SlideShare a Scribd company logo
1 of 28
Download to read offline
Lessons for the optimizer
from TPC-DS benchmark
Sergei Petrunia
Query Optimizer developer
MariaDB Corporation
2019 MariaDB Developers Unconference
New York
The goals
1. Want to evaluate/measure the query optimizer
2. Hard to do, optimizer should handle
– Different query patterns
– Different data distributions, etc
3. How does one do it anyway?
– Look at benchmarks
– Or “optimizer part” of the benchmarks
Benchmarks
1. sysbench
– Popular
– Does only basic queries, few query patterns
2. DBT-3 (aka TPC-H)
– 6 tables, 22 analytic queries
– Was used to see some optimizer problems
– Limited:
●
Uniform data distribution, uncorrelated columns
●
...
TPC-DS benchmark
● Obsoletes DBT-3 benchmark
● Richer dataset
– 25 Tables, 99 queries
– Non-uniform data distributions
● Uses advanced SQL features
– 32 queries use CTE
– 27 queries use Window Functions
– etc
● Could not really run it until MariaDB 10.2 (or MySQL 8)
MariaDB still can’t run all of TPC-DS
●
2 Queries: FULL OUTER JOIN
●
10 Queries: ROLLUP + ORDER BY problem (MDEV-17807)
●
~20 more queries have fixable problems
– “Every derived table must have an alias”, etc
select
...
group by
a,b,c with rollup
order by
a,b,c
ERROR 1221 (HY000): Incorrect usage of CUBE/ROLLUP
and ORDER BY
Oracle MySQL and TPC-DS
● ROLLUP + ORDER BY is supported since 8.0.12
● Doesn’t support FULL OUTER JOIN (2 queries)
● Doesn’t support EXCEPT (1 query)
● Doesn’t support INTERSECT (3 queries)
Running queries from TPC-DS
● Data generator creates CSV files
– Adjust #define for MySQL/MariaDB
● Query generator produces “streams” from templates
– A set of QueryNNN.tpl files
– A stream is a text file with one instance of each of the 99 queries
– One can add hooks at query start/end
● Queries have a few typos
● There’s no tool to run queries/measure time
– Note that the read queries are a subset of benchmark (TpCX$)
Getting it to run
● A collection of scripts at
https://github.com/spetrunia/tpcds-run-tool
● The goal is a fully-automated run
– MariaDB, MySQL, PostgreSQL
● Because we need to play with settings/options
Test runs done
● The dataset
– Scale=1
– 1.2 GB CSV files
– 6 GB when loaded
● The Queries
– 10..20 “Streams”
● Tuning
– Innodb_buffer_pool=8G (50% RAM)
– shared_buffers = 4G (25% RAM)
Test results
Test results
● ...
Test results
● … a bit inconclusive – query times varied across my runs (?)
● Time to run one stream = 20 min – 2 hours
● Searching for the source of randomness
– Started to work on full automation
●
(did I run ANALYZE? Did I have correct with my.cnf
parameters?)
– Started to look at rngseed in dataset/query generator
MariaDB/MySQL
MariaDB 10.2, 10.4, MySQL 8
● Scale=1, 6.1 GB data, 8G buffer pool
● rngseed=1234 for both
● Benchmark takes ~20 min
● Query times are very non-uniform
+-------------+---------------+
| query_name | QueryTime_ms |
+-------------+---------------+
| query72.tpl | 678,321 |
| query23.tpl | 80,025 |
| query2.tpl | 65,156 |
| query39.tpl | 63,761 |
| query78.tpl | 63,473 |
| query4.tpl | 27,549 |
| query31.tpl | 24,344 |
| query47.tpl | 19,156 |
| query11.tpl | 17,484 |
| query74.tpl | 16,571 |
| query21.tpl | 16,212 |
| query59.tpl | 10,522 |
| query88.tpl | 9,965 |
Query#72 dominates
query1.tpl
query13.tpl
query19.tpl
query23.tpl
query28.tpl
query31.tpl
query35.tpl
query40.tpl
query44.tpl
query48.tpl
query53.tpl
query58.tpl
query61.tpl
query65.tpl
query69.tpl
query73.tpl
query78.tpl
query83.tpl
query89.tpl
query92.tpl
query96.tpl
0
100000
200000
300000
400000
500000
600000
700000
800000
Without Query #72
query1.tpl
query13.tpl
query19.tpl
query23.tpl
query28.tpl
query31.tpl
query35.tpl
query40.tpl
query44.tpl
query48.tpl
query53.tpl
query58.tpl
query61.tpl
query65.tpl
query69.tpl
query74.tpl
query79.tpl
query84.tpl
query9.tpl
query93.tpl
query98.tpl
0
10000
20000
30000
40000
50000
60000
70000
80000
90000
PostgreSQL 11
PostgreSQL 11
● There was a “fast” run
● Showing results from the last
two runs (both where “slow”)
– rngseed=5678 for both
– 121 min
– rngseed=1234 (data),
rngseed=4321 (query)
– 145..154 min.
Heaviest queries in the run
● Execution time varies
● Is this a query optimizer issue?
● Or different constants in a skewed dataset?
+-------------+-----------------+-----------------+--------+
| query_name | PG11-seed5678 | PG11-seed1234 | X |
+-------------+-----------------+-----------------+--------+
| query4.tpl | 3,628,830 | 3,578,944 | 1.0139 |
| query11.tpl | 2,004,392 | 2,013,597 | 0.9954 |
| query1.tpl | 87,981 | 1,947,624 | 0.0452 |
| query74.tpl | 693,784 | 641,696 | 1.0812 |
| query47.tpl | 624,717 | 539,941 | 1.1570 |
| query57.tpl | 116,570 | 112,472 | 1.0364 |
| query81.tpl | 22,089 | 47,366 | 0.4663 |
| query6.tpl | 27,896 | 27,009 | 1.0328 |
| query30.tpl | 11,214 | 11,171 | 1.0038 |
| query39.tpl | 10,803 | 10,702 | 1.0094 |
| query95.tpl | 16,418 | 10,065 | 1.6312 |
`
● Do we need a “representative
collection of datasets”?
– Check N datasets?
Compare most heavy queries
● Some queries are present in both lists, but some are only in one.
● Not clear
+-------------+-----------------+-----------------+--------+
| query_name | PG11-seed5678 | PG11-seed1234 | X |
+-------------+-----------------+-----------------+--------+
| query4.tpl | 3,628,830 | 3,578,944 | 1.0139 |
| query11.tpl | 2,004,392 | 2,013,597 | 0.9954 |
| query1.tpl | 87,981 | 1,947,624 | 0.0452 |
| query74.tpl | 693,784 | 641,696 | 1.0812 |
| query47.tpl | 624,717 | 539,941 | 1.1570 |
| query57.tpl | 116,570 | 112,472 | 1.0364 |
| query81.tpl | 22,089 | 47,366 | 0.4663 |
| query6.tpl | 27,896 | 27,009 | 1.0328 |
| query30.tpl | 11,214 | 11,171 | 1.0038 |
| query39.tpl | 10,803 | 10,702 | 1.0094 |
| query95.tpl | 16,418 | 10,065 | 1.6312 |
`
+-------------+---------------+
| query_name | QueryTime_ms |
+-------------+---------------+
| query72.tpl | 678,321 |
| query23.tpl | 80,025 |
| query2.tpl | 65,156 |
| query39.tpl | 63,761 |
| query78.tpl | 63,473 |
| query4.tpl | 27,549 |
| query31.tpl | 24,344 |
| query47.tpl | 19,156 |
| query11.tpl | 17,484 |
| query74.tpl | 16,571 |
| query21.tpl | 16,212 |
| query59.tpl | 10,522 |
MariaDB PostgreSQL
Observations about the benchmark
● rngseed on the dataset matters A LOT
– What is a representative set of rngseed values?
● rngseed on query streams – much less
● Hardware?
● Queries are not equal
– Heavy vs lightweight queries
– Is SUM(query_time) an adequate metric?
●
Wont see that a fast query got 10x slower
Other observations
● Both DBT-3 and TPC-DS workloads are relevant for the optimizer
– Condition selectivities
– Semi-join optimizations
– …
● But don’t match the optimizer issues we see
– ORDER BY … LIMIT optimization
– Long IN-list
– …
Extra: parallel query in PG?
Extra – PostgreSQL 11, parallel query?
● Trying on a run with both rngseed=5678:
● Parallel settings
max_parallel_workers_per_gather=8 (the default was 2)
dynamic_shared_memory_type=posix
show max_worker_processes= 8
● Results
– Only saw one core to be occupied
– The run still took 121 min, didin’t see any speedup
Try a parallel query
select
sum(inv_quantity_on_hand*i_current_price)
from
inventory, item
where
i_item_sk=inv_item_sk;
QUERY PLAN
---------------------------------------------------------------------------------
Aggregate (cost=301495.25..301495.26 rows=1 width=32)
-> Hash Join (cost=1635.00..213408.54 rows=11744894 width=10)
Hash Cond: (inventory.inv_item_sk = item.i_item_sk)
-> Seq Scan on inventory (cost=0.00..180935.94 rows=11744894 width=8)
-> Hash (cost=1410.00..1410.00 rows=18000 width=10)
-> Seq Scan on item (cost=0.00..1410.00 rows=18000 width=10)
● max_parallel_workers_per_gather=0
Try a parallel query
select
sum(inv_quantity_on_hand*i_current_price)
from
inventory, item
where
i_item_sk=inv_item_sk;
QUERY PLAN
----------------------------------------------------------------------------------------------------
Finalize Aggregate (cost=125048.98..125048.99 rows=1 width=32)
-> Gather (cost=125048.55..125048.96 rows=4 width=32)
Workers Planned: 4
-> Partial Aggregate (cost=124048.55..124048.56 rows=1 width=32)
-> Parallel Hash Join (cost=1468.23..102026.87 rows=2936224 width=10)
Hash Cond: (inventory.inv_item_sk = item.i_item_sk)
-> Parallel Seq Scan on inventory (cost=0.00..92849.24 rows=2936224 width=8)
-> Parallel Hash (cost=1335.88..1335.88 rows=10588 width=10)
-> Parallel Seq Scan on item (cost=0.00..1335.88 rows=10588 width=10)
● max_parallel_workers_per_gather=8
Try a parallel query
select
sum(inv_quantity_on_hand*i_current_price)
from
inventory, item
where
i_item_sk=inv_item_sk;
● Results
– max_parallel_workers_per_gather=8: 1.0 sec
– max_parallel_workers_per_gather=0: 3.8 sec
● Didn’t see anything like that in TPC-DS benchmark
Thanks!

