Master process crash with SIGSEGV during CMDCache::Reset and ordered-set aggregate

Products

• VMware Tanzu Greenplum
• VMware Tanzu Greenplum / Gemfire
• VMware Tanzu Data Suite
• VMware Tanzu Data Suite

Issue/Introduction

A database panic occurs with a SIGSEGV signal when executing queries utilizing the GPORCA optimizer. This happens specifically when an ordered-set aggregate (such as percentile_cont, percentile_disc, or median) is used in a Common Table Expression (CTE). 

Coordinator logs will display a PANIC and stack trace similar to following:

"PANIC","XX000","Unexpected internal error: Master process received signal SIGSEGV",,,,,,,0,,,,"
1    0x7f45870bd990 libpthread.so.0 <symbol not found> + 0x870bd990
2    0xe5cb31 postgres _ZN4gpmd15CMDFunctionGPDBD1Ev (CMemoryPool.h:240)
3    0xfdb133 postgres _ZN5gpopt8CMDCache5ResetEv (CMemoryPool.h:240)
4    0x116dac5 postgres _ZN9COptTasks12OptimizeTaskEPv (COptTasks.cpp:996)
...
10   0x116ae8e postgres _ZN9COptTasks18GPOPTOptimizedPlanEP5QueryP11SOptContext (COptTasks.cpp:1334)
...

Environment

Greenplum 7.x prior to 7.8.4

Cause

This issue is caused by a double-free memory bug in the GPORCA optimizer.

Resolution

This defect is resolved in Tanzu Greenplum version 7.8.4.

To resolve the issue:

1. Upgrade the environment to Tanzu Greenplum 7.8.4 or later.

2. Review the Tanzu Greenplum Release Notes for further details.