[YSQL] Failed Test: TestPgRegressJson.testPgRegressJson

[rjalan-yb]

Jira Link: DB-5047

Description

The test case: TestPgRegressJson.testPgRegressJson has failed on alma8-clang15-asan since beginning. Never passed and failed with same json and jsonb error:

pg_regress|pid78021|stdout test yb_pg_create_type            ... ok
pg_regress|pid78021|stdout test yb_pg_create_table           ... ok
pg_regress|pid78021|stdout test yb_pg_copy                   ... ok
pg_regress|pid78021|stdout test yb_pg_json                   ... FAILED
pg_regress|pid78021|stdout test yb_pg_jsonb                  ... FAILED
pg_regress|pid78021|stdout test yb_pg_jsonpath               ... ok
pg_regress|pid78021|stdout test yb_pg_jsonpath_encoding      ... ok
pg_regress|pid78021|stdout test yb_pg_jsonb_jsonpath         ... ok
pg_regress|pid78021|stderr -----------------------------------------------------
pg_regress|pid78021|stderr Suppressions used:
pg_regress|pid78021|stderr   count      bytes template
pg_regress|pid78021|stderr       3        505 pg_strdup
pg_regress|pid78021|stderr -----------------------------------------------------
pg_regress|pid78021|stderr 
pg_regress|pid78021|stdout 
pg_regress|pid78021|stdout ======================
pg_regress|pid78021|stdout  2 of 9 tests failed. 
pg_regress|pid78021|stdout ======================

For Json: yb_pg_json.sql#89

SELECT row_to_json(q)
FROM (SELECT $$a$$ || x AS b,
         y AS c,
--- 211,278 ----
ERROR:  stack depth limit exceeded
HINT:  Increase the configuration parameter "max_stack_depth" (currently 100kB), after ensuring the platform's stack depth limit is adequate.
RESET max_stack_depth;

For JSONB all tsheadline case get errored out with this: yb_pg_jsonb.sql#1248

4838  -- corner cases for ts_headline with jsonb                                                                                                                                                                                                                                                                                                             2051  -- corner cases for ts_headline with jsonb
4839  select ts_headline('null'::jsonb, tsquery('aaa & bbb'));                                                                                                                                                                                                                                                                                               2052  select ts_headline('null'::jsonb, tsquery('aaa & bbb'));
4840   ts_headline                                                                                                                                                                                                                                                                                                                                        |  2053  ERROR:  current transaction is aborted, commands ignored until end of transaction block
4841  -------------    

Failing 98/100 times on alma8-clang15-asan: https://detective-gcp.dev.yugabyte.com/stability/test?analyze_trends=false&branch=master&build_type=all&class=org.yb.pgsql.TestPgRegressJson&fail_tag=all&name=testPgRegressJson&num_commits=500&num_fails=100&platform=all

Passing on 2.17.0 on alma8-clang13-asan:
https://detective-gcp.dev.yugabyte.com/stability/test?analyze_trends=true&branch=2.17.0&build_type=all&class=org.yb.pgsql.TestPgRegressJson&fail_tag=all&name=testPgRegressJson&num_commits=500&platform=all

yb_pg_jsonb.sql was modified in Commit: 509b95010c

[jasonyb]

Commit 8c26092 fixes similar issue, but a deeper fix should be done, such as modifying ASAN's limit.

Internal link: https://yugabyte.slack.com/archives/C4NAYRFS8/p1677706886196309?thread_ts=1677085105.706619&cid=C4NAYRFS8

[sushantrmishra]

SELECT array_to_json(array_agg(x),false) from generate_series(5,10) x;
! array_to_json
! ----------------
! [5,6,7,8,9,10]
! (1 row)
!
SELECT array_to_json('{{1,5},{99,100}}'::int[]);
! array_to_json
! ------------------
! [[1,5],[99,100]]
! (1 row)
!
-- row_to_json
SELECT row_to_json(row(1,'foo'));
! row_to_json
! ---------------------
! {"f1":1,"f2":"foo"}
! (1 row)
!
SELECT row_to_json(q)
FROM (SELECT $$a$$ || x AS b,
y AS c,
--- 211,278 ----
ERROR: stack depth limit exceeded
HINT: Increase the configuration parameter "max_stack_depth" (currently 100kB), after ensuring the platform's stack depth limit is adequate.
RESET max_stack_depth;

