Extend TestSmeshing to also deploy an older version of go-spacmesh (#6403)

## Motivation

To detect errors that would break compatibility with previous versions of the node I extended `TestSmeshing` to deploy 25% of the nodes in the test using an older version of go-spacemesh. So if any change changes protocol rules that would break consensus like the error fixed in #6398 or the error fixed in #5839.

Author: fasmat

Dockerfile

@@ -1,7 +1,7 @@
 # go-spacemesh needs at least ubuntu 22.04. newer versions of ubuntu might work as well, but are untested
 FROM ubuntu:22.04 AS linux
-ENV DEBIAN_FRONTEND noninteractive
-ENV SHELL /bin/bash
+ENV DEBIAN_FRONTEND=noninteractive
+ENV SHELL=/bin/bash
 ARG TZ=Etc/UTC
 ENV TZ=${TZ}
 USER root
@@ -22,11 +22,11 @@ RUN set -ex \
    && locale-gen en_US.UTF-8 \
    && update-locale LANG=en_US.UTF-8 \
    && echo "$TZ" > /etc/timezone
-ENV LANG en_US.UTF-8
-ENV LANGUAGE en_US.UTF-8
-ENV LC_ALL en_US.UTF-8
+ENV LANG=en_US.UTF-8
+ENV LANGUAGE=en_US.UTF-8
+ENV LC_ALL=en_US.UTF-8
 
-FROM golang:1.23 as builder
+FROM golang:1.23 AS builder
 ARG VERSION=""
 ENV VERSION=${VERSION}
 RUN set -ex \

bootstrap.Dockerfile

@@ -1,4 +1,4 @@
-FROM golang:1.23 as builder
+FROM golang:1.23 AS builder
 
 WORKDIR /src
 

common/types/block.go

@@ -89,7 +89,7 @@ type InnerBlock struct {
 }
 
 // RatNum represents a rational number with the numerator and denominator.
-// note: RatNum aims to be a generic representation of a rational number and parse-able by
+// note: RatNum aims to be a generic representation of a rational number and parseable by
 // different programming languages.
 // For doing math around weight inside go-spacemesh codebase, use util.Weight.
 type RatNum struct {

config/config.go

@@ -3,6 +3,7 @@ package config
 
 import (
 	"fmt"
+	"io"
 	"math"
 	"os"
 	"path/filepath"
@@ -276,13 +277,13 @@ func defaultTestConfig() BaseConfig {
 }
 
 // LoadConfig load the config file.
-func LoadConfig(config string, vip *viper.Viper) error {
-	if config == "" {
+func LoadConfig(src io.Reader, vip *viper.Viper) error {
+	if src == nil {
 		return nil
 	}
-	vip.SetConfigFile(config)
-	if err := vip.ReadInConfig(); err != nil {
-		return fmt.Errorf("can't load config at %s: %w", config, err)
+	vip.SetConfigType("json")
+	if err := vip.ReadConfig(src); err != nil {
+		return fmt.Errorf("can't load config: %w", err)
 	}
 	return nil
 }

config/presets/fastnet.go

@@ -26,7 +26,7 @@ func fastnet() config.Config {
 	// set for systest TestEquivocation
 	conf.BaseConfig.MinerGoodAtxsPercent = 50
 
-	// node will select atxs that were received atleast 4 seconds before start of the epoch
+	// node will select atxs that were received at least 4 seconds before start of the epoch
 	// for activeset.
 	// if some atxs weren't received on time it will skew eligibility distribution
 	// and will make some tests fail.

node/node.go

@@ -6,6 +6,7 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"io"
 	"io/fs"
 	"net"
 	"net/http"
@@ -244,31 +245,39 @@ func GetCommand() *cobra.Command {
 }
 
 func configure(c *cobra.Command, configPath string, conf *config.Config) error {
-	preset := conf.Preset // might be set via CLI flag
-	if err := loadConfig(conf, preset, configPath); err != nil {
+	f, err := os.Open(configPath)
+	if err != nil {
+		return fmt.Errorf("opening config file: %w", err)
+	}
+	defer f.Close()
+	if err := LoadConfig(conf, conf.Preset, f); err != nil {
 		return fmt.Errorf("loading config: %w", err)
 	}
+	if err := f.Close(); err != nil {
+		return fmt.Errorf("closing config file: %w", err)
+	}
 	// apply CLI args to config
 	if err := c.ParseFlags(os.Args[1:]); err != nil {
 		return fmt.Errorf("parsing flags: %w", err)
 	}
-
 	if cmd.NoMainNet && onMainNet(conf) && !conf.NoMainOverride {
 		return errors.New("this is a testnet-only build not intended for mainnet")
 	}
-
 	return nil
 }
 
 var grpcLog = grpc_logsettable.ReplaceGrpcLoggerV2()
 
