Optimizing your app by understanding your Postgres | RailsConf 2019 | Samay Sharma

Samay Sharma | RailsConf 2019
Optimizing your app by understanding
your Postgres database
Samay Sharma
Solutions Engineering Manager
RailsConf 2019 | Minneapolis | April 2019

View Slide

2
• Citus - Open Source Extension to
Scale out Postgres
• Manage a team of Solutions
Engineers
• Work with clients
• Fun(?) fact : Recently got married J

View Slide


View Slide

Challenge: Relating issues to the database
• Problems you’re facing from an application perspective
• Tied to database
• How do I find the root cause?
• How do I solve it?
4

View Slide

Solution: Using Postgres’ statistics
• Postgres captures, uses and exposes many statistics.
• Monitoring statistics – Statistics about system activity
• Query planner statistics – Statistics used by the planner to choose the right query plan.
• Server Administrator statistics – Statistics about replication, size of database, tables, etc.
• It’s important to know how to use them.
5

View Slide

Problem: My application is slow
• Symptoms: Application users are seeing slow query performance.
• Your monitoring tool shows you that most of the time is going in database
calls.
• Your page is making 100s of database calls, you don’t know which query
is slow.
6

View Slide

Cache hit ratio
• Gives an idea of how much of your data is coming from the PostgreSQL
buffercache.
• For transactional apps, want to keep if >95%, ideally around 99%.
7

View Slide

How do I measure?
SELECT
sum(heap_blks_read) as heap_read,
sum(heap_blks_hit) as heap_hit,
sum(heap_blks_hit) / (sum(heap_blks_hit) + sum(heap_blks_read)) as ratio
FROM
pg_statio_user_tables;
8

View Slide

Digging further
• If cache hit ratio is low, that could mean a lot of things.
• Maybe you have a lot of bloat.
• Maybe autovacuum is not tuned well enough.
• Maybe your queries are not optimized and you are doing a lot of sequential scans.
• Maybe you have a lot of unused indexes.
• Or, you actually have less resources and you need to scale out with Citus.
10

View Slide

Uses for pg_stat_database
• Find out rows fetched vs returned by queries to the database
• Find out the insert / update / delete ratio for your database
• Total number of transactions executed / throughput
12

View Slide

Uses for pg_stat_user_tables
• Approximate number of live / dead tuples (rows) in a table
• Find out if the table is an insert / update / delete heavy table.
• Find out if your table is being auto vacuumed sufficiently or not.
• Find out if most scans on the table are sequential scans or index scans.
14

View Slide

Uses for pg_stat_user_indexes
• Figuring out unused indexes
• Ratio of index rows read vs fetched
16

View Slide

Back to cache hit rate
• Understand why you had a bad hit rate
• You see database patterns which don’t align with the application – Dive deeper into them
• Is there a lot of bloat, are some tables not vacuumed often enough – Tune autovacuum
• Are you doing a lot of sequential scans - Create indexes
• Maybe you have a lot of unnecessary indexes – Delete them
17

View Slide

Generic vs specific optimizations
• Follow above practices to make generic optimizations when your
database is slow.
• How do you speed up a particular webpage?
• Or how do you find out what specific queries are slow?
18

View Slide

How do I find what queries are slow?
19

View Slide

What is `pg_stat_statements` ?
• PostgreSQL extension which tracks execution statistics of all SQL
statements executed by a server
• Normalized so similar queries grouped together
• Exposed to user w/ view named `pg_stat_statements`
20

View Slide

`pg_stat_statements` view
• Query: User_id, db_id, query_id, query text, # of times executed
• Timing: min_time, max_time, total_time, mean_time, stddev_time
• Shared blocks hit, read, written, dirtied
• Local blocks hit, read, written, dirtied
• I/O times: total time spent in reading & writing blocks.
• Enable track_io_timing to get these, otherwise will be empty
21

View Slide

Examples
• Top 10 time consuming queries
select query, total_time from pg_stat_statements ORDER BY total_time DESC limit 10
• Top 10 queries spending most time on disk reads
select query, blk_read_time from pg_stat_statements ORDER BY blk_read_time DESC limit 10
• Top 10 most common queries
select query, calls from pg_stat_statements ORDER BY calls DESC limit 10
22

View Slide

SELECT rolname,
calls,
total_time,
mean_time,
max_time,
stddev_time,
rows,
regexp_replace(query, '[ \t\n]+', ' ', 'g') AS query_text
FROM pg_stat_statements
JOIN pg_roles r ON r.oid = userid
WHERE calls > 100
AND rolname NOT LIKE '%backup'
ORDER BY mean_time DESC
LIMIT 15;
-[ RECORD 1] ----------------------------------------------------------------------------------------------------
- ---------------------------------------------------------------------------------------------------------------
---
rolname | citus
calls | 141
total_time | 121404.290523
mean_time | 861.023337042553
max_time | 3718.828292
stddev_time | 913.641891936023
rows | 846
query_text | SELECT ads.campaign_id, COUNT(*) FROM ads JOIN impressions i ON (ads.id = ad_id AND ads.company_id
= i.company_id) WHERE ads.campaign_id IN ($1,$2,$3,$4,$5,$6,$7,$8,$9,$10,$11,$12,$13,$14) AND ads.company_id =
$15 AND seen_at > now()::date GROUP BY ads.campaign_id
Source: http://geekmonkey.org/2017/11/optimizing-postgresql-queries-using-pg_stat_statements-and-pg_buffercache/

