Downscale test runner size, increase shards

.github/workflows/optimize-test-sharding.yml

@@ -42,7 +42,7 @@ jobs:
         run: |
           # NOTE: shard count must match shards in run-tests.yml
           go run ./cmd/tools/optimize-test-sharding \
-            -shards 3 \
+            -shards 5 \
             -workflow run-tests.yml \
             -artifact-pattern 'junit-xml--*shard*--functional-test' \
             -file "${{ env.SALT_FILE }}" \

.github/workflows/run-tests.yml

@@ -24,7 +24,8 @@ env:
   DOCKER_COMPOSE_FILE: ./develop/github/docker-compose.yml
   TEMPORAL_VERSION_CHECK_DISABLED: 1
   MAX_TEST_ATTEMPTS: 3
-  SHARD_COUNT: 3 # NOTE: must match shard count in optimize-test-sharding.yml
+  SHARD_COUNT: 5 # NOTE: must match shard count in optimize-test-sharding.yml
+  RUNNER_ARM: ubuntu-24.04-arm
 
 jobs:
   test-setup:
@@ -99,14 +100,7 @@ jobs:
 
       - name: Configure runners
         id: configure_runners
-        run: |
-          # Use 8-core runners for temporalio org, standard runners for forks
-          if [[ "${{ github.repository_owner }}" == "temporalio" ]]; then
-            runner_arm="ubuntu-24.04-arm64-8-cores"
-          else
-            runner_arm="ubuntu-24.04-arm"
-          fi
-          echo "runner_arm=$runner_arm" >> "$GITHUB_OUTPUT"
+        run: echo "runner_arm=$RUNNER_ARM" >> "$GITHUB_OUTPUT"
 
       # Primary DBs always get full tests, extended DBs get smoke tests (1 job) in abridged PRs.
       - name: Build job matrix
@@ -154,6 +148,7 @@ jobs:
               test_timeout: 35m
               github_timeout: 40
               sharded: true
+              reuse_schema: true
             smoke:
               cmd: make functional-test-coverage
               test_timeout: 5m
@@ -539,6 +534,15 @@ jobs:
           # shellcheck disable=SC2046
           docker compose -f ${{ env.DOCKER_COMPOSE_FILE }} up --wait $(docker compose -f ${{ env.DOCKER_COMPOSE_FILE }} ps --services)
 
+      - name: Install schema
+        if: ${{ matrix.reuse_schema }}
+        run: |
+          case "${{ matrix.persistence_driver }}" in
+            cassandra)    make install-schema-functional-cassandra ;;
+            mysql8)       make install-schema-functional-mysql8 ;;
+            postgres12*)  make install-schema-functional-postgresql12 ;;
+          esac
+
       - name: Run functional test
         timeout-minutes: ${{ matrix.github_timeout }}
         run: ./develop/github/monitor_test.sh ${{ matrix.cmd }}
@@ -548,6 +552,8 @@ jobs:
           TEST_ARGS: "${{ matrix.test_args }}"
           TEMPORAL_TEST_LOG_FILE: ${{ github.workspace }}/.testoutput/debug.log
           TEMPORAL_TEST_LOG_LEVEL: info
+          GOMEMLIMIT: 5GiB
+          TEMPORAL_REUSE_SCHEMA: ${{ matrix.reuse_schema && 'true' || '' }}
 
       - name: Dump container logs
         if: ${{ failure() && toJson(matrix.containers) != '[]' }}

Makefile

@@ -595,6 +595,29 @@ install-schema-postgresql12: temporal-sql-tool
 	./temporal-sql-tool -u $(SQL_USER) --pw $(SQL_PASSWORD) -p 5432 --pl postgres12 --db $(VISIBILITY_DB) setup-schema -v 0.0
 	./temporal-sql-tool -u $(SQL_USER) --pw $(SQL_PASSWORD) -p 5432 --pl postgres12 --db $(VISIBILITY_DB) update-schema -d ./schema/postgresql/v12/visibility/versioned
 
+install-schema-functional-cassandra: temporal-cassandra-tool
+	@printf $(COLOR) "Install Cassandra functional test schema..."
+	./temporal-cassandra-tool drop -k $(TEMPORAL_DB) -f
+	./temporal-cassandra-tool create -k $(TEMPORAL_DB) --rf 1
+	./temporal-cassandra-tool -k $(TEMPORAL_DB) setup-schema -v 0.0
+	./temporal-cassandra-tool -k $(TEMPORAL_DB) update-schema -d ./schema/cassandra/temporal/versioned
+
+install-schema-functional-mysql8: temporal-sql-tool
+	@printf $(COLOR) "Install MySQL functional test schema (temporal + visibility into one DB)..."
+	./temporal-sql-tool -u $(SQL_USER) --pw $(SQL_PASSWORD) --pl mysql8 --db $(TEMPORAL_DB) drop -f
+	./temporal-sql-tool -u $(SQL_USER) --pw $(SQL_PASSWORD) --pl mysql8 --db $(TEMPORAL_DB) create
+	./temporal-sql-tool -u $(SQL_USER) --pw $(SQL_PASSWORD) --pl mysql8 --db $(TEMPORAL_DB) setup-schema -v 0.0
+	./temporal-sql-tool -u $(SQL_USER) --pw $(SQL_PASSWORD) --pl mysql8 --db $(TEMPORAL_DB) update-schema -d ./schema/mysql/v8/temporal/versioned
+	./temporal-sql-tool -u $(SQL_USER) --pw $(SQL_PASSWORD) --pl mysql8 --db $(TEMPORAL_DB) update-schema -d ./schema/mysql/v8/visibility/versioned
+
+install-schema-functional-postgresql12: temporal-sql-tool
+	@printf $(COLOR) "Install Postgres functional test schema (temporal + visibility into one DB)..."
+	./temporal-sql-tool -u $(SQL_USER) --pw $(SQL_PASSWORD) -p 5432 --pl postgres12 --db $(TEMPORAL_DB) drop -f
+	./temporal-sql-tool -u $(SQL_USER) --pw $(SQL_PASSWORD) -p 5432 --pl postgres12 --db $(TEMPORAL_DB) create
+	./temporal-sql-tool -u $(SQL_USER) --pw $(SQL_PASSWORD) -p 5432 --pl postgres12 --db $(TEMPORAL_DB) setup-schema -v 0.0
+	./temporal-sql-tool -u $(SQL_USER) --pw $(SQL_PASSWORD) -p 5432 --pl postgres12 --db $(TEMPORAL_DB) update-schema -d ./schema/postgresql/v12/temporal/versioned
+	./temporal-sql-tool -u $(SQL_USER) --pw $(SQL_PASSWORD) -p 5432 --pl postgres12 --db $(TEMPORAL_DB) update-schema -d ./schema/postgresql/v12/visibility/versioned
+
 install-schema-es: temporal-elasticsearch-tool
 	@printf $(COLOR) "Install Elasticsearch schema..."
 	./temporal-elasticsearch-tool -ep http://127.0.0.1:9200 setup-schema

