I'm pretty sure I discovered the issue here is due to a flaw with Jira pagination. When we run a query fetching all issues since the last update, we might get too many records for Jira to return at once, so it returns the first N and then indicates that there are M more records. However, it doesn't look like Jira has a way to cache the query results. So when you re-run the API call and request records starting with the N+1 record, it's possible that the results may have been updated.

For example, suppose I have 5 issues (ISSUE-1 through ISSUE-5) and we want to return all of them ordered by the updated timestamp, but we have maxResults: 3. For the first query, I submit a request with startAt: 0 and get

The API will return that isLast: False and total: 5. I then submit a request with startAt: 3 and would expect to get the next two issues. However, right before this second request is made, ISSUE-2 is updated. This would shift ISSUE-4 into the first 3 records and ISSUE-2 would show up in the next set.

The consequence is that ISSUE-4 is never seen in the paginated results.

To fix this, I believe we'd have to stop using the Paginator and instead run subsequent queries with the maximum updated timestamp of the previous query.

Using the maximum updated timestamp doesn't work either since JQL is limited to querying by the minute and it would be easy to get more than a page of data where all the records were updated in the same minute. Instead, we can query the data in descending order.