More Related Content

What's hot

MySQL Optimizer Cost Model
MySQL Optimizer Cost ModelMySQL Optimizer Cost Model
MySQL Optimizer Cost ModelOlav Sandstå
 
Iceberg + Alluxio for Fast Data Analytics
Iceberg + Alluxio for Fast Data AnalyticsIceberg + Alluxio for Fast Data Analytics
Iceberg + Alluxio for Fast Data AnalyticsAlluxio, Inc.
 
Presto best practices for Cluster admins, data engineers and analysts
Presto best practices for Cluster admins, data engineers and analystsPresto best practices for Cluster admins, data engineers and analysts
Presto best practices for Cluster admins, data engineers and analystsShubham Tagra
 
Percona XtraDB Cluster ( Ensure high Availability )
Percona XtraDB Cluster ( Ensure high Availability )Percona XtraDB Cluster ( Ensure high Availability )
Percona XtraDB Cluster ( Ensure high Availability )Mydbops
 
DB12c: All You Need to Know About the Resource Manager
DB12c: All You Need to Know About the Resource ManagerDB12c: All You Need to Know About the Resource Manager
DB12c: All You Need to Know About the Resource ManagerAndrejs Vorobjovs
 
Hive + Tez: A Performance Deep Dive
Hive + Tez: A Performance Deep DiveHive + Tez: A Performance Deep Dive
Hive + Tez: A Performance Deep DiveDataWorks Summit
 
My first 90 days with ClickHouse.pdf
My first 90 days with ClickHouse.pdfMy first 90 days with ClickHouse.pdf
My first 90 days with ClickHouse.pdfAlkin Tezuysal
 
MySQL Parallel Replication: All the 5.7 and 8.0 Details (LOGICAL_CLOCK)
MySQL Parallel Replication: All the 5.7 and 8.0 Details (LOGICAL_CLOCK)MySQL Parallel Replication: All the 5.7 and 8.0 Details (LOGICAL_CLOCK)
MySQL Parallel Replication: All the 5.7 and 8.0 Details (LOGICAL_CLOCK)Jean-François Gagné
 
Apache Spark Core—Deep Dive—Proper Optimization
Apache Spark Core—Deep Dive—Proper OptimizationApache Spark Core—Deep Dive—Proper Optimization
Apache Spark Core—Deep Dive—Proper OptimizationDatabricks
 
Deep Dive into Spark SQL with Advanced Performance Tuning with Xiao Li & Wenc...
Deep Dive into Spark SQL with Advanced Performance Tuning with Xiao Li & Wenc...Deep Dive into Spark SQL with Advanced Performance Tuning with Xiao Li & Wenc...
Deep Dive into Spark SQL with Advanced Performance Tuning with Xiao Li & Wenc...Databricks
 
Optimizing MariaDB for maximum performance
Optimizing MariaDB for maximum performanceOptimizing MariaDB for maximum performance
Optimizing MariaDB for maximum performanceMariaDB plc
 
When Apache Spark Meets TiDB with Xiaoyu Ma
When Apache Spark Meets TiDB with Xiaoyu MaWhen Apache Spark Meets TiDB with Xiaoyu Ma
When Apache Spark Meets TiDB with Xiaoyu MaDatabricks
 
