There is a region in index builds with no WriteConflictException handling. This can cause exceptions to escape, fail the index build, and crash with the following exception:

Invariant failure {"expr":"status.isA<ErrorCategory::Interruption>() || status.isA<ErrorCategory::ShutdownError>()","msg":"Unexpected error code during index build cleanup: WriteConflict: WriteConflict error: this operation conflicted with another operation. Please retry your operation or multi-document transaction.","file":"src/mongo/db/index_builds_coordinator.cpp","line":2475}

This exception can be propagated from here while inserting keys from the external sorter into the bulk loader.

The nature of the WriteConflict at this point cannot be a logical write-write conflict, because the index builder has exclusive access to the table at this point. It would have to be caused by WiredTiger forcing old transactions to roll-back in the event of cache pressure.