fix: run bacalhau in a separate container

.github/workflows/test.yml

@@ -45,6 +45,7 @@ jobs:
       - name: Run stack
         env:
           DISABLE_TELEMETRY: true
+          API_HOST: ""
         run: ./stack compose-up -d
 
       - name: Run tests

.local.dev

@@ -8,8 +8,8 @@
 
 API_HOST=http://localhost:8002/
 WEB3_RPC_URL=ws://localhost:8548
-SERVER_PORT=8080
-SERVER_URL=http://localhost:8080
+SERVER_PORT=8081
+SERVER_URL=http://localhost:8081
 DIRECTORY_ADDRESS=0x976EA74026E726554dB657fA54763abd0C3a0aa9
 JOB_CREATOR_ADDRESS=0x9965507D1a55bcC2695C58ba16FB37d819B0A4dc
 JOB_CREATOR_PRIVATE_KEY=0x5de4111afa1a4b94908f83103eb1f1706367c2e68ca870fc3fb9a804cdab365a

docker/bacalhau/Dockerfile

@@ -8,4 +8,7 @@ ADD https://github.com/bacalhau-project/bacalhau/releases/download/v1.3.2/bacalh
 RUN tar xfv bacalhau_v1.3.2_linux_amd64.tar.gz
 RUN mv bacalhau /usr/local/bin
 
+HEALTHCHECK --interval=30s --timeout=30s --start-period=5s --retries=3 \
+  CMD wget http://localhost:1234/api/v1/agent/alive -q || exit 1
+
 ENTRYPOINT [ "bacalhau" ]

docker/docker-compose.dev.yml

@@ -37,13 +37,42 @@ services:
       timeout: 10s
       retries: 5
   ipfs:
-    image: ipfs/kubo:v0.30.0
+    image: ipfs/kubo:v0.32.1
     container_name: ipfs
     restart: unless-stopped
     ports:
       - 5001:5001
     volumes:
       - ipfs-data:/data/ipfs
+  bacalhau:
+    image: ghcr.io/lilypad-tech/bacalhau
+    container_name: bacalhau
+    restart: unless-stopped
+    depends_on:
+      ipfs:
+        condition: service_healthy
+    build:
+      context: ..
+      dockerfile: ./docker/bacalhau/Dockerfile
+    extra_hosts:
+      - "localhost:host-gateway"
+    environment:
+      - BACALHAU_ENVIRONMENT=local
+    command:
+      [
+        "serve",
+        "--node-type",
+        "compute,requester",
+        "--peer",
+        "none",
+        "--private-internal-ipfs=false",
+        "--job-selection-accept-networked",
+        "--ipfs-connect",
+        "/dns4/ipfs/tcp/5001",
+      ]
+    volumes:
+      - bacalhau-data:/root/.bacalhau
+      - /var/run/docker.sock:/var/run/docker.sock
   solver:
     image: ghcr.io/lilypad-tech/solver
     container_name: solver
@@ -64,13 +93,12 @@ services:
       - WEB3_RPC_URL=${WEB3_RPC_URL}
       - DISABLE_TELEMETRY=${DISABLE_TELEMETRY}
     ports:
-      - 8080:8080
+      - 8081:8081
     healthcheck:
-      test: ["CMD", "curl", "-f", "http://localhost:8080/"]
+      test: ["CMD", "curl", "-f", "http://localhost:8081/api/v1/job_offers"]
       interval: 30s
       timeout: 10s
       retries: 5
-
   job-creator:
     image: ghcr.io/lilypad-tech/job-creator
     container_name: job-creator
@@ -93,9 +121,12 @@ services:
     container_name: resource-provider
     restart: unless-stopped
     depends_on:
-      - chain
-      - solver
-      - ipfs
+      ipfs:
+        condition: service_healthy
+      solver:
+        condition: service_healthy
+      bacalhau:
+        condition: service_healthy
     build:
       context: ..
       dockerfile: ./docker/resource-provider/Dockerfile
@@ -107,14 +138,17 @@ services:
     extra_hosts:
       - "localhost:host-gateway"
     volumes:
-      - bacalhau-data:/tmp/lilypad/data
-      - /var/run/docker.sock:/var/run/docker.sock
+      - lilypad-data:/tmp/lilypad/data
     environment:
       - WEB3_PRIVATE_KEY=${RESOURCE_PROVIDER_PRIVATE_KEY}
       - IPFS_CONNECT=/dns4/ipfs/tcp/5001
       - LOG_LEVEL=debug
       - DISABLE_TELEMETRY=${DISABLE_TELEMETRY}
+      - BACALHAU_API_HOST=bacalhau
+      - BACALHAU_NODE_CLIENTAPI_HOST=bacalhau
+      - BACALHAU_NODE_CLIENTAPI_PORT=1234
 volumes:
   chain-data:
   ipfs-data:
   bacalhau-data:
+  lilypad-data:

