1. The document summarizes a presentation about parallel query in AWS Aurora. It discusses Aurora architecture, parallel query features and implementation steps, use cases, prerequisites, and provides examples testing performance with and without parallel query enabled. 2. Parallel query allows SQL queries to execute in parallel across multiple Aurora nodes, improving performance for queries with certain characteristics like equal, in, and range filters. 3. Test results show parallel query significantly reducing query execution time from hours to minutes for large analytical queries on a 255GB database.

1. Praveen GR
Associate database consultant,
Mydbops
July 17th, 2021
Mydbops 10th meetup
Parallel query in AWS Aurora
MySQL

2. Interested in Databases
Cloud database
Active Tech Speaker
Active Learner
About Me

3. Consulting
Services
Managed
Services
Focuses on Top Opensource database MySQL,
MongoDB and PostgreSQL
Mydbops Services

4. 500 + Clients In 5 Yrs. of Operations
Our Clients

5. Agenda

6. 1. About aurora
2. Parallel query feature in Aurora and use case.
3. Prerequisites.
4. Implementation.
5. Test case.
6. Limitation.

7. About aurora

8. Aurora architecture
Read Read
Read
Write
Write
Write
Primary instance Aurora reader
Aurora reader
Availability zone a Availability zone b Availability zone c

9. About aurora
1. Volume level sync among the cluster node.
2. Has it's own AZ of the volume in a different zone.
3. The writer supports read and writes. Reader supports only reads.
4. Up to 15 read replica is supported.
5. Automated failure when the writer is down.

10. About aurora
6. Dynamic disk adjustment.
7. Aurora storage is also self-healing.
8. Auto-scaling of the reader.
9. Low-Latency Read Replicas.
10. Custom endpoint with autoscaling.

11. Prominent Aurora MySQL feature

12. Prominent Aurora MySQL feature
1. Hash join in MySQL 5.7.
2. Parallel query
3. Storage Auto-Scaling

13. Parallel query in Aurora

14. Three steps in parallel query
1. SQL execution in parallel.
2. Splits among the cluster.
3. Send required data to the network.

15. Architecture flow
Node 1
Node 4
Node 3
Node 2
Application

16. Use cases

17. Use cases
1. Better with query with equal, in, range.
2. More optimised for analytical queries.
3. Reduced IOPS and CPU utilisation.
4. Uniform load sharing.

18. Prerequisites

19. Prerequisites
1. Aurora version > 2.09 or 1.23
2. Instance class - R series
3. Non-partitioned table.
4. Hash join optimisation enabled.

20. Implementation

21. Implementation
Global variable
1. aurora_parallel_query = ON
2. aurora_disable_hash_join = OFF

22. Test case

23. Monitoring
Status variable
1. Aurora_pq_request_attempted
2. Aurora_pq_request_executed
3. Aurora_pq_request_not_chosen_below_min_rows
4. Aurora_pq_max_concurrent_requests
5. Aurora_pq_request_in_progress

24. Lab Environment
Instance type r3.large
Aurora version 2.09
MySQL version 5.7
Table size 255 GB
Record creation sysbench

25. Test case
Without parallel query
mysql> explain select count(id) from sbtest1 where id > 12345 and k < 6789;
+----+-------------+---------+------------+-------+---------------+---------+---------+------+-----------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+---------+------------+-------+---------------+---------+---------+------+-----------+----------+-------------+
| 1 | SIMPLE | sbtest1 | NULL | range | PRIMARY | PRIMARY | 4 | NULL | 654698147 | 33.33 | Using where |
+----+-------------+---------+------------+-------+---------------+---------+---------+------+-----------+----------+-------------+
1 row in set, 1 warning (0.00 sec)
With parallel query
mysql> explain select count(id) from sbtest1 where id > 12345 and k < 6789;
+----+-------------+---------+------------+------+---------------+------+---------+------+------------+----------+----------------------------------------------------------------
------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra
|
+----+-------------+---------+------------+------+---------------+------+---------+------+------------+----------+----------------------------------------------------------------
------------+
| 1 | SIMPLE | sbtest1 | NULL | ALL | PRIMARY | NULL | NULL | NULL | 1309396294 | 16.66 | Using where; Using parallel query (2 columns, 2 filters, 0
exprs; 0 extra) |
+----+-------------+---------+------------+------+---------------+------+---------+------+------------+----------+----------------------------------------------------------------
------------+
1 row in set, 1 warning (0.00 sec)
Explain plan

26. Test case
Explain Extra option
Columns No.of column in the query.
Filters
No.of column in where clause with
equal, not-equal, range.
Exprs
Column with function or operator , that
can proceed by the parallel query.
Extra
Number of expression that cannot be
proceed by parallel query.

