Skip to content

Remote Delivery Verification

Why independent verification matters

Three different facts must not be collapsed into one status:

  1. files were uploaded to the MDS FTP buffer;
  2. the DNT response confirms validation/ingestion;
  3. the expected files are actually visible in the Marine Data Lake.

The dashboard should therefore expose them separately.

Validation methods

DNT response

The strongest immediate application-level evidence is the DNT response. A successful delivery should show successful validation and ingestion with no error condition.

Copernicus Marine Toolbox (CMT)

CMT can be used to inspect the published dataset and perform dry-run/listing operations. It is useful for validating what Copernicus Marine exposes through its supported client interface.

Public S3-compatible listing

The Marine Data Lake native storage can also be checked through the public CloudFerro S3-compatible endpoint. Anonymous listing can independently confirm that the expected analysis/forecast files exist under the dataset/TAG/year/month hierarchy.

The S3 ETag must not automatically be treated as an MD5 checksum; object-storage multipart semantics can differ.

File Browser

The product File Browser provides another human-readable validation path.

Verification sequence

flowchart LR
    A[FTP transfer complete] --> B[DNT uploaded]
    B --> C[DNT response]
    C -->|Ingested=True| D[MDS application success]
    D --> E[CMT / public S3 check]
    E --> F[PUBLICATION_VERIFIED]

Expected NRT content model

Operationally, the published native area should normally contain:

  • 20 forecast files from the current cycle 00 bulletin;
  • 20 forecast files from the current cycle 12 bulletin;
  • the rolling analysis history, approximately two years.

Unexpected extra forecast files can indicate failed cleanup after an incident or overlapping deliveries.

Timing

Publication is not necessarily instantaneous. Operational material indicates that visibility in the Marine Data Lake can lag the initial upload by up to roughly 1.5 hours.

The verifier should therefore use a reasonable observation window and should not mark a cycle failed solely because public visibility lags a successful DNT response.