feat(ns-plug): dual-send backup to my-new proxy

[edospadoni]

Summary

During the migration window, mirror every configuration-backup upload to the new my.nethesis.it platform via the translation proxy, in addition to the existing upload to the legacy backupd.nethesis.it. The mirror lives in remote-backup's upload action so it covers every backup path (the nightly send-backup cron and manual backups triggered from the UI), and so the eventual cutover is a single localized change to the same file.

Context

The new my.nethesis.it platform ships a first-class configuration-backup subsystem (ingest on collect, managed reads on backend, S3-compatible storage) — see the referenced issues. During the migration window NethSecurity keeps uploading to the legacy backupd and dual-sends to my-new via a translation proxy (nethinfra role). The proxy accepts the existing system_id:secret Basic Auth pair and translates it into the my-new system_key:system_secret as it forwards the request to collect, so nothing changes on the appliance in terms of UCI config or registration.

What changes

packages/ns-plug/files/remote-backup, upload action only:

• After the primary curl ... --upload-file to the legacy backupd completes, its exit code is captured (rc=$?).
• If ns-plug.config.type == enterprise, the same encrypted blob is POSTed to https://my.nethesis.it/proxy/backup with --data-binary "@$file", reusing the existing $curl_args (--silent --location-trusted --user $SYSTEM_ID:$SYSTEM_SECRET — the proxy translates that legacy pair). Content-Type: application/octet-stream and X-Filename are set so the blob and its user-facing name land as proper S3 object + metadata on my-new.
• The mirror is best-effort (|| :): a proxy outage must not fail the primary upload that already succeeded against backupd.
• The script then exit $rc — the primary backupd upload's exit status is what propagates to the caller, so send-backup's set -e still aborts on a real legacy-upload failure and the md5 dedup marker is not advanced on a failed run (the backup is retried next night). The best-effort mirror never masks a primary failure.

packages/ns-plug/files/send-backup is unchanged (identical to main). Putting the dual-send in the transport layer (remote-backup) instead of the cron wrapper is what the review asked for, and it also means a backup triggered manually from the UI is mirrored too, not just the cron one.

Verified end-to-end on a real NethSecurity appliance: script runs under busybox/ash, $curl_args reuse works, a failing primary still fires the best-effort mirror, rc/exit $rc propagates the primary status, and the mirrored blob lands intact on my-new (matching SHA-256, X-Filename → object metadata, application/octet-stream content type).

Not included

• Rewriting the legacy upload path — out of scope; happens in a later cycle once the new platform is promoted (the cutover then replaces this best-effort mirror block with the direct collect call, a single-file change to remote-backup).
• MY registration on the appliance — not needed: the proxy absorbs the credential difference.

Refs: NethServer/my#82 NethServer/my#83

[gsanchietti]

The change should be done inside the remote-backup script, under the upload action