Server Dashboard Guide

Go to Hub Dashboard → your experiment → Details. The current PIN is shown here. You can also set a new PIN from this page at any time without needing to know the old one. This is the easiest option if you have forgotten your PIN.

While logged into the server dashboard, go to Settings → PIN. You can change the PIN here by entering your current PIN and choosing a new one. This option requires knowing the current PIN.

The Form tab provides a graphical interface for defining a parameter grid. Each row represents one hyperparameter:

1. Enter a parameter name (e.g., learning_rate).
2. Add the values you want to test, separated by commas or one per field.
3. Click + Add Parameter to add more rows.

The dashboard computes the total number of combinations (Cartesian product of all values) and shows a live preview of a few example jobs before you submit.

The Replace existing jobs checkbox determines behaviour on submission:

• Checked: All existing PENDING jobs are deleted and replaced with the new set. Running and Done jobs are unaffected.
• Unchecked (default): New jobs are appended to the existing queue.

The JSON tab accepts a plain JSON object where each key is a parameter name and each value is an array of options. The server generates all combinations automatically.

{
  "learning_rate": [0.001, 0.01, 0.1],
  "hidden_units":  [64, 128, 256],
  "dropout":       [0.0, 0.3, 0.5]
}

This creates 3 × 3 × 3 = 27 jobs, one per combination.

Format button: Automatically pretty-prints and re-indents your JSON.

Validate button: Checks the JSON for syntax errors and confirms the number of jobs that will be created before you submit.

Live preview: A sample of the first few generated jobs is shown below the editor so you can verify the combinations look correct.

The JSON tab is the most efficient method for large grids — you can paste a grid directly from your experiment notes or a script without clicking through individual form fields.

Below the Add Jobs section, every job is listed in a table with the following columns:

Click on any row to expand the job detail view, which shows the full parameter list, stdout/stderr output, and any uploaded files.

Use the filter bar above the job list to narrow the view:

• Status filter: Buttons to show All, Pending, Running, Done, or Aborted jobs only.
• Search: Type any text to filter jobs by parameter name or value. For example, typing 0.01 shows only jobs where any parameter equals 0.01.

Only PENDING jobs can be edited — jobs that have not yet been assigned to a worker. Once a worker picks up a job, it moves to RUNNING and can no longer be modified.

To edit a job:

1. Find the job in the job list (use the filter to show PENDING jobs).
2. Click the Edit button in the Actions column.
3. Modify the parameter values in the edit form.
4. Click Save. The change takes effect immediately.

Use this to correct a typo in a parameter value or adjust a combination before it runs — without having to delete and recreate the entire job set.

Only unassigned (PENDING) jobs can be deleted. Jobs that are currently RUNNING, DONE, or ABORTED cannot be deleted.

When your entry script calls jd_upload(file_path), the file is stored on the server and associated with that job. The server dashboard provides a direct interface for browsing and downloading these files.

To access uploaded files:

1. Click on a job row in the job list to expand its detail view.
2. Scroll to the Uploaded Files section.
3. Click any filename to download it directly to your browser.

The last portion of your script's standard output and error output is captured and stored per job. It is visible in the job detail view under Output. This is particularly useful for diagnosing ABORTED jobs without needing to SSH into the worker machine.

For the full output of any job, check the worker log file on the machine where the job ran:

~/jd_data/<exp_id>/jd_worker_logs/jd_worker_<process_id>.log

Open the Workers Management nav tab. Two sub-tabs:

• Active Workers — currently polling; drain, stop, and resume are available.
• Disabled / Stopped — workers that stopped polling (shutdown, crash, or network loss). Read-only: view details and history only.

Filter the table with three dropdowns at the top: Host, Instance, and Slot (parsed from each worker_id like gpunode_A3f9X2_0). Select a host to browse all workers on that machine; add instance and slot to narrow further.

Click Details on any row to open:

• Info — host, instance, slot, job, poll times
• History — audit trail (newest first): registration, jobs, dashboard commands, disable events
• Metrics — latest system metrics (same view as jobs), plus snapshots from history when captured

When a worker stops polling for ~10 minutes (machine powered off, process killed, network loss), the server moves it to Disabled / Stopped. A history entry records why. Disabled workers cannot receive dashboard commands; use Details to review their full audit trail and last known metrics. Restart workers on a live machine with jd_worker_cli — they register as new active workers.

Commands are queued on the server and delivered on each worker’s next poll (about every 3 minutes, or right after the current job finishes). They do not kill processes instantly from the browser.

Use toolbar buttons for all workers, or per-worker buttons in the table. Precedence is stop > drain > run.

After you queue drain or stop, the status badge shows queued … until the worker polls. If you clicked the wrong button, use Cancel (per worker, per host, or “Cancel all pending”) to revert before the command is applied.

Once applied, the worker must be started again from the cluster (jd_worker_cli) — the dashboard cannot restart processes on remote machines.

You can update the 6-digit dashboard PIN from the Settings page. This requires entering your current PIN to authorize the change.

If you have forgotten the current PIN, go to the Hub Dashboard → experiment details to view and reset it from there (no current PIN required).

1. Navigate to Settings in the dashboard nav.
2. Enter your current PIN in the first field.
3. Enter the new PIN twice to confirm.
4. Click Update PIN.

The new PIN takes effect immediately. You will be asked to enter it the next time you open the dashboard in a new browser session.