-// loadConfig loads config and preset (if provided) into the provided config.
+// LoadConfig loads config and preset (if provided) into the provided config.
 // It first loads the preset and then overrides it with values from the config file.
-func loadConfig(cfg *config.Config, preset, path string) error {
+func LoadConfig(cfg *config.Config, preset string, src io.Reader) error {
 	v := viper.New()
-	// read in config from file
-	if err := config.LoadConfig(path, v); err != nil {
-		return err
+	// read in config from src
+	if src != nil {
+		v.SetConfigType("json")
+		if err := v.ReadConfig(src); err != nil {
+			return fmt.Errorf("can't load config: %w", err)
+		}
 	}
 
 	// override default config with preset if provided
@@ -283,7 +292,7 @@ func loadConfig(cfg *config.Config, preset, path string) error {
 		*cfg = p
 	}
 
-	// Unmarshall config file into config struct
+	// Unmarshal config file into config struct
 	hook := mapstructure.ComposeDecodeHookFunc(
 		mapstructure.StringToTimeDurationHookFunc(),
 		mapstructure.StringToSliceHookFunc(","),
@@ -294,15 +303,12 @@ func loadConfig(cfg *config.Config, preset, path string) error {
 		mapstructureutil.AtxVersionsDecodeFunc(),
 		mapstructure.TextUnmarshallerHookFunc(),
 	)
-
 	opts := []viper.DecoderConfigOption{
 		viper.DecodeHook(hook),
 		WithZeroFields(),
 		WithIgnoreUntagged(),
 		WithErrorUnused(),
 	}
-
-	// load config if it was loaded to the viper
 	if err := v.Unmarshal(cfg, opts...); err != nil {
 		return fmt.Errorf("unmarshal config: %w", err)
 	}

node/node_test.go

@@ -139,10 +139,10 @@ func TestSpacemeshApp_AddLogger(t *testing.T) {
 	myLogger := "anton"
 	subLogger := app.addLogger(myLogger, lg)
 	subLogger.Debug("should not get printed")
-	teststr := "should get printed"
-	subLogger.Info(teststr)
+	testStr := "should get printed"
+	subLogger.Info(testStr)
 	r.Equal(
-		fmt.Sprintf("INFO\t%s\t%s\n", myLogger, teststr),
+		fmt.Sprintf("INFO\t%s\t%s\n", myLogger, testStr),
 		buf.String(),
 	)
 }
@@ -339,8 +339,8 @@ func (f *noopHook) OnWrite(*zapcore.CheckedEntry, []zapcore.Field) {}
 // E2E app test of the stream endpoints in the NodeService.
 func TestSpacemeshApp_NodeService(t *testing.T) {
 	logger := logtest.New(t)
-	// errlog is used to simulate errors in the app
-	errlog := log.NewFromLog(
+	// errLog is used to simulate errors in the app
+	errLog := log.NewFromLog(
 		zaptest.NewLogger(t, zaptest.WrapOptions(zap.Hooks(events.EventHook()), zap.WithPanicHook(&noopHook{}))),
 	)
 
@@ -430,9 +430,9 @@ func TestSpacemeshApp_NodeService(t *testing.T) {
 
 	// Report two errors and make sure they're both received
 	eg.Go(func() error {
-		errlog.Error("test123")
-		errlog.Error("test456")
-		errlog.Panic("testPANIC")
+		errLog.Error("test123")
+		errLog.Error("test456")
+		errLog.Panic("testPANIC")
 		return nil
 	})
 
@@ -602,7 +602,7 @@ func TestConfig_Preset(t *testing.T) {
 		require.NoError(t, err)
 
 		conf := config.Config{}
-		require.NoError(t, loadConfig(&conf, name, ""))
+		require.NoError(t, LoadConfig(&conf, name, nil))
 		require.Equal(t, preset, conf)
 	})
 