• ERROR: stack depth limit exceeded
• HINT: Increase the configuration parameter "max_stack_depth" (currently 100kB), after ensuring the platform's stack depth limit is adequate.
  -- Miscellaneous stuff.
  SELECT 'true'::json; -- OK
  ! ERROR: stack depth limit exceeded
  ! HINT: Increase the configuration parameter "max_stack_depth" (currently 100kB), after ensuring the platform's stack depth limit is adequate.
  SELECT 'false'::json; -- OK
  ! ERROR: stack depth limit exceeded
  ! HINT: Increase the configuration parameter "max_stack_depth" (currently 100kB), after ensuring the platform's stack depth limit is adequate.
  SELECT 'null'::json; -- OK
  ! ERROR: stack depth limit exceeded
  ! HINT: Increase the configuration parameter "max_stack_depth" (currently 100kB), after ensuring the platform's stack depth limit is adequate.
  SELECT ' true '::json; -- OK, even with extra whitespace
  ! ERROR: stack depth limit exceeded
  ! HINT: Increase the configuration parameter "max_stack_depth" (currently 100kB), after ensuring the platform's stack depth limit is adequate.
  SELECT 'true false'::json; -- ERROR, too many values
  ! ERROR: stack depth limit exceeded
  ! HINT: Increase the configuration parameter "max_stack_depth" (currently 100kB), after ensuring the platform's stack depth limit is adequate.
  SELECT 'true, false'::json; -- ERROR, too many values
  ! ERROR: stack depth limit exceeded
  ! HINT: Increase the configuration parameter "max_stack_depth" (currently 100kB), after ensuring the platform's stack depth limit is adequate.
  SELECT 'truf'::json; -- ERROR, not a keyword
  ! ERROR: stack depth limit exceeded
  ! HINT: Increase the configuration parameter "max_stack_depth" (currently 100kB), after ensuring the platform's stack depth limit is adequate.
  SELECT 'trues'::json; -- ERROR, not a keyword
  ! ERROR: stack depth limit exceeded
  ! HINT: Increase the configuration parameter "max_stack_depth" (currently 100kB), after ensuring the platform's stack depth limit is adequate.
  SELECT ''::json; -- ERROR, no value
  ! ERROR: stack depth limit exceeded
  ! HINT: Increase the configuration parameter "max_stack_depth" (currently 100kB), after ensuring the platform's stack depth limit is adequate.
  SELECT ' '::json; -- ERROR, no value
  ! ERROR: stack depth limit exceeded
  ! HINT: Increase the configuration parameter "max_stack_depth" (currently 100kB), after ensuring the platform's stack depth limit is adequate.
  --constructors
  -- array_to_json
  SELECT array_to_json(array(select 1 as a));
  ! ERROR: stack depth limit exceeded
  ! HINT: Increase the configuration parameter "max_stack_depth" (currently 100kB), after ensuring the platform's stack depth limit is adequate.
  SELECT array_to_json(array_agg(q),false) from (select x as b, x * 2 as c from generate_series(1,3) x) q;
  ! ERROR: stack depth limit exceeded
  ! HINT: Increase the configuration parameter "max_stack_depth" (currently 100kB), after ensuring the platform's stack depth limit is adequate.
  SELECT array_to_json(array_agg(q),true) from (select x as b, x * 2 as c from generate_series(1,3) x) q;
  ! ERR
  ...[truncated 3298779 chars]...

[OlegLoginov]

RCA:

SET max_stack_depth = '100kB';

SELECT repeat('[', 10000)::json;                                             
ERROR:  stack depth limit exceeded                                           
HINT:  Increase the configuration parameter "max_stack_depth" (currently 100k

show max_stack_depth;                                                        
ERROR:  stack depth limit exceeded                                           
HINT:  Increase the configuration parameter "max_stack_depth" (currently 100k

Here we have expected error stack depth limit exceeded after recursive statement repeat.
In usual (debug/release) mode the next statement starts form a clean stack.
But in ASAN the stack is NOT cleaned. So, ANY next statement has the same stack state as after the previous repeat command. The debug log - see stack_depth value:

show max_stack_depth;
 max_stack_depth
-----------------
 512kB
(1 row)

SELECT repeat('[', 10000)::json;
ERROR:  stack depth limit exceeded
HINT:  stack_depth = 524352 max_stack_depth_bytes = 524288

show max_stack_depth;
ERROR:  stack depth limit exceeded
HINT:  stack_depth = 544000 max_stack_depth_bytes = 524288 

CONCLUSION: In ASAN any statement after recursive command (like repeat above) should generate stack depth limit exceeded error after ANY statement (even after the simplest show statement).

[OlegLoginov]

Possible FIX: let's move all recurrent statements to the end of the tests.
It does not affect testing in debug & release, but hides the "only growing stack" issue in ASAN.

[OlegLoginov]

It potentially affects the following PG tests:
ybd asan --java-test org.yb.pgsql.TestPgRegressJson#testPgRegressJson

src/postgres/src/test/regress/sql/yb_pg_json.sql   
src/postgres/src/test/regress/sql/yb_pg_jsonb.sql  

ybd asan --java-test org.yb.pgsql.TestPgRegressPgMisc#testPgRegressPgMisc

src/postgres/src/test/regress/sql/yb_pg_errors.sql 

Not used:

src/postgres/src/test/regress/sql/json.sql         
src/postgres/src/test/regress/sql/jsonb.sql        

[OlegLoginov]

For the test
ybd asan --java-test org.yb.pgsql.TestPgRegressPgMisc#testPgRegressPgMisc
the error is:

-- Check that stack depth detection mechanism works and                         -- Check that stack depth detection mechanism works and
-- max_stack_depth is not set too high                                          -- max_stack_depth is not set too high
create function infinite_recurse() returns int as                               create function infinite_recurse() returns int as
'select infinite_recurse()' language sql;                                       'select infinite_recurse()' language sql;
\set VERBOSITY terse                                                            \set VERBOSITY terse
select infinite_recurse();                                                      select infinite_recurse();
ERROR:  stack depth limit exceeded                                            | server closed the connection unexpectedly
                                                                              >         This probably means the server terminated abnormally
                                                                              >         before or while processing the request.
                                                                              > connection to server was lost

It should be investigated & fixed in another GHI: #16188

[OlegLoginov]

Alternative fix (D35672) was abandoned.