feat: start server-manager as an internal service

build/docker-compose.yml

@@ -7,8 +7,6 @@ services:
     build:
       context: ..
       dockerfile: build/Dockerfile
-      tags:
-        - "cartesi/rollups-node:devel"
     entrypoint: ["cartesi-rollups-node", "validator"]
     ports:
       - "4000:4000"
@@ -20,8 +18,8 @@ services:
         condition: service_completed_successfully
       hardhat_set_interval:
         condition: service_completed_successfully
-      server_manager:
-        condition: service_healthy
+      machine_snapshot_setup:
+        condition: service_completed_successfully
       database:
         condition: service_healthy
       redis:
@@ -34,51 +32,56 @@ services:
 
       #Advance Runner
       ADVANCE_RUNNER_HEALTHCHECK_PORT: 8080
-      SERVER_MANAGER_ENDPOINT: http://server_manager:5001
+      SERVER_MANAGER_ENDPOINT: http://0.0.0.0:5001
       PROVIDER_HTTP_ENDPOINT: http://hardhat:8545
       SESSION_ID: default_rollups_id
       SNAPSHOT_DIR: /var/opt/cartesi/machine-snapshots
       SNAPSHOT_LATEST: /var/opt/cartesi/machine-snapshots/latest
 
-      #Authority Claimer
+      # Authority Claimer
       AUTHORITY_CLAIMER_HTTP_SERVER_PORT: 8085
       TX_PROVIDER_HTTP_ENDPOINT: http://hardhat:8545
       TX_CHAIN_ID: 31337
       TX_CHAIN_IS_LEGACY: ${TX_LEGACY:-false}
       TX_DEFAULT_CONFIRMATIONS: 2
       TX_SIGNING_MNEMONIC: "test test test test test test test test test test test junk"
 
-      #Dispatcher
+      # Dispatcher
       DISPATCHER_HTTP_SERVER_PORT: 8081
       DAPP_DEPLOYMENT_FILE: /deployments/localhost/dapp.json
       ROLLUPS_DEPLOYMENT_FILE: /opt/cartesi/share/deployments/localhost.json
       RD_EPOCH_DURATION: 86400
       SC_GRPC_ENDPOINT: http://0.0.0.0:50051
       SC_DEFAULT_CONFIRMATIONS: 1
 
-      #GraphQL Server
+      # GraphQL Server
       GRAPHQL_HEALTHCHECK_PORT: 8082
       GRAPHQL_HOST: "0.0.0.0"
       GRAPHQL_PORT: "4000"
 
-      #Indexer
+      # Indexer
       INDEXER_HEALTHCHECK_PORT: 8083
       DAPP_CONTRACT_ADDRESS_FILE: /deployments/localhost/dapp.json
       REDIS_ENDPOINT: redis://redis:6379
 
-      #Inspect Server
+      # Inspect Server
       INSPECT_SERVER_HEALTHCHECK_PORT: 8084
       INSPECT_SERVER_ADDRESS: 0.0.0.0:5005
-      SERVER_MANAGER_ADDRESS: server_manager:5001
+      SERVER_MANAGER_ADDRESS: 0.0.0.0:5001
 
-      #State Server
+      # State Server
       SS_SERVER_ADDRESS: 0.0.0.0:50051
       SF_GENESIS_BLOCK: 0x1
       SF_SAFETY_MARGIN: 1
       BH_HTTP_ENDPOINT: http://hardhat:8545
       BH_WS_ENDPOINT: ws://hardhat:8545
       BH_BLOCK_TIMEOUT: 8
 
+      # Server Manager
+      MANAGER_ADDRESS: http://0.0.0.0:5001
+      SERVER_MANAGER_LOG_LEVEL: warning
+      REMOTE_CARTESI_MACHINE_LOG_LEVEL: info
+
     volumes:
       - machine:/var/opt/cartesi/machine-snapshots
       - blockchain-data:/opt/cartesi/share/deployments:ro
@@ -113,8 +116,8 @@ services:
     depends_on:
       hardhat:
         condition: service_healthy
-      server_manager:
-        condition: service_healthy
+      machine_snapshot_setup:
+        condition: service_completed_successfully
     command:
       [
         "create",
@@ -164,23 +167,13 @@ services:
         "http://hardhat:8545",
       ]
 
-  server_manager:
+  machine_snapshot_setup:
+    image: cartesi/rollups-machine:devel
     build:
       context: ..
       dockerfile: build/machine.Dockerfile
-    restart: always
-    ports:
-      - "5001:5001"
-    healthcheck:
-      test: ["CMD-SHELL", 'bash -c ''echo "" > /dev/tcp/127.0.0.1/5001;''']
-      interval: 10s
-      timeout: 5s
-      retries: 5
     volumes:
       - machine:/var/opt/cartesi/machine-snapshots
-    environment:
-      - SERVER_MANAGER_LOG_LEVEL=warning
-      - REMOTE_CARTESI_MACHINE_LOG_LEVEL=info
 
   database:
     image: postgres:13-alpine

build/machine.Dockerfile

@@ -1,7 +1,7 @@
 # (c) Cartesi and individual authors (see AUTHORS)
 # SPDX-License-Identifier: Apache-2.0 (see LICENSE)
 
-FROM cartesi/server-manager:0.8.2 as build-server-stage
+FROM cartesi/server-manager:0.8.2 as build-machine-stage
 
 USER root
 
@@ -23,8 +23,13 @@ RUN cartesi-machine \
     --store=$SNAPSHOT_DIR \
     -- "ioctl-echo-loop --vouchers=1 --notices=1 --reports=1 --verbose=1"
 