@@ -615,7 +615,7 @@ func TestConfig_Preset(t *testing.T) {
 		cmd.AddFlags(&flags, &conf)
 
 		const lowPeers = 1234
-		require.NoError(t, loadConfig(&conf, name, ""))
+		require.NoError(t, LoadConfig(&conf, name, nil))
 		require.NoError(t, flags.Parse([]string{"--low-peers=" + strconv.Itoa(lowPeers)}))
 		preset.P2P.LowPeers = lowPeers
 		require.Equal(t, preset, conf)
@@ -628,10 +628,8 @@ func TestConfig_Preset(t *testing.T) {
 		conf := config.Config{}
 		const lowPeers = 1234
 		content := fmt.Sprintf(`{"p2p": {"low-peers": %d}}`, lowPeers)
-		path := filepath.Join(t.TempDir(), "config.json")
-		require.NoError(t, os.WriteFile(path, []byte(content), 0o600))
 
-		require.NoError(t, loadConfig(&conf, name, path))
+		require.NoError(t, LoadConfig(&conf, name, strings.NewReader(content)))
 		preset.P2P.LowPeers = lowPeers
 		require.Equal(t, preset, conf)
 	})
@@ -642,10 +640,8 @@ func TestConfig_Preset(t *testing.T) {
 
 		conf := config.Config{}
 		content := fmt.Sprintf(`{"preset": "%s"}`, name)
-		path := filepath.Join(t.TempDir(), "config.json")
-		require.NoError(t, os.WriteFile(path, []byte(content), 0o600))
 
-		require.NoError(t, loadConfig(&conf, name, path))
+		require.NoError(t, LoadConfig(&conf, name, strings.NewReader(content)))
 		require.NoError(t, err)
 		require.Equal(t, preset, conf)
 	})
@@ -719,7 +715,7 @@ func TestConfig_CustomTypes(t *testing.T) {
 			var flags pflag.FlagSet
 			cmd.AddFlags(&flags, &conf)
 
-			require.NoError(t, loadConfig(&conf, "", ""))
+			require.NoError(t, LoadConfig(&conf, "", nil))
 			require.NoError(t, flags.Parse(strings.Fields(tc.cli)))
 			tc.updatePreset(t, &mainnet)
 			require.Equal(t, mainnet, conf)
@@ -730,10 +726,8 @@ func TestConfig_CustomTypes(t *testing.T) {
 			require.Nil(t, mainnet.SMESHING.Opts.ProviderID.Value())
 
 			conf := config.MainnetConfig()
-			path := filepath.Join(t.TempDir(), "config.json")
-			require.NoError(t, os.WriteFile(path, []byte(tc.config), 0o600))
 
-			require.NoError(t, loadConfig(&conf, "", path))
+			require.NoError(t, LoadConfig(&conf, "", strings.NewReader(tc.config)))
 			tc.updatePreset(t, &mainnet)
 			require.Equal(t, mainnet, conf)
 		})
@@ -746,7 +740,7 @@ func TestConfig_CustomTypes(t *testing.T) {
 			var flags pflag.FlagSet
 			cmd.AddFlags(&flags, &conf)
 
-			require.NoError(t, loadConfig(&conf, name, ""))
+			require.NoError(t, LoadConfig(&conf, name, nil))
 			require.NoError(t, flags.Parse(strings.Fields(tc.cli)))
 			tc.updatePreset(t, &preset)
 			require.Equal(t, preset, conf)
@@ -757,10 +751,8 @@ func TestConfig_CustomTypes(t *testing.T) {
 			require.NoError(t, err)
 
 			conf := config.Config{}
-			path := filepath.Join(t.TempDir(), "config.json")
-			require.NoError(t, os.WriteFile(path, []byte(tc.config), 0o600))
 
-			require.NoError(t, loadConfig(&conf, name, path))
+			require.NoError(t, LoadConfig(&conf, name, strings.NewReader(tc.config)))
 			tc.updatePreset(t, &preset)
 			require.Equal(t, preset, conf)
 		})
@@ -803,10 +795,8 @@ func TestConfig_PostProviderID_InvalidValues(t *testing.T) {
 		t.Run(fmt.Sprintf("%s_ConfigFile", tc.name), func(t *testing.T) {
 			conf := config.Config{}
 
-			path := filepath.Join(t.TempDir(), "config.json")
 			cfg := fmt.Sprintf(`{"smeshing": {"smeshing-opts": {"smeshing-opts-provider": %s}}}`, tc.configValue)
-			require.NoError(t, os.WriteFile(path, []byte(cfg), 0o600))
-			err := loadConfig(&conf, "", path)
+			err := LoadConfig(&conf, "", strings.NewReader(cfg))
 			require.ErrorContains(t, err, "invalid provider ID value")
 		})
 	}
@@ -816,18 +806,15 @@ func TestConfig_Load(t *testing.T) {
 	t.Run("invalid fails to load", func(t *testing.T) {
 		conf := config.Config{}
 
