Deploy

/deploy — push artifacts to the H15 SBC and run

The board image already has the libs, postprocess .sos, and a default set of HEFs. /deploy's job is to push the changed artifacts (binary, edited configs, sometimes a new HEF) without breaking what's already there.

Preconditions

• /connect succeeded — ssh -o BatchMode=yes <board> works non-interactively.
• A binary/set of config files staged on the host, ready to push.

Procedure

1. Decide what's changing. Common cases:

   • Binary only (after /swap-model or another main.cpp change).
   • Configs only (after /add-stream or other medialib edits).
   • Both.
   • HEF (less common — usually pre-shipped on the board image).

2. Back up first. For anything you're overwriting, make a .bak on the board side. Cheap insurance. Paths below are an example (the detection case study) — substitute the actual app path you're deploying:

   ssh root@<board> '
     cp /home/root/apps/case_studies/detection/detection_case_study \
        /home/root/apps/case_studies/detection/detection_case_study.bak
     cp <profile_dir>/application_settings.json{,.bak}
     cp <profile_dir>/<profile>.json{,.bak}
   '
   

3. Push. Substitute the actual app/binary names and resource paths — the values below are examples (detection case study, face_landmarks HEF dir).

   • Binary:

     scp <local>/<app_name>.aarch64 root@<board>:<board_app_path>
     ssh root@<board> 'chmod +x <board_app_path>'
     # example: <board_app_path> = /home/root/apps/case_studies/detection/detection_case_study
     

   • Configs (medialib JSONs go into the active profile dir, which is determined by the medialib config — typically /etc/imaging/cfg/hailo15<x>/<sensor>/<lens>/<res>/profiles/<name>/):

     scp <local>/configs/*.json root@<board>:<profile_dir>/
     

   • HEF (if pushing a new one) — destination is the app's resources dir:

     scp <local>/<model>.hef root@<board>:<app_resources_dir>/
     # example: <app_resources_dir> = /home/root/apps/face_landmarks/resources
     

4. Stop any running app instance. Check for a running instance and kill it before re-running:

   ssh root@<board> 'pgrep -af "<app_name>" || echo no app running'
   

5. Smoke test. Run the app for a short window with FPS printing. Important: edited medialib configs need the hash bypass until they're hash-regenerated:

   ssh root@<board> '
     HAILO_MEDIA_LIB_SKIP_METADATA_CONFIG_VALIDATION=1 \
     <board_app_path> -t 12 -p 2>&1 | tail -40
   '
   # example: <board_app_path> = /home/root/apps/case_studies/detection/detection_case_study
   

   Look for:

   • No Failed to initialize media library errors.
   • Starting. line.
   • FPS lines for each expected udp_sinkN and fpsdisplaysink_sensor0_sinkN reaching their target framerate.
   • Stopping. at the end.

6. Tell the user how to run it themselves. Print the full one-liner including the env var, the host-side UDP ports (computed via base_port + sink_num*2), and any app-specific metadata ports (e.g. ZMQ for detection metadata).

Rollback

If the smoke test fails (paths below are an example — substitute the actual app/config paths):

ssh root@<board> '
  cp <board_app_path>.bak <board_app_path>
  cp <profile_dir>/application_settings.json.bak <profile_dir>/application_settings.json
  cp <profile_dir>/<profile>.json.bak <profile_dir>/<profile>.json
'

Gotchas

• scp overwrites without warning. Always back up first.
• The metadata content_hash will reject your edits unless you set HAILO_MEDIA_LIB_SKIP_METADATA_CONFIG_VALIDATION=1. Tell the user this isn't a workaround for production — for a long-lived deployment, regenerate hashes with the Tuning tool.
• Don't push to /usr/ unless you know what you're doing — it can shadow the board image's libs and break unrelated tools. Apps and resources live under /home/root/apps/; configs under /etc/imaging/cfg/.
• Don't run a HEF whose postprocess .so isn't on the board. If pushing a custom postprocess, push the .so to /usr/lib/hailo-post-processes/ too.

Offer to run

When this skill's work is complete, ask the user (AskUserQuestion) whether they want to run the app now and see it live. If yes, invoke the /run-app skill.