common/persistence/cassandra/test.go

@@ -27,16 +27,17 @@ const (
 
 // TestCluster allows executing cassandra operations in testing.
 type TestCluster struct {
-	keyspace       string
-	schemaDir      string
-	session        commongocql.Session
-	cfg            config.Cassandra
-	faultInjection *config.FaultInjection
-	logger         log.Logger
+	keyspace        string
+	schemaDir       string
+	session         commongocql.Session
+	cfg             config.Cassandra
+	faultInjection  *config.FaultInjection
+	skipSchemaSetup bool
+	logger          log.Logger
 }
 
 // NewTestCluster returns a new cassandra test cluster
-func NewTestCluster(keyspace, username, password, host string, port int, schemaDir string, faultInjection *config.FaultInjection, logger log.Logger) *TestCluster {
+func NewTestCluster(keyspace, username, password, host string, port int, schemaDir string, faultInjection *config.FaultInjection, skipSchemaSetup bool, logger log.Logger) *TestCluster {
 	var result TestCluster
 	result.logger = logger
 	result.keyspace = keyspace
@@ -60,6 +61,7 @@ func NewTestCluster(keyspace, username, password, host string, port int, schemaD
 		Keyspace:       keyspace,
 	}
 	result.faultInjection = faultInjection
+	result.skipSchemaSetup = skipSchemaSetup
 	return &result
 }
 
@@ -82,6 +84,10 @@ func (s *TestCluster) DatabaseName() string {
 
 // SetupTestDatabase from PersistenceTestCluster interface
 func (s *TestCluster) SetupTestDatabase() {
+	if s.skipSchemaSetup {
+		s.CreateSession(s.DatabaseName())
+		return
+	}
 	s.CreateSession("system")
 	s.CreateDatabase()
 	s.CreateSession(s.DatabaseName())
@@ -97,7 +103,9 @@ func (s *TestCluster) SetupTestDatabase() {
 
 // TearDownTestDatabase from PersistenceTestCluster interface
 func (s *TestCluster) TearDownTestDatabase() {
-	s.DropDatabase()
+	if !s.skipSchemaSetup {
+		s.DropDatabase()
+	}
 	s.session.Close()
 }
 

common/persistence/persistence-tests/persistence_test_base.go

@@ -61,6 +61,9 @@ type (
 		SchemaDir         string `yaml:"-"`
 		FaultInjection    *config.FaultInjection
 		Logger            log.Logger `yaml:"-"`
+		// SkipSchemaSetup skips database creation, schema loading, and teardown.
+		// Use when the database and schema are pre-created externally (e.g. in CI).
+		SkipSchemaSetup bool `yaml:"-"`
 	}
 )
 
@@ -131,7 +134,7 @@ func NewTestClusterForCassandra(options *TestBaseOptions, logger log.Logger) *ca
 	if options.DBName == "" {
 		options.DBName = "test_" + GenerateRandomDBName(3)
 	}
-	testCluster := cassandra.NewTestCluster(options.DBName, options.DBUsername, options.DBPassword, options.DBHost, options.DBPort, options.SchemaDir, options.FaultInjection, logger)
+	testCluster := cassandra.NewTestCluster(options.DBName, options.DBUsername, options.DBPassword, options.DBHost, options.DBPort, options.SchemaDir, options.FaultInjection, options.SkipSchemaSetup, logger)
 	return testCluster
 }
 
@@ -166,7 +169,7 @@ func NewTestBaseWithSQL(options *TestBaseOptions) *TestBase {
 			panic(fmt.Sprintf("unknown sql store driver: %v", options.SQLDBPluginName))
 		}
 	}
-	testCluster := sql.NewTestCluster(options.SQLDBPluginName, options.DBName, options.DBUsername, options.DBPassword, options.DBHost, options.DBPort, options.ConnectAttributes, options.SchemaDir, options.FaultInjection, logger)
+	testCluster := sql.NewTestCluster(options.SQLDBPluginName, options.DBName, options.DBUsername, options.DBPassword, options.DBHost, options.DBPort, options.ConnectAttributes, options.SchemaDir, options.FaultInjection, options.SkipSchemaSetup, logger)
 	return NewTestBaseForCluster(testCluster, logger)
 }