[YSQL] Assertion failure: Assert(portal->status != PORTAL_DONE) triggered by unique key violation

[mtakahar]

Jira Link: DB-755

Description

Unique key violation leads to server crash when the INSERT is preceded by a PL/pgSQL block that handled non-null violation exception.

Example:

yugabyte(26555)# DROP TABLE t1;
CREATE TABLE t1 (k int PRIMARY KEY);
DO $body$
BEGIN
  INSERT INTO t1 (k) VALUES (null);
EXCEPTION
  WHEN others THEN RAISE NOTICE '%', SQLERRM;
END;
$body$;
INSERT INTO t1 (k) VALUES (1);
INSERT INTO t1 (k) VALUES (1);

DROP TABLE
yugabyte(26555)# CREATE TABLE
yugabyte(26555)# yugabyte$# yugabyte$# yugabyte$# yugabyte$# yugabyte$# yugabyte$# NOTICE:  null value in column "k" violates not-null constraint
DO
yugabyte(26555)# INSERT 0 1
yugabyte(26555)# server closed the connection unexpectedly
	This probably means the server terminated abnormally
	before or while processing the request.
The connection to the server was lost. Attempting reset: Failed.
()> ()> 

The backtrace and the failed ASSERT:

(lldb) Process 26963 stopped
* thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGABRT
    frame #0: 0x00007fff2035f92e libsystem_kernel.dylib`__pthread_kill + 10
libsystem_kernel.dylib`__pthread_kill:
->  0x7fff2035f92e <+10>: jae    0x7fff2035f938            ; <+20>
    0x7fff2035f930 <+12>: movq   %rax, %rdi
    0x7fff2035f933 <+15>: jmp    0x7fff20359ad9            ; cerror_nocancel
    0x7fff2035f938 <+20>: retq   
Target 0: (postgres) stopped.
(lldb)  bt
bt
* thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGABRT
  * frame #0: 0x00007fff2035f92e libsystem_kernel.dylib`__pthread_kill + 10
    frame #1: 0x00007fff2038e5bd libsystem_pthread.dylib`pthread_kill + 263
    frame #2: 0x00007fff202e3406 libsystem_c.dylib`abort + 125
    frame #3: 0x00000001025f63bd postgres`ExceptionalCondition + 125 at /Users/mtakahara/code/yugabyte-db4/src/postgres/src/backend/utils/error/assert.c:54
    frame #4: 0x000000010264a509 postgres`MarkPortalFailed + 57 at /Users/mtakahara/code/yugabyte-db4/src/postgres/src/backend/utils/mmgr/portalmem.c:452
    frame #5: 0x00000001024139bf postgres`PortalRun + 1215 at /Users/mtakahara/code/yugabyte-db4/src/postgres/src/backend/tcop/pquery.c:839
    frame #6: 0x0000000102410e7b postgres`exec_simple_query + 1451 at /Users/mtakahara/code/yugabyte-db4/src/postgres/src/backend/tcop/postgres.c:1161
    frame #7: 0x000000010240f985 postgres`yb_exec_simple_query_impl + 21 at /Users/mtakahara/code/yugabyte-db4/src/postgres/src/backend/tcop/postgres.c:4383
    frame #8: 0x000000010240f8c6 postgres`yb_exec_query_wrapper + 182 at /Users/mtakahara/code/yugabyte-db4/src/postgres/src/backend/tcop/postgres.c:4368
    frame #9: 0x000000010240be15 postgres`yb_exec_simple_query + 69 at /Users/mtakahara/code/yugabyte-db4/src/postgres/src/backend/tcop/postgres.c:4398
    frame #10: 0x000000010240ac3b postgres`PostgresMain + 2907 at /Users/mtakahara/code/yugabyte-db4/src/postgres/src/backend/tcop/postgres.c:5006
    frame #11: 0x000000010233b774 postgres`BackendRun + 836 at /Users/mtakahara/code/yugabyte-db4/src/postgres/src/backend/postmaster/postmaster.c:4470
    frame #12: 0x000000010233a962 postgres`BackendStartup + 626 at /Users/mtakahara/code/yugabyte-db4/src/postgres/src/backend/postmaster/postmaster.c:4136
    frame #13: 0x000000010233959c postgres`ServerLoop + 956 at /Users/mtakahara/code/yugabyte-db4/src/postgres/src/backend/postmaster/postmaster.c:1754
    frame #14: 0x000000010233674a postgres`PostmasterMain + 6906 at /Users/mtakahara/code/yugabyte-db4/src/postgres/src/backend/postmaster/postmaster.c:1417
    frame #15: 0x000000010223c3e4 postgres`PostgresServerProcessMain + 772 at /Users/mtakahara/code/yugabyte-db4/src/postgres/src/backend/main/main.c:234
    frame #16: 0x000000010223c812 postgres`main + 34
    frame #17: 0x00007fff203a9f3d libdyld.dylib`start + 1
(lldb) f 4
f 4
frame #4: 0x000000010264a509 postgres`MarkPortalFailed + 57 at /Users/mtakahara/code/yugabyte-db4/src/postgres/src/backend/utils/mmgr/portalmem.c:452
   449 	MarkPortalFailed(Portal portal)
   450 	{
   451 		/* Perform the state transition */
-> 452 		Assert(portal->status != PORTAL_DONE);
    		^
   453 		portal->status = PORTAL_FAILED;
   454 	
   455 		/*
(lldb) p portal->status
p portal->status
(PortalStatus) $4 = PORTAL_DONE
(lldb) 

[mtakahar]

According to @d-uspenskiy, this issue is related to the code introduced as a part of the change for #9467, and I've verified that his suggested change fixes this, too.

@pkj415 can you please take a look?