[IMPALA-1928] Impala ODBC bad performance with Kerberos mechanism

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Impala 2.1.1, Impala 2.3.0
• Fix Version/s: Impala 2.6.0, Impala 2.5.2
• Component/s: Perf Investigation
• Labels: odbc
• Environment: ODBC Impala 5.2.20 and 5.2.23
• Target Version: Impala 2.6.0, Impala 2.5.2
• Assignee: Mostafa Mokhtar
• Reporter: Jonathan L.
• Created: 31/Mar/15 21:44
• Updated: 30/Jun/16 00:35
• Resolved: 17/May/16 17:16

Description

Using Windows Impala ODBC Driver with no auth mechanism, we are able to run a query that produces about 80millions rows and it finishes in about 20 minutes.

The same query using auth mechanism kerberos (against secured cluster CDH 5.3.1) is taking about 90min.

I've been able to reduce this total time of 90min by 1/3 of the time by increasing the "Rows fetched per block" in the Advanced Options of the ODBC driver from 1000 rows to 100000 rows. This is the best I could get. In short it seems that an ODBC with no auth mechanism configured is faster than with kerberos mechanism.

When running the same query against both cluster (1x with security enabled and one without) the performance is the same. Both are able to complete within aprox 20min.

What can be done in order to improve the performance of the Impala ODBC running with kerberos mechanism?

Issue Links

• is related to: IMPALA-3080 Slow performance on kerberos cluster - Possible inefficient implementation of TSaslTransport::write