[
https://issues.apache.org/jira/browse/ASTERIXDB-1616?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15451212#comment-15451212
]
Wail Alkowaileet commented on ASTERIXDB-1616:
---------------------------------------------
The problem occurs with variable length fields. When you project (return) only
primitive types at level zero (the dataset type), the issue disappears.
I remember I tried to debug it and it seems that the length offset of a string
(for example) is off by one byte (which leads to read some random 4-bytes that
are way larger than the frame size. Hence, it throws IndexOutOfBoundExcpetion
sometimes). However, if you make the schema all open (key-only) this should
solve the problem. Which makes me wonder if that has anything to do with the
parser?
I suspect the problem resides in the ARecordBuilder and friends. I will try to
print the record right before flush (and after the parsing) and see if I get
the same issue.
> NPE when printing record inside open type with unicode fields
> -------------------------------------------------------------
>
> Key: ASTERIXDB-1616
> URL: https://issues.apache.org/jira/browse/ASTERIXDB-1616
> Project: Apache AsterixDB
> Issue Type: Bug
> Reporter: Ian Maxon
> Assignee: Ian Maxon
>
> DDL:
> https://github.com/kevincoakley/asterixdb_tests/blob/master/notebooks/asterixdb-spark/Count%20one_percent%20Tweets%20Spark%20Single.ipynb
> Data:
> https://object.cloud.sdsc.edu/v1/AUTH_kcoakley/asterixdblogs/2015_11_07_00_onepercent.txt
> Basically just a scan+limit on the one_percent dataset will give
> IndexOutOfBounds.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
