test: tss migration e2e test (#2440)

* add new runner for migration tests * skip adding new keygen * skip adding new keygen * modify btc runner * tss fund migrate test for eth * add queries for tss funds migrator info * add changelog * fix lint messages * refactor keygen * refactor keygen * rearrange keygen creations * rearrange keygen creations * modify accounting * rebase develop * add retest after migration * add checks for error messages * add unit test for CheckMigration * fix unit tests for update msgUpdateTSS * fix some lint issues and rename all runners to r * fix some lint issues and rename all runners to r * add changelog * fix lint errors * fix lint errors * removed debug logs * add comments for zetasupervisor lib * add new user for migration tests * add new user for migration tests * change name to start-tss-migration-test * add changes for comments 1 * add changes for comments 1 * rename to CheckIfMigrationDeposit * rename to CheckIfMigrationDeposit * use cctx_Gateway to check cctx * add Tss migration to e2e.yml * move to a sub function * resolve comments 2 * resolve comments 3 * add todo * resolve comments 4 * Update .github/workflows/e2e.yml Co-authored-by: Francisco de Borja Aranda Castillejo <[email protected]> --------- Co-authored-by: Francisco de Borja Aranda Castillejo <[email protected]>
zeta-chain · Jul 17, 2024 · 8e1cfa7 · 8e1cfa7
1 parent abc15c8
commit 8e1cfa7
Show file tree

Hide file tree

Showing 40 changed files with 1,631 additions and 203 deletions.
diff --git a/.github/workflows/e2e.yml b/.github/workflows/e2e.yml
@@ -43,6 +43,10 @@ on:
         type: boolean
         required: false
         default: false
+      tss-migration-test:
+        type: boolean
+        required: false
+        default: false
 
 concurrency: 
   group: e2e-${{ github.head_ref || github.sha }}
@@ -61,13 +65,14 @@ jobs:
       ADMIN_TESTS: ${{ steps.matrix-conditionals.outputs.ADMIN_TESTS }}
       PERFORMANCE_TESTS: ${{ steps.matrix-conditionals.outputs.PERFORMANCE_TESTS }}
       STATEFUL_DATA_TESTS: ${{ steps.matrix-conditionals.outputs.STATEFUL_DATA_TESTS }}
+      TSS_MIGRATION_TESTS: ${{ steps.matrix-conditionals.outputs.TSS_MIGRATION_TESTS }}
+
     steps:
       # use api rather than event context to avoid race conditions (label added after push)
       - id: matrix-conditionals
         uses: actions/github-script@v7
         with:
           script: |
-            console.log(context);
             if (context.eventName === 'pull_request') {
               const { data: pr } = await github.rest.pulls.get({
                 owner: context.repo.owner,
@@ -83,6 +88,7 @@ jobs:
               core.setOutput('ADMIN_TESTS', labels.includes('ADMIN_TESTS'));
               core.setOutput('PERFORMANCE_TESTS', labels.includes('PERFORMANCE_TESTS'));
               core.setOutput('STATEFUL_DATA_TESTS', labels.includes('STATEFUL_DATA_TESTS'));
+              core.setOutput('TSS_MIGRATION_TESTS', labels.includes('TSS_MIGRATION_TESTS'));
             } else if (context.eventName === 'merge_group') {
               core.setOutput('DEFAULT_TESTS', true);
             } else if (context.eventName === 'push' && context.ref === 'refs/heads/develop') {
@@ -111,6 +117,7 @@ jobs:
               core.setOutput('ADMIN_TESTS', context.payload.inputs['admin-test']);
               core.setOutput('PERFORMANCE_TESTS', context.payload.inputs['performance-test']);
               core.setOutput('STATEFUL_DATA_TESTS', context.payload.inputs['stateful-data-test']);
+              core.setOutput('TSS_MIGRATION_TESTS', context.payload.inputs['tss-migration-test']);
             }
 
   e2e:
@@ -140,6 +147,9 @@ jobs:
           - make-target: "start-e2e-import-mainnet-test"
             runs-on: buildjet-16vcpu-ubuntu-2204
             run: ${{ needs.matrix-conditionals.outputs.STATEFUL_DATA_TESTS == 'true' }}
+          - make-target: "start-tss-migration-test"
+            runs-on: ubuntu-20.04
+            run: ${{ needs.matrix-conditionals.outputs.TSS_MIGRATION_TESTS == 'true' }}
     name: ${{ matrix.make-target }}
     uses: ./.github/workflows/reusable-e2e.yml
     with:
@@ -150,7 +160,7 @@ jobs:
   # this allows you to set a required status check
   e2e-ok:
     runs-on: ubuntu-22.04
-    needs: 
+    needs:
       - matrix-conditionals
       - e2e
     if: always()

diff --git a/Makefile b/Makefile
@@ -87,7 +87,7 @@ build-testnet-ubuntu: go.sum
 		docker rm temp-container
 
 install: go.sum
-		@echo "--> Installing zetacored & zetaclientd"
+		@echo "--> Installing zetacored, zetaclientd, and zetaclientd-supervisor"
 		@go install -mod=readonly $(BUILD_FLAGS) ./cmd/zetacored
 		@go install -mod=readonly $(BUILD_FLAGS) ./cmd/zetaclientd
 		@go install -mod=readonly $(BUILD_FLAGS) ./cmd/zetaclientd-supervisor
@@ -254,6 +254,11 @@ start-stress-test: zetanode
 	@echo "--> Starting stress test"
 	cd contrib/localnet/ && $(DOCKER) compose --profile stress -f docker-compose.yml up -d
 
