[CALCITE-2256] Exception in Druid Adapter: JsonSegmentMetadata, numRows field out of range

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 1.16.0
• Component/s: druid-adapter
• Assignee: Unassigned
• Reporter: SaschaC
• Created: 15/Apr/18 10:50
• Updated: 22/Feb/19 01:21

Description

When trying to connect to a Druid datasource, Calcite pulls metadata from Druid as a first action. As Druid is a database for big-data, it is not uncommon that a datasource would contain billions of records. The metadata about a datasource returns a records count, namely the total number of records residing in that datasource. This is returned in a property named "numRows".

The Calcite Druid Adapter tries to read the numRows field (that Druid returns within the response Json document of a metadata request) into an integer field which overflows for record counts bigger than 4 billion and triggers the below exception

Caused by: com.fasterxml.jackson.databind.JsonMappingException: Numeric value (13452491153) out of range of int
 at [Source: (sun.net.www.protocol.http.HttpURLConnection$HttpInputStream); line: 565, column: 26]
 at [Source: (sun.net.www.protocol.http.HttpURLConnection$HttpInputStream); line: 565, column: 15] (through reference chain: java.util.ArrayList[0]->org.apache.calcite.adapter.druid.DruidConnectionImpl$JsonSegmentMetadata["numRows"])

In class org.apache.calcite.adapter.druid,DruidConnectionImpl, inner class JsonSegmentMetadata has integer fields for reading in Druid metadata like size or numRows which aught to be of type long so that they can hold larger numbers:

private static class JsonSegmentMetadata {
  public String id;
  public List<String> intervals;
  public Map<String, JsonColumn> columns;
  *public int size;*
  *public int numRows;*
  public Map<String, JsonAggregator> aggregators;
}