ArcGIS Enterprise on AWS crashes when attempting to export image from image service

NCOneMap · ‎03-28-2022

We have a successful Enterprise deployment in AWS. I have embedded raster proxies (that access image tiles in AWS S3) into a local mosaic dataset and can view them as expected in Pro. I have an identical mosaic dataset that is in an enterprise geodatabase and is used for an image service (v10.9.1). The only difference is that the path to the cache directory is different (z:\mrfcache). The service published correctly. However, when I try and view the service in a map, the raster proxy cache tile is never created and the ArcGIS Server logs indicate an error while trying to complete the ExportImage function before the service crashes (a dump file is also created which has been sent to Esri). All of the vector services work fine. My intuition is thinking a permissions issue but unsure of what to check. Esri tech support has indicated that they wouldn't help in this matter so any help is greatly appreciated. @VedantBajaj

https://services.gis.nc.gov/secure/rest/services/Elevation/DEM20ft_DEM/ImageServer

NCOneMap · ‎03-28-2022

Update: The service draws correctly when accessing the imagery in S3 directly (i.e., /vsis/ paths).

JonEmch · ‎03-31-2022

Could you DM me the case number of this issue so that I can get more context? Thank you!

Keep on keeping on!

mbullEsriUK · ‎02-19-2023

Hi. Did you ever manage to identify the issue with this? I think I am hitting the same problem. I deployed a standalone ArcGIS Image Server and JumpBox using ArcGIS Enterprise CloudBuilder for Microsoft Azure (v10.9.1). On the JumpBox I installed ArcGIS Pro (v2.9.6) and created my fGDB and Mosaic Dataset specifying the pixel cache as d:\mrfcache - I can see the cache being created on the JumpBox VM. When I publish the Mosaic Dataset as an Image Service, I get the same error you were seeing, no image is returned and no cache is created. If I create the same Mosaic Dataset without specifying a pixel cache directory and publish it, the service works.

NCOneMap · ‎02-20-2023

Yes, after much discussion with Esri they identified it as a bug (BUG-000148116) and ultimately provided a hotfix (S-1091-HF-000005735). I haven't confirmed it but I am hoping that they have incorporated this correction into one of their later 10.9.1 machine images so that we can integrate it directly into our autoscaling environment. Currently, I am having to run a script after a scale out event to install the hotfix which is not ideal.

mbullEsriUK · ‎02-20-2023

Thanks for taking the time to respond, it's greatly appreciated. I had some discussions internally today and we came to the same conclusion that it was that BUG and Hotfix.