[ https://issues.apache.org/jira/browse/DRILL-4653?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15486473#comment-15486473 ]
ASF GitHub Bot commented on DRILL-4653: --------------------------------------- Github user ssriniva123 commented on the issue: https://github.com/apache/drill/pull/518 Paul, Thanks for taking the time out in writing out a detailed email. Here are some of my thoughts. - Drill uses the com.fasterxml.jackson.core.json.UTF8StreamJsonParser for parsing of JSON records. This parser does not rely on line delimiters for record separators but instead uses the JSON structure as a natural way to signal End of record (EOR). There are methods internal to the parser which check for line feeds but is not exposed to callers. - The CountingJsonReader uses the parser.skipChildren() method to skip the rest of the children for this record, hence it is not possible to accurately count and match the no of braces to cleanly skip that bad record. - One thought is to tap the inputsource of the parser on an exception condition, but is not encouraged. My thought process was exactly along the lines you have been thinking. On an exception scenario the code attempts to locate a closing bracket(}) followed by a opening bracket ({). This is what is being done in the BaseJsonProcessor.processJSONException method. Please note that it works in all cases except when we do not have proper brackets to signify end of a JSON record. Hope this explanation helps clarify. > Malformed JSON should not stop the entire query from progressing > ---------------------------------------------------------------- > > Key: DRILL-4653 > URL: https://issues.apache.org/jira/browse/DRILL-4653 > Project: Apache Drill > Issue Type: Improvement > Components: Storage - JSON > Affects Versions: 1.6.0 > Reporter: subbu srinivasan > Fix For: Future > > > Currently Drill query terminates upon first encounter of a invalid JSON line. > Drill has to continue progressing after ignoring the bad records. Something > similar to a setting of (ignore.malformed.json) would help. -- This message was sent by Atlassian JIRA (v6.3.4#6332)