STAC ICD for PRIP can be downloaded from this link.
Below table is available to collect the comments on this version of the document.
| Id. | Author | Section/Page number | Title | Comment | Answer |
|---|---|---|---|---|---|
| 1 | m-mohr | p. 7 | EOPF_SPECIFICATION | Missing link, there's a <tbd> instead. | |
2 | CloudFerro | p. 21 | REQ-0050 | timestamp precision should be defined to avoid differences between PRIP implementations | Proposal to keep the same precision as for CDSE (second) |
| 3 | CloudFerro | p. 25 | REQ-0085 | thumbnail size and resolution must be defined to create consistency between product types. if no thumbnail is generated by PS then this asset should be empty or removed. whatever is included in the products and does not fit the requirements should be listed as asset type "preview" or similar. | See higher level document : STAC Requirements and Recommendations for EOF Services |
| 4 | CloudFerro | p. 25 | REQ-0090 | there are various media types within ZARR, so setting mime type of all objects to "application/vnd-zarr" may cause some strange behaviour of standard client libs | |
| 5 | CloudFerro | p. 28 | REQ-0207 | it is not needed for PRIP to enable searching by any common attribute, limited subset will suffice. PRIP API is not for end-users and used only for systematic downloads | ESA asked for this level of searching capabilities |
| 6. | CloudFerro | p. 29 | REQ-0305, REQ-0410 | multipart uploads to object storage may affect the object checksum so in some cases it may be unreliable (TBC) with multipart upload single-part upload md5 != multipart-download md5 | It's up to client's implementation and should be adressed by "tailored ICD" |
| 7 | CloudFerro | p. 31 | REQ-0391 | no relation between accounts in central identity and s3 credentials make the management of access prone to errors and delays in provisioning of access. s3 credentials should be self-managed and reflect access roles (similar to CF S3 keys manager) | No known standard to include s3 credentials into IAM → out of scope of this ICD. It must be described by infrastructure provider. |
| 8 | CloudFerro | p. 31 | REQ-0400 | regarding: " PRIP client point of view, retrieving an asset needs to: List all S3 objects sharing the prefix provided by the asset link, leveraging the ListObjectsV2 S3 operation "prefix listing is time consuming and might be expensive (and will raise the maintenance cost of the PRIP) for multiple requests and will affect timeliness. compressed product recommended. | Question already discussed with ESA, and submitted to its arbitration. In which way it can raise the maintenance cost ? |
| 9. | CloudFerro | STAC-PRIP-API-REQ-0210 | Access control to the STAC Catalogue is implemented at the Collection level while CDSE manages it on the product_type level. How collection and product types will be joined? | One collection corresponds to a unique product_type. Besides, PRIP is not intended for for end-users, thus access control is less critical. | |
| 10. | CloudFerro | STAC-PRIP-API-REQ-0340 | Quota Management expressed as n terra bytes per month <- what if a reprocessing campaign is published. Should every time the quota should be adjusted? | Quota adjustement should be managed at operations level and arbitrated by ESA. Not to describe in the ICD | |
| 11. | CloudFerro | lack of requirement regarding the maximum allowable number of items returned from a STAC query | We propoe to add an additional requirement on the maximum number of returned items as configurable | ||
| 12. | CloudFerro | p. 22 | publication date - Timestamp of the availability of the current Item in the PRIP instance Catalogue. update date - If STAC Item properties have been updated, timestamp of the last update. Regarding the change of the modification date it is not clear if the LTA/CDSE should re-download the file. | ||
| 13 | CloudFerro | File extension should be if the zipped/concatenated product is to be exposed. The file:size and file:checksum https://github.com/stac-extensions/file | Zarr item is provided in extended form (folders & sub-folders) → neither file extension nor checksum are relevant here | ||
| 14 | CloudFerro | Section 3.4.3.4 | Downloading ZARR content | More info needed regarding "Note: It is assumed that the AWS CLI is configured with the expected S3 API endpoint and authentication credentials." | Should depend on the provider → not candidate for the ICD. rather provided in tailored ICD |
| 15 | Cloudferro | Bulk download via a single zipped file without compressions. | Downloading multiple small zarr chunk files will generate huge overload regarding number of network requests and md5/checksums checks. | Maybe we could adapt chunks size for storage usage, in order to reduce the number of objects | |