docker/docker-compose.yml

@@ -1,7 +1,7 @@
 # This is a docker-compose file for use by Resource Providers
 services:
   ipfs:
-    image: ipfs/kubo:v0.30.0
+    image: ipfs/kubo:v0.32.1
     container_name: ipfs
     restart: unless-stopped
     ports:
@@ -10,6 +10,30 @@ services:
       - 127.0.0.1:8080:8080
     volumes:
       - ipfs-data:/data/ipfs
+  bacalhau:
+    image: ghcr.io/lilypad-tech/bacalhau
+    container_name: bacalhau
+    restart: unless-stopped
+    depends_on:
+      ipfs:
+        condition: service_healthy
+    environment:
+      - BACALHAU_ENVIRONMENT=local
+    command:
+      [
+        "serve",
+        "--node-type",
+        "compute,requester",
+        "--peer",
+        "none",
+        "--private-internal-ipfs=false",
+        "--job-selection-accept-networked",
+        "--ipfs-connect",
+        "/dns4/ipfs/tcp/5001",
+      ]
+    volumes:
+      - bacalhau-data:/root/.bacalhau
+      - /var/run/docker.sock:/var/run/docker.sock
   resource-provider:
     image: ghcr.io/lilypad-tech/resource-provider:latest
     container_name: resource-provider
@@ -22,12 +46,14 @@ services:
       args:
         - COMPUTE_MODE=gpu
     volumes:
-      - bacalhau-data:/tmp/lilypad/data
-      - /var/run/docker.sock:/var/run/docker.sock
+      - lilypad-data:/tmp/lilypad/data
     environment:
       - WEB3_PRIVATE_KEY
       - WEB3_RPC_URL
       - IPFS_CONNECT=/dns4/ipfs/tcp/5001
+      - BACALHAU_API_HOST=bacalhau
+      - BACALHAU_NODE_CLIENTAPI_HOST=bacalhau
+      - BACALHAU_NODE_CLIENTAPI_PORT=1234
   watchtower:
     image: containrrr/watchtower
     container_name: watchtower
@@ -37,3 +63,4 @@ services:
 volumes:
   ipfs-data:
   bacalhau-data:
+  lilypad-data:

docker/resource-provider/Dockerfile

@@ -46,11 +46,10 @@ ENV PATH="/usr/local/bin:${PATH}"
 # Create a startup script to run both services simultaneously
 RUN touch run
 RUN echo "#!/bin/bash" >> run
-# Launch Bacalhau
-RUN echo "/usr/local/bin/bacalhau serve --node-type compute,requester --peer none --private-internal-ipfs=false --job-selection-accept-networked &" >> run
 
-# Wait for Bacalhau to be ready by checking the correct API endpoint
-RUN echo "while ! curl -s http://0.0.0.0:1234/api/v1/agent/alive | grep '\"Status\": \"OK\"'; do echo 'Waiting for Bacalhau...'; sleep 2; done" >> run
+# Ensure bacalhau is initialized
+RUN echo "export BACALHAU_ENVIRONMENT=local" >> run
+RUN echo "bacalhau id" >> run
 
 # Launch Lilypad
 RUN echo "/usr/local/bin/lilypad resource-provider --network ${NETWORK} --disable-pow=${DISABLE_POW} --disable-telemetry=${DISABLE_TELEMETRY} &" >> run

pkg/executor/bacalhau/bacalhau.go

@@ -208,7 +208,7 @@ func (executor *BacalhauExecutor) getJobID(
 	outputError := strings.Join(strings.Fields(strings.Join(splitOutputs[1:], " ")), " ")
 
 	if outputError != "" {
-		return "", fmt.Errorf("error running command %s -> %s, %s", deal.ID, outputError, runOutput)
+		log.Error().Msgf("error parsing output %s -> %s, %s", deal.ID, outputError, runOutput)
 	}
 
 	id := strings.TrimSpace(string(runOutput))

stack

@@ -312,6 +312,7 @@ function resource-provider-docker-run() {
 
 function bacalhau-node(){
   export BACALHAU_SERVE_IPFS_PATH=/tmp/lilypad/data/ipfs
+  export BACALHAU_ENVIRONMENT=local
   export LOG_LEVEL=debug
   bacalhau serve --node-type compute,requester --peer none --private-internal-ipfs=false --job-selection-accept-networked --ipfs-connect "/ip4/127.0.0.1/tcp/5001"
 }

test/ratelimit_test.go

@@ -58,7 +58,7 @@ func makeCalls(t *testing.T, path string, ch chan rateResult) {
 
 	// Make 100 requests
 	for range 100 {
-		requestURL := fmt.Sprintf("http://localhost:%d%s", 8080, path)
+		requestURL := fmt.Sprintf("http://localhost:%d%s", 8081, path)
 		res, err := http.Get(requestURL)
 
 		if err != nil {