We don't ensure that the optime of the failing operation is logged, which can make remediation difficult during crash loops (where we need to find the exact oplog entry that is causing the crash). We should make sure that the optime is always logged, regardless of redaction.