+start-tss-migration-test: zetanode
+	@echo "--> Starting migration test"
+	export E2E_ARGS="--test-tss-migration" && \
+	cd contrib/localnet/ && $(DOCKER) compose up -d
+
 ###############################################################################
 ###                         Upgrade Tests              						###
 ###############################################################################

diff --git a/changelog.md b/changelog.md
@@ -80,6 +80,8 @@
 * [2369](https://github.com/zeta-chain/node/pull/2369) - fix random cross-chain swap failure caused by using tiny UTXO
 * [2549](https://github.com/zeta-chain/node/pull/2459) - add separate accounts for each policy in e2e tests
 * [2415](https://github.com/zeta-chain/node/pull/2415) - add e2e test for upgrade and test admin functionalities
+* [2440](https://github.com/zeta-chain/node/pull/2440) - Add e2e test for TSS migration
+
 
 ### Fixes
 

diff --git a/cmd/zetaclientd-supervisor/lib.go b/cmd/zetaclientd-supervisor/lib.go
@@ -12,6 +12,7 @@ import (
 	"runtime"
 	"strings"
 	"sync"
+	"syscall"
 	"time"
 
 	"github.com/cosmos/cosmos-sdk/client/grpc/tmservice"
@@ -20,6 +21,7 @@ import (
 	"github.com/rs/zerolog"
 	"google.golang.org/grpc"
 
+	observertypes "github.com/zeta-chain/zetacore/x/observer/types"
 	"github.com/zeta-chain/zetacore/zetaclient/config"
 )
 
@@ -66,6 +68,7 @@ type zetaclientdSupervisor struct {
 	upgradesDir        string
 	upgradePlanName    string
 	enableAutoDownload bool
+	restartChan        chan os.Signal
 }
 
 func newZetaclientdSupervisor(
@@ -81,19 +84,24 @@ func newZetaclientdSupervisor(
 	if err != nil {
 		return nil, fmt.Errorf("grpc dial: %w", err)
 	}
-
+	// these signals will result in the supervisor process only restarting zetaclientd
+	restartChan := make(chan os.Signal, 1)
 	return &zetaclientdSupervisor{
 		zetacoredConn:      conn,
 		logger:             logger,
 		reloadSignals:      make(chan bool, 1),
 		upgradesDir:        defaultUpgradesDir,
 		enableAutoDownload: enableAutoDownload,
+		restartChan:        restartChan,
 	}, nil
 }
 
 func (s *zetaclientdSupervisor) Start(ctx context.Context) {
 	go s.watchForVersionChanges(ctx)
 	go s.handleCoreUpgradePlan(ctx)
+	go s.handleNewKeygen(ctx)
+	go s.handleNewTSSKeyGeneration(ctx)
+	go s.handleTSSUpdate(ctx)
 }
 
 func (s *zetaclientdSupervisor) WaitForReloadSignal(ctx context.Context) {
@@ -169,6 +177,125 @@ func (s *zetaclientdSupervisor) watchForVersionChanges(ctx context.Context) {
 	}
 }
 
