Aggregations: DateHistogram with negative 'pre_offset' or 'post_offset' value ends with "Message not fully read (response) for"

[marcelog]

Hey guys,

We're using version 1.3.2 and Oracle JRE: 1.7.0_67. The cluster has 3 nodes (2 data nodes and 1 used only to route searches).

There is a separate (different named) cluster of 1 node using marvel to collect statistics from the first one. No other nodes in the network, and no custom/manually made java clients around.

As the title says, when doing a DateHistogramAggregation using a negative value for either pre_offset or post_offset yields the message:

" [transport.netty ] Message not fully read (response) for [69208] handler org.elasticsearch.search.action.SearchServiceTransportAction$6@f58070b, error [false], resetting" in the logs.

Note that every other query works flawlessly. This is only reproduceable when using the arguments pre_offset and/or post_offset in a DateHistogram with a negative value and with more than 1 node in the cluster.

Facts so far:

• Gist with the output of _nodes?jvm=true&pretty.
• Gist with sample query
• Same query, but without negative values in pre_offset and post_offset works (it doesn't return the results we expect, of course, but no errors/warnings are shown in the logs)
• This previous issue doesn't seem to be related, there are no other nodes in the network and there are no "custom" java clients around either.
• The sample query works when there is only one node in the cluster and we start to get these error messages when adding more nodes (2 will suffice to reproduce the issue).

Any ideas? In the meantime, we solved the issue by using pre_zone and post_zone instead of pre_offset and post_offset. Negative values are ok, and everything is running smoothly with those (we tried index, search, and snapshot operations). No error messages in the logs.

Thanks in advance,

[marcelog]

Wow that was really fast! Thanks :)