Queries with large result sets and multiple _revinclude parameters result in Internal server error

[klimkjar]

Describe the bug
When performing a search with more than one _revinclude search parameter and _count set to a largish value (on our server anything 100 or above seems to be enough), the server returns 500 Internal server error. Reproduced in two of our testing environments with different datasets.

The exception in the AI logs states 'The count must be greater than zero.' A stack trace is attached.

FHIR Version?
R4

Data provider?
CosmosDB

To Reproduce
Steps to reproduce the behavior:

1. Perform a search as described, for instance /Practitioner/?_count=1000&_revinclude=Contract:subject&_revinclude=Person:practitioner

Expected behavior
The server returns a set of search result, marked as Incomplete if all of the _revinclude data is not present. When only a single _revinclude statement is used, this is the behavior we see.

Actual behavior
The server returns 500 Internal server error

exception_stacktrace.txt

AB#98330

[klimkjar]

This is on version 3.3.18, and we've seen it on earlier versions as well.

[EXPEkesheth]

@klimkjar Thanks for reporting the issue. Is this a consistent repro? Can you share details on how often you see this issue?

[klimkjar]

@EXPEkesheth Yes, this is a consistent repro. I can reproduce it 100% of the time by setting _count large enough on a search with a large enough number of matches. Will try to hook up an instance running in a debugger to one of the test datasets to see if I can get more details, but a bit pressed for time so will probably only get around to it next week.

In the meantime we're working around it by requesting a lower _count or fetching related resources in separate queries.