+func (s *zetaclientdSupervisor) handleTSSUpdate(ctx context.Context) {
+	maxRetries := 11
+	retryInterval := 5 * time.Second
+
+	// TODO : use retry library under pkg/retry
+	// https://github.com/zeta-chain/node/issues/2492
+	for i := 0; i < maxRetries; i++ {
+		client := observertypes.NewQueryClient(s.zetacoredConn)
+		tss, err := client.TSS(ctx, &observertypes.QueryGetTSSRequest{})
+		if err != nil {
+			s.logger.Warn().Err(err).Msg("unable to get original tss")
+			time.Sleep(retryInterval)
+			continue
+		}
+		i = 0
+		for {
+			select {
+			case <-time.After(time.Second):
+			case <-ctx.Done():
+				return
+			}
+			tssNew, err := client.TSS(ctx, &observertypes.QueryGetTSSRequest{})
+			if err != nil {
+				s.logger.Warn().Err(err).Msg("unable to get tss")
+				continue
+			}
+
+			if tssNew.TSS.TssPubkey == tss.TSS.TssPubkey {
+				continue
+			}
+
+			tss = tssNew
+			s.logger.Info().
+				Msgf("tss address is updated from %s to %s", tss.TSS.TssPubkey, tssNew.TSS.TssPubkey)
+			time.Sleep(6 * time.Second)
+			s.logger.Info().Msg("restarting zetaclientd to update tss address")
+			s.restartChan <- syscall.SIGHUP
+		}
+	}
+	s.logger.Warn().Msg("handleTSSUpdate exiting without success")
+}
+
+func (s *zetaclientdSupervisor) handleNewTSSKeyGeneration(ctx context.Context) {
+	maxRetries := 11
+	retryInterval := 5 * time.Second
+
+	// TODO : use retry library under pkg/retry
+	for i := 0; i < maxRetries; i++ {
+		client := observertypes.NewQueryClient(s.zetacoredConn)
+		alltss, err := client.TssHistory(ctx, &observertypes.QueryTssHistoryRequest{})
+		if err != nil {
+			s.logger.Warn().Err(err).Msg("unable to get tss original history")
+			time.Sleep(retryInterval)
+			continue
+		}
+		i = 0
+		tssLenCurrent := len(alltss.TssList)
+		for {
+			select {
+			case <-time.After(time.Second):
+			case <-ctx.Done():
+				return
+			}
+			tssListNew, err := client.TssHistory(ctx, &observertypes.QueryTssHistoryRequest{})
+			if err != nil {
+				s.logger.Warn().Err(err).Msg("unable to get tss new history")
+				continue
+			}
+			tssLenUpdated := len(tssListNew.TssList)
+
+			if tssLenUpdated == tssLenCurrent {
+				continue
+			}
+			if tssLenUpdated < tssLenCurrent {
+				tssLenCurrent = len(tssListNew.TssList)
+				continue
+			}
+
+			tssLenCurrent = tssLenUpdated
+			s.logger.Info().Msgf("tss list updated from %d to %d", tssLenCurrent, tssLenUpdated)
+			time.Sleep(5 * time.Second)
+			s.logger.Info().Msg("restarting zetaclientd to update tss list")
+			s.restartChan <- syscall.SIGHUP
+		}
+	}
+	s.logger.Warn().Msg("handleNewTSSKeyGeneration exiting without success")
+}
+
+func (s *zetaclientdSupervisor) handleNewKeygen(ctx context.Context) {
+	client := observertypes.NewQueryClient(s.zetacoredConn)
+	prevKeygenBlock := int64(0)
+	for {
+		select {
+		case <-time.After(time.Second):
+		case <-ctx.Done():
+			return
+		}
+		resp, err := client.Keygen(ctx, &observertypes.QueryGetKeygenRequest{})
+		if err != nil {
+			s.logger.Warn().Err(err).Msg("unable to get keygen")
+			continue
+		}
+		if resp.Keygen == nil {
+			s.logger.Warn().Err(err).Msg("keygen is nil")
+			continue
+		}
+
+		if resp.Keygen.Status != observertypes.KeygenStatus_PendingKeygen {
+			continue
+		}
+		keygenBlock := resp.Keygen.BlockNumber
+		if prevKeygenBlock == keygenBlock {
+			continue
+		}
+		prevKeygenBlock = keygenBlock
+		s.logger.Info().Msgf("got new keygen at block %d", keygenBlock)
+		s.restartChan <- syscall.SIGHUP
+	}
+}
 func (s *zetaclientdSupervisor) handleCoreUpgradePlan(ctx context.Context) {
 	client := upgradetypes.NewQueryClient(s.zetacoredConn)
 

diff --git a/cmd/zetaclientd-supervisor/main.go b/cmd/zetaclientd-supervisor/main.go
@@ -36,10 +36,6 @@ func main() {
 	shutdownChan := make(chan os.Signal, 1)
 	signal.Notify(shutdownChan, syscall.SIGINT, syscall.SIGTERM)
 
-	// these signals will result in the supervisor process only restarting zetaclientd
-	restartChan := make(chan os.Signal, 1)
-	signal.Notify(restartChan, syscall.SIGHUP)
-
 	hotkeyPassword, tssPassword, err := promptPasswords()
 	if err != nil {
 		logger.Error().Err(err).Msg("unable to get passwords")
@@ -53,6 +49,8 @@ func main() {
 		os.Exit(1)
 	}
 	supervisor.Start(ctx)
+	// listen for SIGHUP to trigger a restart of zetaclientd
+	signal.Notify(supervisor.restartChan, syscall.SIGHUP)
 
 	shouldRestart := true
 	for shouldRestart {
@@ -82,7 +80,7 @@ func main() {
 				select {
 				case <-ctx.Done():
 					return nil
-				case sig := <-restartChan:
+				case sig := <-supervisor.restartChan:
 					logger.Info().Msgf("got signal %d, sending SIGINT to zetaclientd", sig)
 				case sig := <-shutdownChan:
 					logger.Info().Msgf("got signal %d, shutting down", sig)