MySQL Performance Tuning. Part 1: MySQL Configuration (includes MySQL 5.7)
MySQL Performance Tuning. Part 1: MySQL Configuration (includes MySQL 5.7)MySQL Performance Tuning. Part 1: MySQL Configuration (includes MySQL 5.7)
MySQL Performance Tuning. Part 1: MySQL Configuration (includes MySQL 5.7)Aurimas Mikalauskas
 
PostgreSQL WAL for DBAs
PostgreSQL WAL for DBAs PostgreSQL WAL for DBAs
PostgreSQL WAL for DBAs PGConf APAC
 
Distributed Databases Deconstructed: CockroachDB, TiDB and YugaByte DB
Distributed Databases Deconstructed: CockroachDB, TiDB and YugaByte DBDistributed Databases Deconstructed: CockroachDB, TiDB and YugaByte DB
Distributed Databases Deconstructed: CockroachDB, TiDB and YugaByte DBYugabyteDB
 
How to Take Advantage of Optimizer Improvements in MySQL 8.0
How to Take Advantage of Optimizer Improvements in MySQL 8.0How to Take Advantage of Optimizer Improvements in MySQL 8.0
How to Take Advantage of Optimizer Improvements in MySQL 8.0Norvald Ryeng
 
Spark and Object Stores —What You Need to Know: Spark Summit East talk by Ste...
Spark and Object Stores —What You Need to Know: Spark Summit East talk by Ste...Spark and Object Stores —What You Need to Know: Spark Summit East talk by Ste...
Spark and Object Stores —What You Need to Know: Spark Summit East talk by Ste...Spark Summit
 
Tez Shuffle Handler: Shuffling at Scale with Apache Hadoop
Tez Shuffle Handler: Shuffling at Scale with Apache HadoopTez Shuffle Handler: Shuffling at Scale with Apache Hadoop
Tez Shuffle Handler: Shuffling at Scale with Apache HadoopDataWorks Summit
 

What's hot (20)

MySQL Optimizer Cost Model
MySQL Optimizer Cost ModelMySQL Optimizer Cost Model
MySQL Optimizer Cost Model
 
Iceberg + Alluxio for Fast Data Analytics
Iceberg + Alluxio for Fast Data AnalyticsIceberg + Alluxio for Fast Data Analytics
Iceberg + Alluxio for Fast Data Analytics
 
Presto best practices for Cluster admins, data engineers and analysts
Presto best practices for Cluster admins, data engineers and analystsPresto best practices for Cluster admins, data engineers and analysts
Presto best practices for Cluster admins, data engineers and analysts
 
Percona XtraDB Cluster ( Ensure high Availability )
Percona XtraDB Cluster ( Ensure high Availability )Percona XtraDB Cluster ( Ensure high Availability )
Percona XtraDB Cluster ( Ensure high Availability )
 
DB12c: All You Need to Know About the Resource Manager
DB12c: All You Need to Know About the Resource ManagerDB12c: All You Need to Know About the Resource Manager
DB12c: All You Need to Know About the Resource Manager
 
Hive + Tez: A Performance Deep Dive
Hive + Tez: A Performance Deep DiveHive + Tez: A Performance Deep Dive
Hive + Tez: A Performance Deep Dive
 
My first 90 days with ClickHouse.pdf
My first 90 days with ClickHouse.pdfMy first 90 days with ClickHouse.pdf
My first 90 days with ClickHouse.pdf
 
MySQL Parallel Replication: All the 5.7 and 8.0 Details (LOGICAL_CLOCK)
MySQL Parallel Replication: All the 5.7 and 8.0 Details (LOGICAL_CLOCK)MySQL Parallel Replication: All the 5.7 and 8.0 Details (LOGICAL_CLOCK)
MySQL Parallel Replication: All the 5.7 and 8.0 Details (LOGICAL_CLOCK)
 
Apache Spark Core—Deep Dive—Proper Optimization
Apache Spark Core—Deep Dive—Proper OptimizationApache Spark Core—Deep Dive—Proper Optimization
Apache Spark Core—Deep Dive—Proper Optimization
 
Deep Dive into Spark SQL with Advanced Performance Tuning with Xiao Li & Wenc...
Deep Dive into Spark SQL with Advanced Performance Tuning with Xiao Li & Wenc...Deep Dive into Spark SQL with Advanced Performance Tuning with Xiao Li & Wenc...
Deep Dive into Spark SQL with Advanced Performance Tuning with Xiao Li & Wenc...
 
Presto
PrestoPresto
Presto
 
Optimizing MariaDB for maximum performance
Optimizing MariaDB for maximum performanceOptimizing MariaDB for maximum performance
Optimizing MariaDB for maximum performance
 
When Apache Spark Meets TiDB with Xiaoyu Ma
When Apache Spark Meets TiDB with Xiaoyu MaWhen Apache Spark Meets TiDB with Xiaoyu Ma
When Apache Spark Meets TiDB with Xiaoyu Ma
 
MySQL Performance Tuning. Part 1: MySQL Configuration (includes MySQL 5.7)
MySQL Performance Tuning. Part 1: MySQL Configuration (includes MySQL 5.7)MySQL Performance Tuning. Part 1: MySQL Configuration (includes MySQL 5.7)
MySQL Performance Tuning. Part 1: MySQL Configuration (includes MySQL 5.7)
 
PostgreSQL WAL for DBAs
PostgreSQL WAL for DBAs PostgreSQL WAL for DBAs
PostgreSQL WAL for DBAs
 
Distributed Databases Deconstructed: CockroachDB, TiDB and YugaByte DB
Distributed Databases Deconstructed: CockroachDB, TiDB and YugaByte DBDistributed Databases Deconstructed: CockroachDB, TiDB and YugaByte DB
Distributed Databases Deconstructed: CockroachDB, TiDB and YugaByte DB
 
How to Take Advantage of Optimizer Improvements in MySQL 8.0
How to Take Advantage of Optimizer Improvements in MySQL 8.0How to Take Advantage of Optimizer Improvements in MySQL 8.0
How to Take Advantage of Optimizer Improvements in MySQL 8.0
 
Apache Spark Architecture
Apache Spark ArchitectureApache Spark Architecture
Apache Spark Architecture
 
Spark and Object Stores —What You Need to Know: Spark Summit East talk by Ste...
Spark and Object Stores —What You Need to Know: Spark Summit East talk by Ste...Spark and Object Stores —What You Need to Know: Spark Summit East talk by Ste...
Spark and Object Stores —What You Need to Know: Spark Summit East talk by Ste...
 
