`phala logs` returns empty — composition API returns containers: [] despite container running

[0xmatchmaker]

Summary

Container logs are unavailable via phala logs for CVMs that use dockerfile_inline in their docker-compose.yaml, even though the containers are running correctly.

Environment

• CLI version: latest (phala npm package)
• OS image: dstack-dev-0.5.6
• Instance type: tdx.small

Steps to Reproduce

1. Deploy a CVM with dockerfile_inline:

services:
  my-app:
    build:
      context: .
      dockerfile_inline: |
        FROM python:3.12-slim
        WORKDIR /app
        RUN echo 'print("hello")' > app.py
        CMD ["python3", "-u", "app.py"]
    ports:
      - "8080:8080"

2. Verify container is running: phala ps my-app → shows container running
3. Try to get logs: phala logs --cvm-id my-app → "No logs available"
4. Confirm app works via HTTP endpoint → returns valid response

Root Cause Analysis

Three API data sources are all empty:

1. GET /cvms/{id}/composition returns containers: [] — CLI can't discover container log endpoints
2. syslog_endpoint is null — fallback streaming path unavailable
3. services is [] despite public_logs: true

Meanwhile, phala ps successfully lists the container, suggesting the guest agent can enumerate containers but the composition/syslog APIs don't reflect this.

Workarounds

• phala logs --serial works (kernel/boot logs)
• Expose an HTTP status endpoint in the app
• Use phala ssh if SSH keys are configured

Expected Behavior

phala logs should return container stdout/stderr regardless of build method (dockerfile_inline vs pre-built image).

[Leechael]

Thanks for the detailed report. I reproduced the exact scenario using the compose file from the issue, deployed on dstack-dev-0.5.8, tested with phala CLI v1.1.13:

$ phala ps <cvm-id>
NAME             IMAGE          CREATED        STATUS                    STATE
dstack-my-app-1  dstack-my-app  4 minutes ago  Exited (0) 4 minutes ago  exited

$ phala logs --cvm-id <cvm-id>
hello

Both commands work correctly. Could you share more details to help reproduce the issue?

• CLI version (phala --version)
• Exact steps and timing — specifically, how long after deployment did you run phala logs? The dockerfile_inline build can take 30–60 seconds; querying before it completes would return empty results.
• Output of phala cvms get <cvm-id> --json at the time the issue occurred