View Slide

EXPLAIN ANALYZE and tune them!
EXPLAIN ANALYZE SELECT * FROM tbl where col1 = 1;
QUERY PLAN
-----------------------------------------------------------------------------------------------------------
Seq Scan on tbl (cost=0.00..169247.80 rows=9584 width=8) (actual time=0.641..622.851 rows=10000 loops=1)
Filter: (col1 = 1)
Rows Removed by Filter: 9990000
Planning time: 0.051 ms
Execution time: 623.185 ms
(5 rows)
24

View Slide

Interesting things about EXPLAIN
EXPLAIN ANALYZE SELECT * FROM tbl where col1 = 1 and col2 = 0;
QUERY PLAN
----------------------------------------------------------------------------------------------------------
Filter: ((col1 = 1) AND (col2 = 0))
(5 rows)
25

View Slide

Estimates vs actual
• Postgres tells you what it thinks even in the EXPLAIN plans.
• Sometimes reason for slow queries might just be that Postgres has
incorrect statistics. Notice and fix those mismatches.
• Try to ANALYZE with larger default_statistics_target.
26

View Slide

Multi-column statistics
• Postgres only captures only single column statistics.
• Sometimes, those are not enough because they can’t cover correlations.
• But, you can tell PG two columns are related.
27

View Slide

CREATE STATISTICS
CREATE STATISTICS s1 (dependencies) on col1, col2 from tbl;
ANALYZE tbl;
EXPLAIN ANALYZE SELECT * FROM tbl where col1 = 1 and col2 = 0;
QUERY PLAN
-----------------------------------------------------------------------------------------------------------
Filter: ((col1 = 1) AND (col2 = 0))
(5 rows)
28

View Slide

Example of impact
29

View Slide

Summary of specific query optimization
• Find the right queries to optimize using pg_stat_statements
• Use EXPLAIN ANALYZE and tune the slowest parts
• Look at differences between estimated and actual and find ways to fix
them.
30

View Slide

Bonus: Quick Estimates
31

View Slide

pg_stats
select * from pg_stats where tablename = 'tbl' and attname = 'col1';
-[ RECORD 1 ]----------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------
schemaname | public
tablename | tbl
attname | col1
inherited | f
null_frac | 0
avg_width | 4
n_distinct | 1000
most_common_vals | {945,394,677,839,551,636,805,3,743,799,299,967,295,396,400,403,663,740}
most_common_freqs |
{0.00176667,0.0017,0.0016,0.0016,0.00156667,0.00156667,0.00156667,0.00153333,0.00153333,0.00153333,0.0015,0.0015,0.00146667,0.00146667,0.00146667,0.00146667,0.00146667,0.00146667}
histogram_bounds |
{0,10,20,29,39,49,58,68,78,88,99,109,119,129,139,149,158,168,177,186,195,205,215,225,235,245,254,263,273,283,292,303,313,323,331,341,350,360,371,380,391,406,415,425,434,443,454,465,474,483
,492,502,511,521,531,541,550,562,571,581,591,600,610,620,630,642,652,662,672,682,692,702,712,722,731,742,754,763,773,783,792,803,813,824,834,845,855,866,876,886,897,907,917,928,938,948,958
,969,980,991,999}
correlation | 1
most_common_elems |
most_common_elem_freqs |
elem_count_histogram |
32

View Slide

Uses for pg_stats
• What is the null ratio for a particular column?
• Find the approximate number of distinct values for a column.
• What are the most common values for a column and what are the
approximate number of rows with those values?
33

View Slide

Approximate count
• Do you know how to get a very fast approximate count of rows in a table?
SELECT
relname, reltuples
FROM
pg_class
WHERE
relname = 'tbl’;
relname | reltuples
---------+-----------
tbl | 1e+07
(1 row)
34

View Slide

Summary
• Postgres exposes a lot of stats to you. In this talk, we covered mainly
performance.
• Pg_stat_* tables to find index, table, IO, vacuum statistics.
• Enable pg_stat_statements to find out your slow queries.
35

View Slide

Summary
• Use EXPLAIN ANALYZE or EXPLAIN (ANALYZE, BUFFERS) to see what
the planner is actually doing.
• Use pg_stats and pg_class to get good approximations and understand
postgres’ understanding of your data.
36

View Slide

aka.ms/citus-azure.int

View Slide

Questions?
@citusdata
@samay_sharma
github.com/citusdata/citus
@microsoft

View Slide

Useful readings – Based on talk feedback
• Measuring Bloat (and more) : https://www.citusdata.com/blog/2017/10/20/monitoring-your-bloat-in-postgres/
• Queries for unused indexes and cache hit ratio : https://www.citusdata.com/blog/2017/09/29/what-performance-
can-you-expect-from-postgres/
• Tuning Autovacuum : https://www.citusdata.com/blog/2016/11/04/autovacuum-not-the-enemy/
• Multi-column statistics : https://www.citusdata.com/blog/2018/03/06/postgres-planner-and-its-usage-of-
statistics/
39

View Slide

Optimizing your app by understanding your Postgres | RailsConf 2019 | Samay Sharma

Optimizing your app by understanding your Postgres | RailsConf 2019 | Samay Sharma

Citus Data

More Decks by Citus Data

Other Decks in Technology

Featured

Transcript