Mailing-List: contact dev-help@flink.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@flink.apache.org
Date: Mon, 6 Jun 2016 21:08:20 +0000 (UTC)
From: "Elias Levy (JIRA)" <jira@apache.org>
To: dev@flink.apache.org
Message-ID: <JIRA.12976202.1465247255000.5840.1465247300935@Atlassian.JIRA>
In-Reply-To: <JIRA.12976202.1465247255000@Atlassian.JIRA>
References: <JIRA.12976202.1465247255000@Atlassian.JIRA> <JIRA.12976202.1465247255795@arcas>
Subject: [jira] [Created] (FLINK-4027) FlinkKafkaProducer09 sink can lose
 messages
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
archived-at: Mon, 06 Jun 2016 21:08:22 -0000

Elias Levy created FLINK-4027:
---------------------------------

             Summary: FlinkKafkaProducer09 sink can lose messages
                 Key: FLINK-4027
                 URL: https://issues.apache.org/jira/browse/FLINK-4027
             Project: Flink
          Issue Type: Bug
          Components: Kafka Connector
    Affects Versions: 1.0.3
            Reporter: Elias Levy
            Priority: Critical


The FlinkKafkaProducer09 sink appears to not offer at-least-once guarantees.

The producer is publishing messages asynchronously.  A callback can record publishing errors, which will be raised when detected.  But as far as I can tell, there is no barrier to wait for async errors from the sink when checkpointing or to track the event time of acked messages to inform the checkpointing process.

If a checkpoint occurs while there are pending publish requests, and the requests return a failure after the checkpoint occurred, those message will be lost as the checkpoint will consider them processed by the sink.


--
This message was sent by Atlassian JIRA
(v6.3.4#6332)