Bug#27970159: DUPLICATE ENTRY FOR KEY '<GROUP_KEY>' ERROR

Description: Query involving GROUP BY on timestamp column
results in duplicate entry for key (ER_DUP_ENTRY) error.

Analysis: The problem arises when timestamp values are
inserted to a table with one timezone and these values are
fetched later with a new timezone, such that the inserted
timestamp values falls in the DST switch hour in the new
timezone. Same timestamp value can exist twice during the
DST switching hour. Eg: 2019-10-27 03:30 in EET timezone
where DST ends on 27th October 2019 at 04:00. Time 03:30
appears 1st during DST hour and the same time appears again
after the clock is switched back to 03:00 on 2019-10-27 at
4:00.

Lets consider the following example:-
CREATE TABLE t1 (c1 TIMESTAMP);
SET TIME_ZONE = '+00:00';
INSERT INTO t1 VALUES ('2019-10-27 00:30:00');
INSERT INTO t1 VALUES ('2019-10-27 00:30:00');
INSERT INTO t1 VALUES ('2019-10-27 01:30:00');
INSERT INTO t1 VALUES ('2019-10-27 01:30:00');
SET TIME_ZONE = 'SYSTEM';

Table t1 is inserted with 4 timestamp values and later the
timezone is changed to EET (system timezone). Once timezone
is changed to EET, both the values 00:30:00 and 01:30:00 map
to the same hour, 03:30:00, in EET timezone due to the DST
switching in EET on 2019-10-27 at 4:00.

mysql> SELECT * FROM t1;
+---------------------+
| c1                  |
+---------------------+
| 2019-10-27 03:30:00 |
| 2019-10-27 03:30:00 |
| 2019-10-27 03:30:00 |
| 2019-10-27 03:30:00 |
+---------------------+
4 rows in set (0.00 sec)

Now lets consider a query with GROUP BY on the TIMESTAMP
column, c1 :-
SELECT COUNT(*) FROM t1 GROUP BY c1;

As part of the above query execution, these timestamp values
in binary representation are fetched from the table where it
is stored as 00:30:00 and 01:30:00 in the binary format, B1
and B2. However all 4 values map to the same timestamp value,
03:30, in the current timezone (EET). When converted to its
respective binary format, all 4 values get the same binary
representation B2.

As part of grouping the data for GROUP BY execution,
optimizer creates hash map with a key on the GROUP BY field
values, i.e, the binary representation of the timestamp
values fetched from the table.

As part of grouping, 1st row ('2019-10-27 00:30:00') is
encountered with key as B1. Since its the 1st occurrence of
B1, the record is inserted into the hash. But this record
inserted is the binary representation of the timestamp
value in EET timezone, i.e, B2. Then the 2nd row
('2019-10-27 00:30:00') with same key B1 is encountered.
The hash search with key as B1 cannot fetch the record in
the hash since the 1st record was inserted as B2. Due to
this, the record is again inserted into the hash and results
in a duplicate key error.

Fix:- When server wants to return DUPLICATE ENTRY FOR KEY
'<GROUP_KEY>' ERROR, if the grouping involves a TIMESTAMP
field, throw an error: "Grouping on temporal is
non-deterministic for time zones having DST. Please consider
switching to UTC for this query."

Approved by: Guilhem Bichot <guilhem.bichot@oracle.com>
Approved by: Jens Even Blomsoy <jens.even.blomsoy@oracle.com>

