" json:"session_ttl_min,omitempty"`

diff --git a/agent/config/default.go b/agent/config/default.go
@@ -58,6 +58,7 @@ func DefaultSource() Source {
 		segment_limit = 64
 
 		server = false
+		server_rejoin_age_max = "168h"
 		syslog_facility = "LOCAL0"
 
 		tls = {

diff --git a/agent/config/runtime.go b/agent/config/runtime.go
@@ -1358,6 +1358,18 @@ type RuntimeConfig struct {
 	// hcl: ports { server = int }
 	ServerPort int
 
+	// ServerRejoinAgeMax is used to specify the duration of time a server
+	// is allowed to be down/offline before a startup operation is refused.
+	//
+	// For example: if a server has been offline for 5 days, and this option
+	// is configured to 3 days, then any subsequent startup operation will fail
+	// and require an operator to manually intervene.
+	//
+	// The default is: 7 days
+	//
+	// hcl: server_rejoin_age_max = "duration"
+	ServerRejoinAgeMax time.Duration
+
 	// Services contains the provided service definitions:
 	//
 	// hcl: services = [

diff --git a/agent/config/runtime_test.go b/agent/config/runtime_test.go
@@ -25,13 +25,12 @@ import (
 	"github.com/stretchr/testify/require"
 	"golang.org/x/time/rate"
 
-	hcpconfig "github.com/hashicorp/consul/agent/hcp/config"
-
 	"github.com/hashicorp/consul/acl"
 	"github.com/hashicorp/consul/agent/cache"
 	"github.com/hashicorp/consul/agent/checks"
 	"github.com/hashicorp/consul/agent/consul"
 	consulrate "github.com/hashicorp/consul/agent/consul/rate"
+	hcpconfig "github.com/hashicorp/consul/agent/hcp/config"
 	"github.com/hashicorp/consul/agent/structs"
 	"github.com/hashicorp/consul/agent/token"
 	"github.com/hashicorp/consul/lib"
@@ -6419,6 +6418,7 @@ func TestLoad_FullConfig(t *testing.T) {
 		SerfPortWAN:             8302,
 		ServerMode:              true,
 		ServerName:              "Oerr9n1G",
+		ServerRejoinAgeMax:      604800 * time.Second,
 		ServerPort:              3757,
 		Services: []*structs.ServiceDefinition{
 			{
@@ -7163,7 +7163,8 @@ func TestRuntimeConfig_Sanitize(t *testing.T) {
 				},
 			},
 		},
-		Locality: &Locality{Region: strPtr("us-west-1"), Zone: strPtr("us-west-1a")},
+		Locality:           &Locality{Region: strPtr("us-west-1"), Zone: strPtr("us-west-1a")},
+		ServerRejoinAgeMax: 24 * 7 * time.Hour,
 	}
 
 	b, err := json.MarshalIndent(rt.Sanitized(), "", "    ")

diff --git a/agent/config/testdata/TestRuntimeConfig_Sanitize.golden b/agent/config/testdata/TestRuntimeConfig_Sanitize.golden
@@ -332,6 +332,7 @@
     "ServerMode": false,
     "ServerName": "",
     "ServerPort": 0,
+    "ServerRejoinAgeMax": "168h0m0s",
     "Services": [
         {
             "Address": "",

diff --git a/agent/config/testdata/full-config.hcl b/agent/config/testdata/full-config.hcl
@@ -394,6 +394,7 @@ serf_lan = "99.43.63.15"
 serf_wan = "67.88.33.19"
 server = true
 server_name = "Oerr9n1G"
+server_rejoin_age_max = "604800s"
 service = {
     id = "dLOXpSCI"
     name = "o1ynPkp0"