-		path := filepath.Join(t.TempDir(), "config.json")
-		require.NoError(t, os.WriteFile(path, []byte("}"), 0o600))
-
-		err := loadConfig(&conf, "", path)
-		require.ErrorContains(t, err, path)
+		err := LoadConfig(&conf, "", strings.NewReader("}"))
+		require.ErrorContains(t, err, "invalid character '}' looking for beginning of value")
 	})
 	t.Run("missing default doesn't fail", func(t *testing.T) {
 		conf := config.Config{}
 		var flags pflag.FlagSet
 		cmd.AddFlags(&flags, &conf)
 
-		require.NoError(t, loadConfig(&conf, "", ""))
+		require.NoError(t, LoadConfig(&conf, "", nil))
 		require.NoError(t, flags.Parse([]string{}))
 	})
 }
@@ -848,7 +835,7 @@ func TestConfig_GenesisAccounts(t *testing.T) {
 		args = append(args, fmt.Sprintf("-a %s=%d", key, value))
 	}
 
-	require.NoError(t, loadConfig(&conf, "", ""))
+	require.NoError(t, LoadConfig(&conf, "", nil))
 	require.NoError(t, flags.Parse(args))
 	for _, key := range keys {
 		require.EqualValues(t, value, conf.Genesis.Accounts[key])
@@ -858,9 +845,7 @@ func TestConfig_GenesisAccounts(t *testing.T) {
 func TestHRP(t *testing.T) {
 	conf := config.Config{}
 	data := `{"main": {"network-hrp": "TEST"}}`
-	cfg := filepath.Join(t.TempDir(), "config.json")
-	require.NoError(t, os.WriteFile(cfg, []byte(data), 0o600))
-	require.NoError(t, loadConfig(&conf, "", cfg))
+	require.NoError(t, LoadConfig(&conf, "", strings.NewReader(data)))
 	app := New(WithConfig(&conf))
 	require.NotNil(t, app)
 	require.Equal(t, "TEST", types.NetworkHRP())

syncer/atxsync/syncer.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"math/rand/v2"
 	"time"
 
 	"go.uber.org/zap"
@@ -107,9 +108,10 @@ func (s *Syncer) Download(parent context.Context, publish types.EpochID, downloa
 		return fmt.Errorf("failed to get last request time for epoch %v: %w", publish, err)
 	}
 	// in case of immediate we will request epoch info without waiting EpochInfoInterval
-	immediate := len(state) == 0 || (errors.Is(err, sql.ErrNotFound) || !lastSuccess.After(downloadUntil))
+	immediate := len(state) == 0 || errors.Is(err, sql.ErrNotFound) || !lastSuccess.After(downloadUntil)
 	if !immediate && total == downloaded {
-		s.logger.Debug("sync for epoch was completed before",
+		s.logger.Debug(
+			"sync for epoch was completed before",
 			log.ZContext(parent),
 			zap.Uint32("epoch_id", publish.Uint32()),
 		)
@@ -150,22 +152,24 @@ func (s *Syncer) downloadEpochInfo(
 ) error {
 	interval := s.cfg.EpochInfoInterval
 	if immediate {
-		interval = 0
+		interval = 1 * time.Second // not really immediate, to avoid an endless loop that doesn't wait between requests
 	}
+
 	for {
-		if interval != 0 {
-			s.logger.Debug(
-				"waiting between epoch info requests",
-				zap.Uint32("epoch_id", publish.Uint32()),
-				zap.Duration("duration", interval),
-			)
-		}
+		// randomize interval to avoid sync spikes
+		minWait := time.Duration(float64(interval) * 0.9)
+		maxWait := time.Duration(float64(interval) * 1.1)
+		wait := minWait + rand.N(maxWait-minWait+1)
+		s.logger.Debug(
+			"waiting between epoch info requests",
+			zap.Uint32("epoch_id", publish.Uint32()),
+			zap.Duration("duration", wait),
+		)
+
 		select {
 		case <-ctx.Done():
 			return nil
-		// TODO(dshulyak) this has to be randomized in a followup
-		// when sync will be schedulled in advance, in order to smooth out request rate across the network
-		case <-time.After(interval):
+		case <-time.After(wait):
 		}
 
 		peers := s.fetcher.SelectBestShuffled(s.cfg.EpochInfoPeers)