Software Sushi · Getting started

Evidently Data Drift packaged by Software Sushi

Compare tabular datasets, inspect local drift reports, and export HTML and JSON with Evidently 0.7.23.

This is paid repackaged open-source software. Software Sushi charges for its packaging and setup. Evidently is available separately at no charge. Software Sushi is independent of Evidently AI.

1. Launch and connect

Choose t3.medium or larger, 4 GiB RAM and at least 32 GiB root storage. Use x86_64, encrypted storage and IMDSv2. No GPU, LLM key, model download or Evidently account is required. Software costs USD 0.03 per running instance-hour on each offered instance type. EC2, EBS and data transfer are billed separately.

For Systems Manager Session Manager, attach an instance profile with AmazonSSMManagedInstanceCore permissions and provide outbound access to the Systems Manager endpoints. Session Manager initially opens as ssm-user; the commands below switch to ubuntu where needed. Alternatively, use SSH as ubuntu with your own key and approved administrative network access. Keep inbound application ports closed. Wait for first boot, then run the supplied example:

sudo cloud-init status --wait
sudo -iu ubuntu sushi-evidently status
sudo -iu ubuntu bash /opt/sushi/docs/demo.sh

2. Open the monitoring UI

The native UI and API listen at 127.0.0.1:8000. Forward instance port 8000 to your computer through an authenticated SSH tunnel, or Session Manager with the AWS CLI and Session Manager plugin. The instance login banner prints a command bound to that instance. Open http://localhost:8000; if local port 8000 is occupied, choose a different local port in both the tunnel and browser.

Open the project created by the demo, select Reports, then View beside its controlled tabular drift report. A new project dashboard starts empty until you add monitoring panels; the example report is on the Reports tab. The amount feature shifts strongly, while stable stays unchanged.

The helper loads the write token from ~/.config/sushi/evidently/access.json, mode 0600, without printing it. If the native UI requests a secret for a write, enter the token through your authenticated session. Keep it out of URLs, logs, shell history, screenshots and support messages.

3. Understand and recheck the example

The example writes two 200-row CSV files. Reference amount is 0 through 199; current amount is 1000 through 1199. Both stable columns are 0 through 199. It calculates two-sample Kolmogorov–Smirnov tests at a 0.05 threshold, then independently rereads CSVs, exact expected p-values, saved reports, API widgets and HTML exports.

Each new demo writes a dedicated directory under ~/evidently-results/. Outputs include CSVs, JSON, HTML and a SHA-256 manifest. To reread a saved result, set MANIFEST to the exact manifest path printed by your demo, then run:

sushi-evidently read "$MANIFEST"

This command reads an existing result. It does not create another report. The controlled example does not choose suitable thresholds or prove model quality for your own data.

4. Use your data and retain reports

Use /opt/evidently/venv/bin/python as ubuntu with evidently.Report, evidently.metrics.ValueDrift, and Pandas DataFrames. The installed guide at /opt/sushi/docs/README.md describes native workspace access, exports, token handling and Python examples.

Projects persist under /var/lib/evidently/workspace across service and instance restarts. There is no automatic expiry or external backup. Deleting demo files does not remove the workspace report; deleting the root volume can remove all workspace data. Monitor disk use and export reports deliberately.

For a consistent backup, stop evidently.service and back up /var/lib/evidently, /etc/evidently and the customer access file together, then restart. Preserve ownership and modes during restore. Encrypt backups because they contain the write token. Do not regenerate identities to work around a failed first boot.

5. Diagnose a problem

sudo systemctl status sushi-evidently-firstboot.service evidently.service
sudo journalctl -u sushi-evidently-firstboot.service -u evidently.service --since '15 minutes ago'
sudo -iu ubuntu sushi-evidently status

Check missing files, permissions and disk space. Keep failed task outputs for diagnosis. Review logs before sharing and remove customer data or credentials you added. Evidently Cloud, managed alerting, high availability and upstream commercial support are not included.

For AMI packaging and setup questions, contact info@softwaresushi.com.