Trimming of the draftcndb index reaches a change that contains a cookie that doesn't include anything from the same server-id. eg:
This causes the inner loop in the trimming code to never delete anything, and the trimming loops terminate. The next pass through the trimming code starts with the same change, and the trimming code will perpetually not match it.
The result is the changelogdb will grow until disk space is exhausted. Searching of cn=changelog gets slower and slower as the size of the database grows.