Error 'read beyond eof in table' when running SELECT query on partitioned table

Products

• VMware Tanzu Greenplum

Issue/Introduction

When attempting a select on a partitioned table, the error "read beyond eof in table XXXXX" below is seen:

2023-10-11 16:57:15.191445 PDT,"pivotal","test1",p3241626,th-1132627584,"[local]",,2023-10-11 16:57:14 PDT,0,con86,cmd2,seg-1,,,,sx1,"ERROR","XX000","read beyond eof in table ""test_vacuum_prt_1_prt_3"" file ""base/219425/118627.1"", read position 0 (small offset 0), actual read length 0 (large rea     d length 65536) (cdbbufferedread.c:202)  (seg0 slice1 127.0.1.1:6002 pid=3241628) (cdbbufferedread.c:202)",,,,,,"select count(*) from test_vacuum_prt",0,,"cdbbufferedread.c",202,"Stack trace:
1    0x55ac7eb62b0d postgres errstart + 0x473
2    0x55ac7ec376a5 postgres cdbdisp_get_PQerror + 0x12b
3    0x55ac7ec374b5 postgres cdbdisp_dumpDispatchResult + 0x5b
4    0x55ac7ec3794d postgres cdbdisp_dumpDispatchResults + 0x104
5    0x55ac7ec332cb postgres cdbdisp_getDispatchResults + 0xc2
6    0x55ac7ec3348b postgres CdbDispatchHandleError + 0x1ab
7    0x55ac7e775382 postgres mppExecutorCleanup + 0xef
8    0x55ac7e7532f5 postgres standard_ExecutorRun + 0x5b5
9    0x55ac7e752d3d postgres ExecutorRun + 0x54
10   0x55ac7e9d5986 postgres <symbol not found> + 0x7e9d5986
11   0x55ac7e9d5576 postgres PortalRun + 0x30a
12   0x55ac7e9cc756 postgres <symbol not found> + 0x7e9cc756
13   0x55ac7e9d2376 postgres PostgresMain + 0xe31
14   0x55ac7e92a641 postgres <symbol not found> + 0x7e92a641
15   0x55ac7e929b6e postgres <symbol not found> + 0x7e929b6e
16   0x55ac7e925279 postgres <symbol not found> + 0x7e925279
17   0x55ac7e924575 postgres PostmasterMain + 0x1568
18   0x55ac7e8052a1 postgres <symbol not found> + 0x7e8052a1
19   0x7fc3bf395d90 libc.so.6 <symbol not found> + 0xbf395d90
20   0x7fc3bf395e40 libc.so.6 __libc_start_main + 0x80
21   0x55ac7e449205 postgres _start + 0x25

This error occurs when a vacuum on one of the partitions is concurrently running during the time of the SELECT statement.

For more info see: https://github.com/greenplum-db/gpdb/issues/16545

Environment

Product Version: 6.23

Resolution

As of October 30, 2023, fix is still pending.

Current workarounds are to:

• avoid vacuum of partition tables during select
• enclose select in a begin/end block, including a table lock beforehand:

  begin;
  lock table foo in access share mode;
  select count(*) from foo;
  end;