GPDB 6.4 goes into recovery mode by executing queries included with a user defined function of PL/R

Products

• VMware Tanzu Greenplum

Issue/Introduction

Master process got a PANIC and the database goes into recovery mode with the following error message when running queries included with user defined function of PL/R:

2021-06-30 15:40:54.423740 KST,"etluser","bdagpdb",p241108,th1191909504,"0.0.0.0","57951",2021-06-30 15:40:45 KST,0,con288,cmd11,seg-1,,,,sx1,"LOG","00000","execute <unnamed>: select array_outlier_check(array[0.17830450565958245::double precision,0.1015089255490409,0.17129999686423503,0.11107548206268802],    1.2686413983454440)",,,,,,"select array_outlier_check(array[0.17830450565958245::double precision,0.1015089255490409,0.17129999686423503,0.11107548206268802], 1.2686413983454440)",0,,"postgres.c",2740,
~~ snip
2021-06-30 15:40:54.645472 KST,,,p241108,th0,,,2021-06-30 15:40:45 KST,0,con288,cmd12,seg-1,,,,,"PANIC","XX000","Unexpected internal error: Master process received signal SIGSEGV",,,,,,,0,,,,"1  0x7fc34367e5d0 libpthread.so.0 <symbol not found> + 0x4367e5d0
2  0x7fc3388ba7a3 ??:0
3  0x7fc3388bd64d ??:0
4  0x7fc3388b9126 ??:0
5  0x6d06ae postgres <symbol not found> (execQual.c:2162)
6  0x6d5cdd postgres ExecProject (execQual.c:6333)
7  0x6f4efd postgres ExecResult (nodeResult.c:218)
8  0x6c88a8 postgres ExecProcNode (execProcnode.c:970)
9  0x6bfe39 postgres <symbol not found> (tuptable.h:159)
10  0x6c0bac postgres standard_ExecutorRun (execMain.c:952)
11  0x88f647 postgres <symbol not found> (pquery.c:1152)
12  0x891631 postgres PortalRun (pquery.c:999)
13  0x88dfd8 postgres PostgresMain (postgres.c:2765)

The problematic query is shown below and plr-3.0.3-gp6-rhel7-x86_64.gppkg has been installed on GPDB v6.4 in order to run these queries and functions:

CREATE FUNCTION public.array_outlier_check(var1 double precision[], var2 double precision) RETURNS double precision[]
    LANGUAGE plr NO SQL
    AS $$
        var <- ifelse(var1>var2,1,0)
        return(var)
$$;
ALTER FUNCTION public.array_outlier_check(var1 double precision[], var2 double precision) OWNER TO etluser;
--
-- Name: array_outlier_check(double precision[], numeric); Type: FUNCTION; Schema: public; Owner: etluser
--
CREATE FUNCTION public.array_outlier_check(var1 double precision[], var2 numeric) RETURNS numeric[]
    LANGUAGE plr NO SQL
    AS $$
    var <- ifelse(var1>var2,1,0)
    return (var)
$$;
ALTER FUNCTION public.array_outlier_check(var1 double precision[], var2 numeric) OWNER TO etluser;

This article will discuss the cause and resolution for this issue.

Environment

Product Version: 6.4
OS: RHEL or CentOS 7.x

Resolution

The cause of thus issue is a bug in PL/R when converting the R object to a numeric array. We recommend updating PL/R to v3.0.4 where this issue has been fixed.

For the workaround, the return type should be changed from numeric[] to float8[] in this case if were not able to right away. For example:

From
CREATE FUNCTION public.array_outlier_check(var1 double precision[], var2 numeric) RETURNS numeric[]

To
CREATE FUNCTION public.array_outlier_check(var1 double precision[], var2 numeric) RETURNS float8[]