Tez Shuffle Handler: Shuffling at Scale with Apache Hadoop
Tez Shuffle Handler: Shuffling at Scale with Apache HadoopTez Shuffle Handler: Shuffling at Scale with Apache Hadoop
Tez Shuffle Handler: Shuffling at Scale with Apache Hadoop
 

Similar to LESSONS FROM TPC-DS BENCHMARK

Database and application performance vivek sharma
Database and application performance vivek sharmaDatabase and application performance vivek sharma
Database and application performance vivek sharmaaioughydchapter
 
PostgreSQL query planner's internals
PostgreSQL query planner's internalsPostgreSQL query planner's internals
PostgreSQL query planner's internalsAlexey Ermakov
 
Spark Summit EU talk by Berni Schiefer
Spark Summit EU talk by Berni SchieferSpark Summit EU talk by Berni Schiefer
Spark Summit EU talk by Berni SchieferSpark Summit
 
Islamabad PUG - 7th Meetup - performance tuning
Islamabad PUG - 7th Meetup - performance tuningIslamabad PUG - 7th Meetup - performance tuning
Islamabad PUG - 7th Meetup - performance tuningUmair Shahid
 
Islamabad PUG - 7th meetup - performance tuning
Islamabad PUG - 7th meetup - performance tuningIslamabad PUG - 7th meetup - performance tuning
Islamabad PUG - 7th meetup - performance tuningUmair Shahid
 
PostgreSQL and Benchmarks
PostgreSQL and BenchmarksPostgreSQL and Benchmarks
PostgreSQL and BenchmarksJignesh Shah
 
Auto-Pilot for Apache Spark Using Machine Learning
Auto-Pilot for Apache Spark Using Machine LearningAuto-Pilot for Apache Spark Using Machine Learning
Auto-Pilot for Apache Spark Using Machine LearningDatabricks
 
Performance Tuning and Optimization
Performance Tuning and OptimizationPerformance Tuning and Optimization
Performance Tuning and OptimizationMongoDB
 
Scott Clark, Co-Founder and CEO, SigOpt at MLconf SF 2016
Scott Clark, Co-Founder and CEO, SigOpt at MLconf SF 2016Scott Clark, Co-Founder and CEO, SigOpt at MLconf SF 2016
Scott Clark, Co-Founder and CEO, SigOpt at MLconf SF 2016MLconf
 
MLConf 2016 SigOpt Talk by Scott Clark
MLConf 2016 SigOpt Talk by Scott ClarkMLConf 2016 SigOpt Talk by Scott Clark
MLConf 2016 SigOpt Talk by Scott ClarkSigOpt
 
Oracle Database Performance Tuning Basics
Oracle Database Performance Tuning BasicsOracle Database Performance Tuning Basics
Oracle Database Performance Tuning Basicsnitin anjankar
 
Lessons Learned While Scaling Elasticsearch at Vinted
Lessons Learned While Scaling Elasticsearch at VintedLessons Learned While Scaling Elasticsearch at Vinted
Lessons Learned While Scaling Elasticsearch at VintedDainius Jocas
 
PostgreSQL 9.5 - Major Features
PostgreSQL 9.5 - Major FeaturesPostgreSQL 9.5 - Major Features
PostgreSQL 9.5 - Major FeaturesInMobi Technology
 
Graphite & Metrictank - Meetup Tel Aviv Yafo
Graphite & Metrictank - Meetup Tel Aviv YafoGraphite & Metrictank - Meetup Tel Aviv Yafo
Graphite & Metrictank - Meetup Tel Aviv YafoDieter Plaetinck
 

Similar to LESSONS FROM TPC-DS BENCHMARK (20)

Database and application performance vivek sharma
Database and application performance vivek sharmaDatabase and application performance vivek sharma
Database and application performance vivek sharma
 
PostgreSQL query planner's internals
PostgreSQL query planner's internalsPostgreSQL query planner's internals
PostgreSQL query planner's internals
 
Spark Summit EU talk by Berni Schiefer
Spark Summit EU talk by Berni SchieferSpark Summit EU talk by Berni Schiefer
Spark Summit EU talk by Berni Schiefer
 
Islamabad PUG - 7th Meetup - performance tuning
Islamabad PUG - 7th Meetup - performance tuningIslamabad PUG - 7th Meetup - performance tuning
Islamabad PUG - 7th Meetup - performance tuning
 
Islamabad PUG - 7th meetup - performance tuning
Islamabad PUG - 7th meetup - performance tuningIslamabad PUG - 7th meetup - performance tuning
Islamabad PUG - 7th meetup - performance tuning
 
Intro_2.ppt
Intro_2.pptIntro_2.ppt
Intro_2.ppt
 
Intro.ppt
Intro.pptIntro.ppt
Intro.ppt
 
Intro.ppt
Intro.pptIntro.ppt
Intro.ppt
 
MySQL performance tuning
MySQL performance tuningMySQL performance tuning
MySQL performance tuning
 
PostgreSQL and Benchmarks
PostgreSQL and BenchmarksPostgreSQL and Benchmarks
PostgreSQL and Benchmarks
 
Apache Cassandra at Macys
Apache Cassandra at MacysApache Cassandra at Macys
Apache Cassandra at Macys
 
Auto-Pilot for Apache Spark Using Machine Learning
Auto-Pilot for Apache Spark Using Machine LearningAuto-Pilot for Apache Spark Using Machine Learning
Auto-Pilot for Apache Spark Using Machine Learning
 
Performance Tuning and Optimization
Performance Tuning and OptimizationPerformance Tuning and Optimization
Performance Tuning and Optimization
 
Scott Clark, Co-Founder and CEO, SigOpt at MLconf SF 2016
Scott Clark, Co-Founder and CEO, SigOpt at MLconf SF 2016Scott Clark, Co-Founder and CEO, SigOpt at MLconf SF 2016
Scott Clark, Co-Founder and CEO, SigOpt at MLconf SF 2016
 
MLConf 2016 SigOpt Talk by Scott Clark
MLConf 2016 SigOpt Talk by Scott ClarkMLConf 2016 SigOpt Talk by Scott Clark
MLConf 2016 SigOpt Talk by Scott Clark
 
Oracle Database Performance Tuning Basics
Oracle Database Performance Tuning BasicsOracle Database Performance Tuning Basics
Oracle Database Performance Tuning Basics
 
Lessons Learned While Scaling Elasticsearch at Vinted
Lessons Learned While Scaling Elasticsearch at VintedLessons Learned While Scaling Elasticsearch at Vinted
Lessons Learned While Scaling Elasticsearch at Vinted
 
PostgreSQL 9.5 - Major Features
PostgreSQL 9.5 - Major FeaturesPostgreSQL 9.5 - Major Features
PostgreSQL 9.5 - Major Features
 
Graphite & Metrictank - Meetup Tel Aviv Yafo
Graphite & Metrictank - Meetup Tel Aviv YafoGraphite & Metrictank - Meetup Tel Aviv Yafo
Graphite & Metrictank - Meetup Tel Aviv Yafo
 
techniques.ppt
techniques.ppttechniques.ppt
techniques.ppt
 

More from Sergey Petrunya

New optimizer features in MariaDB releases before 10.12
New optimizer features in MariaDB releases before 10.12New optimizer features in MariaDB releases before 10.12
New optimizer features in MariaDB releases before 10.12Sergey Petrunya
 
