Not sure if you deal with hints now, but if you are, if there is a much bigger execution time (last line) than the first node total runtime, it might be because it's waiting for locks. See https://explain.dalibo.com/plan/20205991261113ga for an example. The query lasts 22 seconds for the user, but PostgreSQL finishes after 11 seconds. The remaining is lost on waiting for locks (I had a VACUUM FULL running on the table).
Not sure if you deal with hints now, but if you are, if there is a much bigger execution time (last line) than the first node total runtime, it might be because it's waiting for locks. See https://explain.dalibo.com/plan/20205991261113ga for an example. The query lasts 22 seconds for the user, but PostgreSQL finishes after 11 seconds. The remaining is lost on waiting for locks (I had a VACUUM FULL running on the table).