Building mosaic datasets using PostgreSQL Enterprise database

[EliseNilsen]

We have, until recently, been working on-premise with Microsoft SQL Server, using File Geodatabases for building and sharing mosaic datasets. Now, we are trying to move the workflow over to PostgreSQL Enterprise Database on Amazon Web Service but are encountering some issues.

The SDE consists of a mosaic dataset and a referenced mosaic dataset. The rasters are added to the mosaic dataset using UNC paths and are located in the same PostgreSQL Enterprise database as the mosaic datasets. Both the path to the database containing the mosaics and the rasters and the database itself are
registered with ArcGIS Server Manager.

When analyzing the mosaic dataset, I get an error saying that "The mosaic dataset item references files that are inaccessible."

In the Repair Mosaic Dataset, the UNC path to the database containing the mosaic appears as broken. The mosaic visually looks fine in ArcGIS Pro, but when sharing the referenced mosaic as a web layer, the images get this checkerboard pattern, as often happens if DOS paths are used when adding rasters to mosaic datasets and published.

Is there a different practice when building mosaic in a PostgreSQL Enterprise database than in a regular File Geodatabase? What can I do to make the rasters added to the source mosaic dataset to be recognized?

[VinceAngelo]

Are you running an RDS in AWS or PostgreSQL in a Windows EC2 with file sharing?

If it's the former, then this won't work, since an RDS doesn't have the ability to share a UNC path. If it's the latter, you probably ought to be using an RDS and an S3 bucket.

- V

[EliseNilsen]

Hi @VinceAngelo  and thank you for your reply!

We run PostgreSQL in an EC2 and have file share on another EC2. Why is it not preferable to use PostgreSQL in a Windows EC2 with file sharing? Regarding RDS in AWS and S3, do you have some tips on how we best configure that to host mosaic datasets in PostgreSQL?

Elise

[VinceAngelo]

It's more expensive to allocate an EC2 to run a PG instance than to use an equivalent RDS. S3 is much less expensive (albeit with higher latency) than EBS (SSD-backed) storage. Performance (and functionality) of a file share from an EC2/EBS volume is highly dependent on the location of the client (in the same AWS enclosure? in a different one? across a WAN?). 

The mosaic dataset only really stores the metadata, not the images themselves, which are the vast majority of the storage requirement (probably 99.9+%). I don't have a ton of experience with MDS using RDS/S3, but I'm sure there's documentation and others with more experience for best practice recommendations.

[EliseNilsen]

Thank you for your answers @VinceAngelo! Unfortunately, we have not been able to locate relevant documentation about hosting and publishing mosaic datasets from PostgreSQL (with the mosaic datasets and source rasters in the same Enterprise database). Hope others with experience will jump on this thread.