MariaDB's join optimizer: how it works and current fixes
MariaDB's join optimizer: how it works and current fixesMariaDB's join optimizer: how it works and current fixes
MariaDB's join optimizer: how it works and current fixesSergey Petrunya
 
Improved histograms in MariaDB 10.8
Improved histograms in MariaDB 10.8Improved histograms in MariaDB 10.8
Improved histograms in MariaDB 10.8Sergey Petrunya
 
Improving MariaDB’s Query Optimizer with better selectivity estimates
Improving MariaDB’s Query Optimizer with better selectivity estimatesImproving MariaDB’s Query Optimizer with better selectivity estimates
Improving MariaDB’s Query Optimizer with better selectivity estimatesSergey Petrunya
 
JSON Support in MariaDB: News, non-news and the bigger picture
JSON Support in MariaDB: News, non-news and the bigger pictureJSON Support in MariaDB: News, non-news and the bigger picture
JSON Support in MariaDB: News, non-news and the bigger pictureSergey Petrunya
 
Optimizer Trace Walkthrough
Optimizer Trace WalkthroughOptimizer Trace Walkthrough
Optimizer Trace WalkthroughSergey Petrunya
 
ANALYZE for Statements - MariaDB's hidden gem
ANALYZE for Statements - MariaDB's hidden gemANALYZE for Statements - MariaDB's hidden gem
ANALYZE for Statements - MariaDB's hidden gemSergey Petrunya
 
Optimizer features in recent releases of other databases
Optimizer features in recent releases of other databasesOptimizer features in recent releases of other databases
Optimizer features in recent releases of other databasesSergey Petrunya
 
MariaDB 10.4 - что нового
MariaDB 10.4 - что новогоMariaDB 10.4 - что нового
MariaDB 10.4 - что новогоSergey Petrunya
 
Using histograms to get better performance
Using histograms to get better performanceUsing histograms to get better performance
Using histograms to get better performanceSergey Petrunya
 
MariaDB Optimizer - further down the rabbit hole
MariaDB Optimizer - further down the rabbit holeMariaDB Optimizer - further down the rabbit hole
MariaDB Optimizer - further down the rabbit holeSergey Petrunya
 
Query Optimizer in MariaDB 10.4
Query Optimizer in MariaDB 10.4Query Optimizer in MariaDB 10.4
Query Optimizer in MariaDB 10.4Sergey Petrunya
 
MariaDB 10.3 Optimizer - where does it stand
MariaDB 10.3 Optimizer - where does it standMariaDB 10.3 Optimizer - where does it stand
MariaDB 10.3 Optimizer - where does it standSergey Petrunya
 
MyRocks in MariaDB | M18
MyRocks in MariaDB | M18MyRocks in MariaDB | M18
MyRocks in MariaDB | M18Sergey Petrunya
 
New Query Optimizer features in MariaDB 10.3
New Query Optimizer features in MariaDB 10.3New Query Optimizer features in MariaDB 10.3
New Query Optimizer features in MariaDB 10.3Sergey Petrunya
 
Histograms in MariaDB, MySQL and PostgreSQL
Histograms in MariaDB, MySQL and PostgreSQLHistograms in MariaDB, MySQL and PostgreSQL
Histograms in MariaDB, MySQL and PostgreSQLSergey Petrunya
 
Common Table Expressions in MariaDB 10.2
Common Table Expressions in MariaDB 10.2Common Table Expressions in MariaDB 10.2
Common Table Expressions in MariaDB 10.2Sergey Petrunya
 
MyRocks in MariaDB: why and how
MyRocks in MariaDB: why and howMyRocks in MariaDB: why and how
MyRocks in MariaDB: why and howSergey Petrunya
 

More from Sergey Petrunya (20)

New optimizer features in MariaDB releases before 10.12
New optimizer features in MariaDB releases before 10.12New optimizer features in MariaDB releases before 10.12
New optimizer features in MariaDB releases before 10.12
 
MariaDB's join optimizer: how it works and current fixes
MariaDB's join optimizer: how it works and current fixesMariaDB's join optimizer: how it works and current fixes
MariaDB's join optimizer: how it works and current fixes
 
Improved histograms in MariaDB 10.8
Improved histograms in MariaDB 10.8Improved histograms in MariaDB 10.8
Improved histograms in MariaDB 10.8
 
Improving MariaDB’s Query Optimizer with better selectivity estimates
Improving MariaDB’s Query Optimizer with better selectivity estimatesImproving MariaDB’s Query Optimizer with better selectivity estimates
Improving MariaDB’s Query Optimizer with better selectivity estimates
 
JSON Support in MariaDB: News, non-news and the bigger picture
JSON Support in MariaDB: News, non-news and the bigger pictureJSON Support in MariaDB: News, non-news and the bigger picture
JSON Support in MariaDB: News, non-news and the bigger picture
 
Optimizer Trace Walkthrough
Optimizer Trace WalkthroughOptimizer Trace Walkthrough
Optimizer Trace Walkthrough
 
ANALYZE for Statements - MariaDB's hidden gem
ANALYZE for Statements - MariaDB's hidden gemANALYZE for Statements - MariaDB's hidden gem
ANALYZE for Statements - MariaDB's hidden gem
 
Optimizer features in recent releases of other databases
Optimizer features in recent releases of other databasesOptimizer features in recent releases of other databases
Optimizer features in recent releases of other databases
 
MariaDB 10.4 - что нового
MariaDB 10.4 - что новогоMariaDB 10.4 - что нового
MariaDB 10.4 - что нового
 
Using histograms to get better performance
Using histograms to get better performanceUsing histograms to get better performance
Using histograms to get better performance
 
MariaDB Optimizer - further down the rabbit hole
MariaDB Optimizer - further down the rabbit holeMariaDB Optimizer - further down the rabbit hole
MariaDB Optimizer - further down the rabbit hole
 
Query Optimizer in MariaDB 10.4
Query Optimizer in MariaDB 10.4Query Optimizer in MariaDB 10.4
Query Optimizer in MariaDB 10.4
 
MariaDB 10.3 Optimizer - where does it stand
MariaDB 10.3 Optimizer - where does it standMariaDB 10.3 Optimizer - where does it stand
MariaDB 10.3 Optimizer - where does it stand
 
MyRocks in MariaDB | M18
MyRocks in MariaDB | M18MyRocks in MariaDB | M18
MyRocks in MariaDB | M18
 
New Query Optimizer features in MariaDB 10.3
New Query Optimizer features in MariaDB 10.3New Query Optimizer features in MariaDB 10.3
New Query Optimizer features in MariaDB 10.3
 
MyRocks in MariaDB
MyRocks in MariaDBMyRocks in MariaDB
MyRocks in MariaDB
 
Histograms in MariaDB, MySQL and PostgreSQL
Histograms in MariaDB, MySQL and PostgreSQLHistograms in MariaDB, MySQL and PostgreSQL
Histograms in MariaDB, MySQL and PostgreSQL
 