-FROM cartesi/server-manager:0.8.2 as server-stage
+FROM debian:bookworm-20230725-slim as machine-stage
 
-WORKDIR /opt/cartesi/bin
-COPY --from=build-server-stage --chown=cartesi:cartesi /tmp/dapp-bin /var/opt/cartesi/machine-snapshots/0_0
+RUN addgroup --system --gid 102 cartesi && \
+    adduser --system --uid 102 --ingroup cartesi --disabled-login --no-create-home --home /nonexistent --gecos "cartesi user" --shell /bin/false cartesi
+
+COPY --from=build-machine-stage --chown=cartesi:cartesi /tmp/dapp-bin /var/opt/cartesi/machine-snapshots/0_0
 RUN ln -s /var/opt/cartesi/machine-snapshots/0_0 /var/opt/cartesi/machine-snapshots/latest
+
+WORKDIR /var/opt/cartesi/machine-snapshots
+ENTRYPOINT [ "/bin/bash" ]

internal/node/node.go

@@ -13,51 +13,63 @@ import (
 	"github.com/cartesi/rollups-node/internal/services"
 )
 
+const (
+	serverManagerPort    = "5001"
+	serverManagerAddress = "0.0.0.0:" + serverManagerPort
+)
+
 var ValidatorServices = []services.Service{
-	StateServer,
+	StateServer, // must start before Dispatcher
 	AdvanceRunner,
 	AuthorityClaimer,
 	Dispatcher,
 	GraphQLServer,
 	Indexer,
 	InspectServer,
+	ServerManager,
 }
 
 var (
 	AdvanceRunner = services.NewService(
 		"advance-runner",
-		"cartesi-rollups-advance-runner",
 		healthcheckPort("advance-runner"),
+		"cartesi-rollups-advance-runner",
 	)
 	AuthorityClaimer = services.NewService(
 		"authority-claimer",
-		"cartesi-rollups-authority-claimer",
 		healthcheckPort("authority-claimer"),
+		"cartesi-rollups-authority-claimer",
 	)
 	Dispatcher = services.NewService(
 		"dispatcher",
-		"cartesi-rollups-dispatcher",
 		healthcheckPort("dispatcher"),
+		"cartesi-rollups-dispatcher",
 	)
 	GraphQLServer = services.NewService(
 		"graphql-server",
-		"cartesi-rollups-graphql-server",
 		healthcheckPort("graphql"),
+		"cartesi-rollups-graphql-server",
 	)
 	Indexer = services.NewService(
 		"indexer",
-		"cartesi-rollups-indexer",
 		healthcheckPort("indexer"),
+		"cartesi-rollups-indexer",
 	)
 	InspectServer = services.NewService(
 		"inspect-server",
-		"cartesi-rollups-inspect-server",
 		healthcheckPort("inspect-server"),
+		"cartesi-rollups-inspect-server",
 	)
 	StateServer = services.NewService(
 		"state-server",
-		"cartesi-rollups-state-server",
 		stateServerHealthcheckPort(),
+		"cartesi-rollups-state-server",
+	)
+	ServerManager = services.NewService(
+		"server-manager",
+		serverManagerPort,
+		"server-manager",
+		"--manager-address="+serverManagerAddress,
 	)
 )
 

internal/services/service.go

@@ -24,19 +24,24 @@ const (
 )
 
 type Service struct {
-	name            string
-	binaryName      string
+	// name identifies the service
+	name string
+	// healthcheckPort is a port used to verify if the service is ready
 	healthcheckPort string
+	// path to the service binary
+	path string
+	// args to the service binary
+	args []string
 }
 
-func NewService(name, binaryName, healthcheckPort string) Service {
-	return Service{name, binaryName, healthcheckPort}
+func NewService(name, healthcheckPort, path string, args ...string) Service {
+	return Service{name, healthcheckPort, path, args}
 }
 
 // Start will execute a binary and wait for its completion or until the context
 // is canceled
 func (s Service) Start(ctx context.Context) error {
-	cmd := exec.CommandContext(ctx, s.binaryName)
+	cmd := exec.CommandContext(ctx, s.path, s.args...)
 	cmd.Stderr = os.Stderr
 	cmd.Stdout = os.Stdout
 	cmd.Cancel = func() error {

internal/services/service_test.go

@@ -45,7 +45,7 @@ func (s *ServiceTestSuite) SetupTest() {
 func (s *ServiceTestSuite) TestServiceStops() {
 	service := Service{
 		name:            "fake-service",
-		binaryName:      "fake-service",
+		path:            "fake-service",
 		healthcheckPort: fmt.Sprint(s.servicePort),
 	}
 	ctx, cancel := context.WithCancel(context.Background())
@@ -69,7 +69,7 @@ func (s *ServiceTestSuite) TestServiceStops() {
 func (s *ServiceTestSuite) TestServiceTimeout() {
 	service := Service{
 		name:            "fake-service",
-		binaryName:      "fake-service",
+		path:            "fake-service",
 		healthcheckPort: "0000", // wrong port
 	}
 	ctx, cancel := context.WithCancel(context.Background())
@@ -94,7 +94,7 @@ func (s *ServiceTestSuite) TestServiceTimeout() {
 func (s *ServiceTestSuite) TestServiceReady() {
 	service := Service{
 		name:            "fake-service",
-		binaryName:      "fake-service",
+		path:            "fake-service",
 		healthcheckPort: fmt.Sprint(s.servicePort),
 	}
 	ctx, cancel := context.WithCancel(context.Background())