[ https://issues.apache.org/jira/browse/KAFKA-1744?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14201374#comment-14201374 ]
Evan Huus commented on KAFKA-1744: ---------------------------------- [~nehanarkhede] this was discovered in the golang consumer I maintain - the scala consumer (as I linked) seems to handle this case already. I have not checked the java consumer. The [spec for the fetch API|https://cwiki.apache.org/confluence/display/KAFKA/A+Guide+To+The+Kafka+Protocol#AGuideToTheKafkaProtocol-FetchAPI] implies (though it does not explicitly state) that if I perform a fetch request for offset X, the fetch response will contain messages whose offset is strictly >= X. If this is not true (in practice I have seen messages with offsets < X) I would suggest explicitly noting this in the spec to avoid confusion. Alternatively it may be a real bug in the broker, in which case the spec is fine and the broker should be fixed. I don't have enough information to say for sure. > Fetch Response contains messages prior to the requested offset > -------------------------------------------------------------- > > Key: KAFKA-1744 > URL: https://issues.apache.org/jira/browse/KAFKA-1744 > Project: Kafka > Issue Type: Bug > Components: core > Affects Versions: 0.8.1.1 > Reporter: Evan Huus > > As reported in https://github.com/Shopify/sarama/issues/166 there are cases > where a FetchRequest for a particular offset returns some messages prior to > that offset. > The spec does not seem to indicate that this is possible; it does state that > "As an optimization the server is allowed to return a partial message at the > end of the message set." but otherwise implies that a request for offset X > will only return complete messages starting at X. > The scala consumer does seem to handle this case gracefully though, if I am > reading it correctly (my scala is not the best): > https://github.com/apache/kafka/blob/0.8.1/core/src/main/scala/kafka/consumer/ConsumerIterator.scala#L96-L99 > So is this a bug or just a case that needs to be added to the spec? Something > like "As an optimization the server is allowed to return some messages in the > message set prior to the requested offset. Clients should handle this case."? > Although I can't imagine why sending extra data would be faster than only > sending the necessary messages... -- This message was sent by Atlassian JIRA (v6.3.4#6332)