Say Hello to MyRocks
Say Hello to MyRocksSay Hello to MyRocks
Say Hello to MyRocks
 
Common Table Expressions in MariaDB 10.2
Common Table Expressions in MariaDB 10.2Common Table Expressions in MariaDB 10.2
Common Table Expressions in MariaDB 10.2
 
MyRocks in MariaDB: why and how
MyRocks in MariaDB: why and howMyRocks in MariaDB: why and how
MyRocks in MariaDB: why and how
 

Recently uploaded

Cyber security and its impact on E commerce
Cyber security and its impact on E commerceCyber security and its impact on E commerce
Cyber security and its impact on E commercemanigoyal112
 
BATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASE
BATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASEBATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASE
BATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASEOrtus Solutions, Corp
 
Xen Safety Embedded OSS Summit April 2024 v4.pdf
Xen Safety Embedded OSS Summit April 2024 v4.pdfXen Safety Embedded OSS Summit April 2024 v4.pdf
Xen Safety Embedded OSS Summit April 2024 v4.pdfStefano Stabellini
 
Unveiling the Future: Sylius 2.0 New Features
Unveiling the Future: Sylius 2.0 New FeaturesUnveiling the Future: Sylius 2.0 New Features
Unveiling the Future: Sylius 2.0 New FeaturesŁukasz Chruściel
 
Catch the Wave: SAP Event-Driven and Data Streaming for the Intelligence Ente...
Catch the Wave: SAP Event-Driven and Data Streaming for the Intelligence Ente...Catch the Wave: SAP Event-Driven and Data Streaming for the Intelligence Ente...
Catch the Wave: SAP Event-Driven and Data Streaming for the Intelligence Ente...confluent
 
Maximizing Efficiency and Profitability with OnePlan’s Professional Service A...
Maximizing Efficiency and Profitability with OnePlan’s Professional Service A...Maximizing Efficiency and Profitability with OnePlan’s Professional Service A...
Maximizing Efficiency and Profitability with OnePlan’s Professional Service A...OnePlan Solutions
 
Intelligent Home Wi-Fi Solutions | ThinkPalm
Intelligent Home Wi-Fi Solutions | ThinkPalmIntelligent Home Wi-Fi Solutions | ThinkPalm
Intelligent Home Wi-Fi Solutions | ThinkPalmSujith Sukumaran
 
Software Project Health Check: Best Practices and Techniques for Your Product...
Software Project Health Check: Best Practices and Techniques for Your Product...Software Project Health Check: Best Practices and Techniques for Your Product...
Software Project Health Check: Best Practices and Techniques for Your Product...Velvetech LLC
 
SpotFlow: Tracking Method Calls and States at Runtime
SpotFlow: Tracking Method Calls and States at RuntimeSpotFlow: Tracking Method Calls and States at Runtime
SpotFlow: Tracking Method Calls and States at Runtimeandrehoraa
 
MYjobs Presentation Django-based project
MYjobs Presentation Django-based projectMYjobs Presentation Django-based project
MYjobs Presentation Django-based projectAnoyGreter
 
英国UN学位证,北安普顿大学毕业证书1:1制作
英国UN学位证,北安普顿大学毕业证书1:1制作英国UN学位证,北安普顿大学毕业证书1:1制作
英国UN学位证,北安普顿大学毕业证书1:1制作qr0udbr0
 
Global Identity Enrolment and Verification Pro Solution - Cizo Technology Ser...
Global Identity Enrolment and Verification Pro Solution - Cizo Technology Ser...Global Identity Enrolment and Verification Pro Solution - Cizo Technology Ser...
Global Identity Enrolment and Verification Pro Solution - Cizo Technology Ser...Cizo Technology Services
 
Best Web Development Agency- Idiosys USA.pdf
Best Web Development Agency- Idiosys USA.pdfBest Web Development Agency- Idiosys USA.pdf
Best Web Development Agency- Idiosys USA.pdfIdiosysTechnologies1
 
办理学位证(UQ文凭证书)昆士兰大学毕业证成绩单原版一模一样
办理学位证(UQ文凭证书)昆士兰大学毕业证成绩单原版一模一样办理学位证(UQ文凭证书)昆士兰大学毕业证成绩单原版一模一样
办理学位证(UQ文凭证书)昆士兰大学毕业证成绩单原版一模一样umasea
 
Folding Cheat Sheet #4 - fourth in a series
Folding Cheat Sheet #4 - fourth in a seriesFolding Cheat Sheet #4 - fourth in a series
Folding Cheat Sheet #4 - fourth in a seriesPhilip Schwarz
 
Cloud Data Center Network Construction - IEEE
Cloud Data Center Network Construction - IEEECloud Data Center Network Construction - IEEE
Cloud Data Center Network Construction - IEEEVICTOR MAESTRE RAMIREZ
 
Dealing with Cultural Dispersion — Stefano Lambiase — ICSE-SEIS 2024
Dealing with Cultural Dispersion — Stefano Lambiase — ICSE-SEIS 2024Dealing with Cultural Dispersion — Stefano Lambiase — ICSE-SEIS 2024
Dealing with Cultural Dispersion — Stefano Lambiase — ICSE-SEIS 2024StefanoLambiase
 
Call Us🔝>༒+91-9711147426⇛Call In girls karol bagh (Delhi)
Call Us🔝>༒+91-9711147426⇛Call In girls karol bagh (Delhi)Call Us🔝>༒+91-9711147426⇛Call In girls karol bagh (Delhi)
Call Us🔝>༒+91-9711147426⇛Call In girls karol bagh (Delhi)jennyeacort
 
Implementing Zero Trust strategy with Azure
Implementing Zero Trust strategy with AzureImplementing Zero Trust strategy with Azure
Implementing Zero Trust strategy with AzureDinusha Kumarasiri
 
Alluxio Monthly Webinar | Cloud-Native Model Training on Distributed Data
Alluxio Monthly Webinar | Cloud-Native Model Training on Distributed DataAlluxio Monthly Webinar | Cloud-Native Model Training on Distributed Data
Alluxio Monthly Webinar | Cloud-Native Model Training on Distributed DataAlluxio, Inc.
 

Recently uploaded (20)

Cyber security and its impact on E commerce
Cyber security and its impact on E commerceCyber security and its impact on E commerce
Cyber security and its impact on E commerce
 
BATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASE
BATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASEBATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASE
BATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASE
 
Xen Safety Embedded OSS Summit April 2024 v4.pdf
Xen Safety Embedded OSS Summit April 2024 v4.pdfXen Safety Embedded OSS Summit April 2024 v4.pdf
Xen Safety Embedded OSS Summit April 2024 v4.pdf
 
Unveiling the Future: Sylius 2.0 New Features
Unveiling the Future: Sylius 2.0 New FeaturesUnveiling the Future: Sylius 2.0 New Features
Unveiling the Future: Sylius 2.0 New Features
 
Catch the Wave: SAP Event-Driven and Data Streaming for the Intelligence Ente...
Catch the Wave: SAP Event-Driven and Data Streaming for the Intelligence Ente...Catch the Wave: SAP Event-Driven and Data Streaming for the Intelligence Ente...
Catch the Wave: SAP Event-Driven and Data Streaming for the Intelligence Ente...
 
