Maybe you just discovered by yourself what Aaron described in a previous message (Message-Id: <1405173409302.ce95d5ef@Nodemailer> on logstash mailing list):
> [deleting a type of documents from an index] is a horrible idea in the same way that running DELETE FROM TABLE WHERE TYPE='cisco' AND DATE<... in a SQL database is a bad idea. You get thousands, if not millions of individual deletes to manage, and they are horrible in terms of disk I/O performance to the system and the cluster. Worse still, in elasticsearch, the deletes aren't immediate. They are only flagged for deletion. The actual delete takes place during the next segment merge, which is up to 30 minutes later. It could render you with horrible I/O during the seek as well as the subsequent delete.