Change-Id: Ic15f04ea9e9b3c04bcc1ec97e47172726a082f9b

Author: Arun Kuruvila

mysql-test/r/timezone.result

@@ -45,3 +45,20 @@ unix_timestamp('2038-01-19 04:14:07'),
 unix_timestamp('2038-01-19 04:14:08');
 unix_timestamp('1970-01-01 01:00:00')	unix_timestamp('1970-01-01 01:00:01')	unix_timestamp('2038-01-19 04:14:07')	unix_timestamp('2038-01-19 04:14:08')
 0	1	2147483647	0
+CREATE TABLE t1 (c1 TIMESTAMP);
+SET TIME_ZONE = '+00:00';
+INSERT INTO t1 VALUES('2019-10-27 00:47:42'), ('2019-10-27 00:47:42');
+INSERT INTO t1 VALUES('2019-10-27 01:47:42'), ('2019-10-27 01:47:42');
+SET TIME_ZONE = 'SYSTEM';
+SELECT DISTINCT c1 FROM t1;
+c1
+2019-10-27 02:47:42
+SELECT COUNT(*) FROM t1 GROUP BY c1;
+ERROR HY000: Grouping on temporal is non-deterministic for timezones having DST. Please consider switching to UTC for this query.
+SET sql_mode='';
+Warnings:
+Warning	3090	Changing sql mode 'NO_AUTO_CREATE_USER' is deprecated. It will be removed in a future release.
+SELECT COUNT(*) FROM t1 GROUP BY c1;
+ERROR HY000: Grouping on temporal is non-deterministic for timezones having DST. Please consider switching to UTC for this query.
+SET sql_mode=DEFAULT;
+DROP TABLE t1;

mysql-test/t/timezone.test

@@ -62,3 +62,28 @@ select unix_timestamp('1970-01-01 01:00:00'),
        unix_timestamp('2038-01-19 04:14:08');
 
 # End of 4.1 tests
+
+#
+# Bug #27970159: DUPLICATE ENTRY FOR KEY '<GROUP_KEY>' ERROR
+#
+
+CREATE TABLE t1 (c1 TIMESTAMP);
+SET TIME_ZONE = '+00:00';
+
+INSERT INTO t1 VALUES('2019-10-27 00:47:42'), ('2019-10-27 00:47:42');
+INSERT INTO t1 VALUES('2019-10-27 01:47:42'), ('2019-10-27 01:47:42');
+
+SET TIME_ZONE = 'SYSTEM';
+
+SELECT DISTINCT c1 FROM t1;
+
+--error ER_GROUPING_ON_TIMESTAMP_IN_DST
+SELECT COUNT(*) FROM t1 GROUP BY c1;
+
+SET sql_mode='';
+--error ER_GROUPING_ON_TIMESTAMP_IN_DST
+SELECT COUNT(*) FROM t1 GROUP BY c1;
+SET sql_mode=DEFAULT;
+
+#Cleanup
+DROP TABLE t1;

sql/share/errmsg-utf8.txt

@@ -7774,6 +7774,9 @@ ER_XA_REPLICATION_FILTERS
 ER_CANT_OPEN_ERROR_LOG
   eng "Could not open file '%s' for error logging%s%s"
 
+ER_GROUPING_ON_TIMESTAMP_IN_DST
+  eng "Grouping on temporal is non-deterministic for timezones having DST. Please consider switching to UTC for this query."
+
 #
 #  End of 5.7 error messages.
 #

sql/sql_executor.cc

@@ -1,4 +1,4 @@
-/* Copyright (c) 2000, 2018, Oracle and/or its affiliates. All rights reserved.
+/* Copyright (c) 2000, 2019, Oracle and/or its affiliates. All rights reserved.
 
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -3506,6 +3506,26 @@ end_update(JOIN *join, QEP_TAB *const qep_tab, bool end_of_records)
   init_tmptable_sum_functions(join->sum_funcs);
   if ((error=table->file->ha_write_row(table->record[0])))
   {
+    /*
+      If the error is HA_ERR_FOUND_DUPP_KEY and the grouping involves a
+      TIMESTAMP field, throw a meaningfull error to user with the actual
+      reason and the workaround. I.e, "Grouping on temporal is
+      non-deterministic for timezones having DST. Please consider switching
+      to UTC for this query". This is a temporary measure until we implement
+      WL#13148 (Do all internal handling TIMESTAMP in UTC timezone), which
+      will make such problem impossible.
+    */
+    if (error == HA_ERR_FOUND_DUPP_KEY)
+    {
+      for (group=table->group ; group ; group=group->next)
+      {
+        if (group->field->type() == MYSQL_TYPE_TIMESTAMP)
+        {
+          my_error(ER_GROUPING_ON_TIMESTAMP_IN_DST, MYF(0));
+          DBUG_RETURN(NESTED_LOOP_ERROR);
+        }
+      }
+    }
     if (create_ondisk_from_heap(join->thd, table,
                                 tmp_tbl->start_recinfo,
                                 &tmp_tbl->recinfo,