Maximizing Efficiency and Profitability with OnePlan’s Professional Service A...
Maximizing Efficiency and Profitability with OnePlan’s Professional Service A...Maximizing Efficiency and Profitability with OnePlan’s Professional Service A...
Maximizing Efficiency and Profitability with OnePlan’s Professional Service A...
 
Intelligent Home Wi-Fi Solutions | ThinkPalm
Intelligent Home Wi-Fi Solutions | ThinkPalmIntelligent Home Wi-Fi Solutions | ThinkPalm
Intelligent Home Wi-Fi Solutions | ThinkPalm
 
Software Project Health Check: Best Practices and Techniques for Your Product...
Software Project Health Check: Best Practices and Techniques for Your Product...Software Project Health Check: Best Practices and Techniques for Your Product...
Software Project Health Check: Best Practices and Techniques for Your Product...
 
SpotFlow: Tracking Method Calls and States at Runtime
SpotFlow: Tracking Method Calls and States at RuntimeSpotFlow: Tracking Method Calls and States at Runtime
SpotFlow: Tracking Method Calls and States at Runtime
 
MYjobs Presentation Django-based project
MYjobs Presentation Django-based projectMYjobs Presentation Django-based project
MYjobs Presentation Django-based project
 
英国UN学位证,北安普顿大学毕业证书1:1制作
英国UN学位证,北安普顿大学毕业证书1:1制作英国UN学位证,北安普顿大学毕业证书1:1制作
英国UN学位证,北安普顿大学毕业证书1:1制作
 
Global Identity Enrolment and Verification Pro Solution - Cizo Technology Ser...
Global Identity Enrolment and Verification Pro Solution - Cizo Technology Ser...Global Identity Enrolment and Verification Pro Solution - Cizo Technology Ser...
Global Identity Enrolment and Verification Pro Solution - Cizo Technology Ser...
 
Best Web Development Agency- Idiosys USA.pdf
Best Web Development Agency- Idiosys USA.pdfBest Web Development Agency- Idiosys USA.pdf
Best Web Development Agency- Idiosys USA.pdf
 
办理学位证(UQ文凭证书)昆士兰大学毕业证成绩单原版一模一样
办理学位证(UQ文凭证书)昆士兰大学毕业证成绩单原版一模一样办理学位证(UQ文凭证书)昆士兰大学毕业证成绩单原版一模一样
办理学位证(UQ文凭证书)昆士兰大学毕业证成绩单原版一模一样
 
Folding Cheat Sheet #4 - fourth in a series
Folding Cheat Sheet #4 - fourth in a seriesFolding Cheat Sheet #4 - fourth in a series
Folding Cheat Sheet #4 - fourth in a series
 
Cloud Data Center Network Construction - IEEE
Cloud Data Center Network Construction - IEEECloud Data Center Network Construction - IEEE
Cloud Data Center Network Construction - IEEE
 
Dealing with Cultural Dispersion — Stefano Lambiase — ICSE-SEIS 2024
Dealing with Cultural Dispersion — Stefano Lambiase — ICSE-SEIS 2024Dealing with Cultural Dispersion — Stefano Lambiase — ICSE-SEIS 2024
Dealing with Cultural Dispersion — Stefano Lambiase — ICSE-SEIS 2024
 
Call Us🔝>༒+91-9711147426⇛Call In girls karol bagh (Delhi)
Call Us🔝>༒+91-9711147426⇛Call In girls karol bagh (Delhi)Call Us🔝>༒+91-9711147426⇛Call In girls karol bagh (Delhi)
Call Us🔝>༒+91-9711147426⇛Call In girls karol bagh (Delhi)
 
Implementing Zero Trust strategy with Azure
Implementing Zero Trust strategy with AzureImplementing Zero Trust strategy with Azure
Implementing Zero Trust strategy with Azure
 
Alluxio Monthly Webinar | Cloud-Native Model Training on Distributed Data
Alluxio Monthly Webinar | Cloud-Native Model Training on Distributed DataAlluxio Monthly Webinar | Cloud-Native Model Training on Distributed Data
Alluxio Monthly Webinar | Cloud-Native Model Training on Distributed Data
 