27. Performance Analysis
Without parallel query
mysql> select sql_no_cache sum(k) from sbtest1 where upper(k)=231212 and upper(c) is not null;
+--------+
| sum(k) |
+--------+
| NULL |
+--------+
1 row in set (2 hours 33 min 22.40 sec)
With parallel query
mysql> select sql_no_cache sum(k) from sbtest1 where upper(k)=231212 and upper(c) is not null;
+--------+
| sum(k) |
+--------+
| NULL |
+--------+
1 row in set (1 min 6.61 sec)
With function

28. Performance Analysis
Without parallel query
CPU is in complete saturation for the query execution.

29. Performance Analysis
Without parallel query
Read latency of the server is high.

30. Performance Analysis
With parallel query
Single spike for query processing.

31. Performance Analysis
With parallel query

32. Test case ( using eq Ref )
Without parallel query
mysql> explain select sql_no_cache count(*) from sbtest1 where k=7256238746;
+----+-------------+---------+------------+------+---------------+------+---------+------+------------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+---------+------------+------+---------------+------+---------+------+------------+----------+-------------+
| 1 | SIMPLE | sbtest1 | NULL | ALL | NULL | NULL | NULL | NULL | 1205294616 | 10.00 | Using where |
+----+-------------+---------+------------+------+---------------+------+---------+------+------------+----------+-------------+
1 row in set, 1 warning (0.00 sec)
With parallel query
mysql> explain select count(*) from sbtest1 where k=7256238746;
+----+-------------+---------+------------+------+---------------+------+---------+------+------------+----------+----------------------------------------------------------------
------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra
|
+----+-------------+---------+------------+------+---------------+------+---------+------+------------+----------+----------------------------------------------------------------
------------+
| 1 | SIMPLE | sbtest1 | NULL | ALL | NULL | NULL | NULL | NULL | 1205294616 | 10.00 | Using where; Using parallel query (1 columns, 0 filters, 1
exprs; 0 extra) |
+----+-------------+---------+------------+------+---------------+------+---------+------+------------+----------+----------------------------------------------------------------
------------+
1 row in set, 1 warning (0.01 sec)
Explain plan

33. Performance
Without parallel query
mysql> select sql_no_cache count(*) from sbtest1 where k=7256238746;
+----------+
| count(*) |
+----------+
| 0 |
+----------+
1 row in set (2 hours 25 min 9.11 sec)
With parallel query
mmysql> select count(*) from sbtest1 where k=7256238746;
+----------+
| count(*) |
+----------+
| 0 |
+----------+
1 row in set (2 min 12.22 sec)

34. Performance Analysis
Without parallel query
CPU is in complete saturation for the query execution.

35. Performance
Without parallel query
Read latency of the server is high.

36. Performance
With parallel query
Single spike for query processing.

37. Performance Analysis
With parallel query

38. Test case ( Using Join Cond )
Without parallel query
mysql> explain select count(t1.k) from sbtest1 t1 inner join sbtest3 t2 on t1.id=t2.id where t1.k=247423;
+----+-------------+-------+------------+--------+---------------+---------+---------+--------------+----------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+--------+---------------+---------+---------+--------------+----------+----------+-------------+
| 1 | SIMPLE | t2 | NULL | index | PRIMARY | k_1 | 4 | NULL | 17588790 | 100.00 | Using index |
| 1 | SIMPLE | t1 | NULL | eq_ref | PRIMARY | PRIMARY | 4 | sbtest.t2.id | 1 | 10.00 | Using where |
+----+-------------+-------+------------+--------+---------------+---------+---------+--------------+----------+----------+-------------+
2 rows in set, 1 warning (0.00 sec)
With parallel query
mysql> explain select count(t1.k) from sbtest1 t1 inner join sbtest3 t2 on t1.id=t2.id where t1.k=247423;
+----+-------------+-------+------------+-------+---------------+------+---------+------+------------+----------+-----------------------------------------------------------------
---------------------------------------------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra
|
+----+-------------+-------+------------+-------+---------------+------+---------+------+------------+----------+-----------------------------------------------------------------
---------------------------------------------------------+
| 1 | SIMPLE | t2 | NULL | index | PRIMARY | k_1 | 4 | NULL | 17588790 | 100.00 | Using index
|
| 1 | SIMPLE | t1 | NULL | ALL | PRIMARY | NULL | NULL | NULL | 1205294616 | 0.00 | Using where; Using join buffer (Hash Join Outer table t1); Using
parallel query (2 columns, 1 filters, 1 exprs; 0 extra) |
+----+-------------+-------+------------+-------+---------------+------+---------+------+------------+----------+-----------------------------------------------------------------
---------------------------------------------------------+
2 rows in set, 1 warning (0.00 sec)
Explain plan

39. Without parallel query
With parallel query
mysql> select count(t1.k) from sbtest1 t1 inner join sbtest3 t2 on t1.id=t2.id where t1.k=247423;
+-------------+
| count(t1.k) |
+-------------+
| 0 |
+-------------+
1 row in set (2 min 38.48 sec)
mysql> select sql_no_cache count(t1.k) from sbtest1 t1 inner join sbtest3 t2 on t1.id=t2.id where
t1.k=247423;
+-------------+
| count(t1.k) |
+-------------+
| 0 |
+-------------+
1 row in set (10 min 57.72 sec)
Performance Analysis

40. Performance
Without parallel query
Performance Analysis

41. Performance
Without parallel query
Performance Analysis

42. With parallel query
Performance Analysis

43. Performance
With parallel query
Performance Analysis

44. Performance Improvement
Efficiency
120
100
80
60
40
20
0
PQ Normal​
Method
Title
120
100
80
60
40
20
Normal PQ​
CPU

45. Performance Summary
Without parallel Query With parallel query
Function 2 hrs 33 mins 1 min
Eq Ref 2 hrs 25 mins 2 mins
Join condition 10 mins 2 mins

46. Limitations

47. Limitation
1.Row format should not be compressed. Supports only dynamic
2. Won't work for smaller tables.
3. Limited number of parallel query execution.
4. Aurora version should > 2.09.

48. Reach Us : Info@mydbops.com
Thank You