LESSONS FROM TPC-DS BENCHMARK

  • 1. Lessons for the optimizer from TPC-DS benchmark Sergei Petrunia Query Optimizer developer MariaDB Corporation 2019 MariaDB Developers Unconference New York
  • 2. The goals 1. Want to evaluate/measure the query optimizer 2. Hard to do, optimizer should handle – Different query patterns – Different data distributions, etc 3. How does one do it anyway? – Look at benchmarks – Or “optimizer part” of the benchmarks
  • 3. Benchmarks 1. sysbench – Popular – Does only basic queries, few query patterns 2. DBT-3 (aka TPC-H) – 6 tables, 22 analytic queries – Was used to see some optimizer problems – Limited: ● Uniform data distribution, uncorrelated columns ● ...
  • 4. TPC-DS benchmark ● Obsoletes DBT-3 benchmark ● Richer dataset – 25 Tables, 99 queries – Non-uniform data distributions ● Uses advanced SQL features – 32 queries use CTE – 27 queries use Window Functions – etc ● Could not really run it until MariaDB 10.2 (or MySQL 8)
  • 5. MariaDB still can’t run all of TPC-DS ● 2 Queries: FULL OUTER JOIN ● 10 Queries: ROLLUP + ORDER BY problem (MDEV-17807) ● ~20 more queries have fixable problems – “Every derived table must have an alias”, etc select ... group by a,b,c with rollup order by a,b,c ERROR 1221 (HY000): Incorrect usage of CUBE/ROLLUP and ORDER BY
  • 6. Oracle MySQL and TPC-DS ● ROLLUP + ORDER BY is supported since 8.0.12 ● Doesn’t support FULL OUTER JOIN (2 queries) ● Doesn’t support EXCEPT (1 query) ● Doesn’t support INTERSECT (3 queries)
  • 7. Running queries from TPC-DS ● Data generator creates CSV files – Adjust #define for MySQL/MariaDB ● Query generator produces “streams” from templates – A set of QueryNNN.tpl files – A stream is a text file with one instance of each of the 99 queries – One can add hooks at query start/end ● Queries have a few typos ● There’s no tool to run queries/measure time – Note that the read queries are a subset of benchmark (TpCX$)
  • 8. Getting it to run ● A collection of scripts at https://github.com/spetrunia/tpcds-run-tool ● The goal is a fully-automated run – MariaDB, MySQL, PostgreSQL ● Because we need to play with settings/options
  • 9. Test runs done ● The dataset – Scale=1 – 1.2 GB CSV files – 6 GB when loaded ● The Queries – 10..20 “Streams” ● Tuning – Innodb_buffer_pool=8G (50% RAM) – shared_buffers = 4G (25% RAM)
  • 12. Test results ● … a bit inconclusive – query times varied across my runs (?) ● Time to run one stream = 20 min – 2 hours ● Searching for the source of randomness – Started to work on full automation ● (did I run ANALYZE? Did I have correct with my.cnf parameters?) – Started to look at rngseed in dataset/query generator
  • 14. MariaDB 10.2, 10.4, MySQL 8 ● Scale=1, 6.1 GB data, 8G buffer pool ● rngseed=1234 for both ● Benchmark takes ~20 min ● Query times are very non-uniform +-------------+---------------+ | query_name | QueryTime_ms | +-------------+---------------+ | query72.tpl | 678,321 | | query23.tpl | 80,025 | | query2.tpl | 65,156 | | query39.tpl | 63,761 | | query78.tpl | 63,473 | | query4.tpl | 27,549 | | query31.tpl | 24,344 | | query47.tpl | 19,156 | | query11.tpl | 17,484 | | query74.tpl | 16,571 | | query21.tpl | 16,212 | | query59.tpl | 10,522 | | query88.tpl | 9,965 |
  • 18. PostgreSQL 11 ● There was a “fast” run ● Showing results from the last two runs (both where “slow”) – rngseed=5678 for both – 121 min – rngseed=1234 (data), rngseed=4321 (query) – 145..154 min.
  • 19. Heaviest queries in the run ● Execution time varies ● Is this a query optimizer issue? ● Or different constants in a skewed dataset? +-------------+-----------------+-----------------+--------+ | query_name | PG11-seed5678 | PG11-seed1234 | X | +-------------+-----------------+-----------------+--------+ | query4.tpl | 3,628,830 | 3,578,944 | 1.0139 | | query11.tpl | 2,004,392 | 2,013,597 | 0.9954 | | query1.tpl | 87,981 | 1,947,624 | 0.0452 | | query74.tpl | 693,784 | 641,696 | 1.0812 | | query47.tpl | 624,717 | 539,941 | 1.1570 | | query57.tpl | 116,570 | 112,472 | 1.0364 | | query81.tpl | 22,089 | 47,366 | 0.4663 | | query6.tpl | 27,896 | 27,009 | 1.0328 | | query30.tpl | 11,214 | 11,171 | 1.0038 | | query39.tpl | 10,803 | 10,702 | 1.0094 | | query95.tpl | 16,418 | 10,065 | 1.6312 | ` ● Do we need a “representative collection of datasets”? – Check N datasets?
  • 20. Compare most heavy queries ● Some queries are present in both lists, but some are only in one. ● Not clear +-------------+-----------------+-----------------+--------+ | query_name | PG11-seed5678 | PG11-seed1234 | X | +-------------+-----------------+-----------------+--------+ | query4.tpl | 3,628,830 | 3,578,944 | 1.0139 | | query11.tpl | 2,004,392 | 2,013,597 | 0.9954 | | query1.tpl | 87,981 | 1,947,624 | 0.0452 | | query74.tpl | 693,784 | 641,696 | 1.0812 | | query47.tpl | 624,717 | 539,941 | 1.1570 | | query57.tpl | 116,570 | 112,472 | 1.0364 | | query81.tpl | 22,089 | 47,366 | 0.4663 | | query6.tpl | 27,896 | 27,009 | 1.0328 | | query30.tpl | 11,214 | 11,171 | 1.0038 | | query39.tpl | 10,803 | 10,702 | 1.0094 | | query95.tpl | 16,418 | 10,065 | 1.6312 | ` +-------------+---------------+ | query_name | QueryTime_ms | +-------------+---------------+ | query72.tpl | 678,321 | | query23.tpl | 80,025 | | query2.tpl | 65,156 | | query39.tpl | 63,761 | | query78.tpl | 63,473 | | query4.tpl | 27,549 | | query31.tpl | 24,344 | | query47.tpl | 19,156 | | query11.tpl | 17,484 | | query74.tpl | 16,571 | | query21.tpl | 16,212 | | query59.tpl | 10,522 | MariaDB PostgreSQL
  • 21. Observations about the benchmark ● rngseed on the dataset matters A LOT – What is a representative set of rngseed values? ● rngseed on query streams – much less ● Hardware? ● Queries are not equal – Heavy vs lightweight queries – Is SUM(query_time) an adequate metric? ● Wont see that a fast query got 10x slower
  • 22. Other observations ● Both DBT-3 and TPC-DS workloads are relevant for the optimizer – Condition selectivities – Semi-join optimizations – … ● But don’t match the optimizer issues we see – ORDER BY … LIMIT optimization – Long IN-list – …
  • 24. Extra – PostgreSQL 11, parallel query? ● Trying on a run with both rngseed=5678: ● Parallel settings max_parallel_workers_per_gather=8 (the default was 2) dynamic_shared_memory_type=posix show max_worker_processes= 8 ● Results – Only saw one core to be occupied – The run still took 121 min, didin’t see any speedup
  • 25. Try a parallel query select sum(inv_quantity_on_hand*i_current_price) from inventory, item where i_item_sk=inv_item_sk; QUERY PLAN --------------------------------------------------------------------------------- Aggregate (cost=301495.25..301495.26 rows=1 width=32) -> Hash Join (cost=1635.00..213408.54 rows=11744894 width=10) Hash Cond: (inventory.inv_item_sk = item.i_item_sk) -> Seq Scan on inventory (cost=0.00..180935.94 rows=11744894 width=8) -> Hash (cost=1410.00..1410.00 rows=18000 width=10) -> Seq Scan on item (cost=0.00..1410.00 rows=18000 width=10) ● max_parallel_workers_per_gather=0
  • 26. Try a parallel query select sum(inv_quantity_on_hand*i_current_price) from inventory, item where i_item_sk=inv_item_sk; QUERY PLAN ---------------------------------------------------------------------------------------------------- Finalize Aggregate (cost=125048.98..125048.99 rows=1 width=32) -> Gather (cost=125048.55..125048.96 rows=4 width=32) Workers Planned: 4 -> Partial Aggregate (cost=124048.55..124048.56 rows=1 width=32) -> Parallel Hash Join (cost=1468.23..102026.87 rows=2936224 width=10) Hash Cond: (inventory.inv_item_sk = item.i_item_sk) -> Parallel Seq Scan on inventory (cost=0.00..92849.24 rows=2936224 width=8) -> Parallel Hash (cost=1335.88..1335.88 rows=10588 width=10) -> Parallel Seq Scan on item (cost=0.00..1335.88 rows=10588 width=10) ● max_parallel_workers_per_gather=8
  • 27. Try a parallel query select sum(inv_quantity_on_hand*i_current_price) from inventory, item where i_item_sk=inv_item_sk; ● Results – max_parallel_workers_per_gather=8: 1.0 sec – max_parallel_workers_per_gather=0: 3.8 sec ● Didn’t see anything like that in TPC-DS benchmark