moving ,” and - percona · moving a “large,” “ ... relatively low barrier to entry, almost...

Big Bad PostgreSQL: A Case Study

1

Moving a“large,”

“complicated,” andmission-criticaldatawarehouse

from Oracleto PostgreSQL

for cost control.

About the Speaker

Scalable Internet ArchitecturesWith an estimated one billion users worldwide, the Internet today is nothing less than a global subculture with immense diversity, incredible size, and wide geographic reach. With arelatively low barrier to entry, almost anyone can register a domain name today and potentiallyprovide services to people around the entire world tomorrow. But easy entry to web-basedcommerce and services can be a double-edged sword. In such a market, it is typically muchharder to gauge interest in advance, and the negative impact of unexpected customer trafficcan turn out to be devastating for the unprepared.

In Scalable Internet Architectures, renowned software engineer and architect TheoSchlossnagle outlines the steps and processes organizations can follow to build online services that can scale well with demand—both quickly and economically. By making intelligent decisions throughout the evolution of an architecture, scalability can be a matter of engineering rather than redesign, costly purchasing, or black magic.

Filled with numerous examples, anecdotes, and lessons gleaned from the author’s years of experience building large-scale Internet services, Scalable Internet Architectures is boththought-provoking and instructional. Readers are challenged to understand first, before theystart a large project, how what they are building will be used, so that from the beginning they can design for scalability those parts which need to scale. With the right approach, itshould take no more effort to design and implement a solution that scales than it takes to build something that will not—and if this is the case, Schlossnagle writes, respect yourself and build it right.

Schlossnagle

DEVELOPER’S LIBRARY

$49.99 USA / $61.99 CAN / £35.99 Net UK

Internet/Programming

www.developers-library.com

DEVELOPER’S LIBRARY

Cover image © Digital Vision/Getty Images

Theo Schlossnagle is a principal at OmniTI Computer Consulting, where he providesexpert consulting services related to scalable Internet architectures, database replication,and email infrastructure. He is the creator of the Backhand Project and the Ecelerity MTA,and spends most of his time solving the scalability problems that arise in high performance and highly distributed systems.

Scalable Internet Architectures

ScalabilityPerformanceSecurity

www.omniti.com

Scalable InternetArchitectures

Theo Schlossnagle

S32699X_Scalable_Internet.qxd 6/23/06 3:31 PM Page 1 • Principal @ OmniTI

• Open Source

mod_backhand, spreadlogd,OpenSSH+SecurID, Daiquiri,Wackamole, libjlog, Spread, Reconnoiter, etc.

• Closed Source

Message Systems MTA, Message Central

• Author

Scalable Internet Architectures

Datawarehouse

Log Importer

Data Exporter

OLTP warm backup

OLTP

Oracle 8i

0.5 TB

Hitachi

0.25 TB

JBOD

Oracle 8i

0.75 TB

JBOD

MySQL 4.1

1.2 TB

IDE RAID

MySQL

log importer

1.2 TB

SATA

RAID

Oracle 8i

0.5 TB

Hitachi

1.5 TB

MTI

Overall Architecture

OLTP instance:drives the site

Warm spare

bulk selects / data exports

Log import andprocessing

Database Situation

Database Situation

• The problems:

• The database is growing.

• The OLTP and ODS/warehouse are too slow.

• A lot of application code against the OLTP system.

• Minimal application code against the ODS system.

Database Situation

• The problems:





• Oracle:

• Licensed per processor.

• Really, really, really expensive on a large scale.

Database Situation

• The problems:





• Oracle:

• Licensed per processor.

• Really, really, really expensive on a large scale.

• PostgreSQL:

• No licensing costs.

• Good support for complex queries.

Database Choices

Database Choices

• Must keep Oracle on OLTP• Complex, Oracle-specific web application.• Need more processors.

Database Choices


• ODS: Oracle not required.• Complex queries from limited sources.• Needs more space and power.

Database Choices


• ODS: Oracle not required.• Complex queries from limited sources.• Needs more space and power.

• Result:• Move ODS Oracle licenses to OLTP• Run PostgreSQL on ODS

PostgreSQL gotchas

PostgreSQL gotchas

• For an OLTP system that does thousands of updates per second, vacuuming is a hassle.

PostgreSQL gotchas


• No upgrades?!

PostgreSQL gotchas


• No upgrades?!

• Less community experience with large databases.

PostgreSQL gotchas


• No upgrades?!

• Less community experience with large databases.

• Replication features less evolved.

PostgreSQL ♥ ODS

PostgreSQL ♥ ODS

• Mostly inserts.

PostgreSQL ♥ ODS

• Mostly inserts.

• Updates/Deletes controlled, not real-time.

PostgreSQL ♥ ODS

• Mostly inserts.


• pl/perl (leverage DBI/DBD for remote database connectivity).

PostgreSQL ♥ ODS

• Mostly inserts.



• Monster queries.

PostgreSQL ♥ ODS

• Mostly inserts.



• Monster queries.

• Extensible.

Choosing Linux

Choosing Linux

• Popular, liked, good community support.

Choosing Linux


• Chronic problems:

Choosing Linux



• kernel panics

Choosing Linux



• kernel panics

• filesystems remounting read-only

Choosing Linux



• kernel panics


• filesystems don’t support snapshots

Choosing Linux



• kernel panics



• LVM is clunky on enterprise storage

Choosing Linux



• kernel panics



• LVM is clunky on enterprise storage

• 20 outages in 4 months

Choosing Solaris 10

Choosing Solaris 10

• Switched to Solaris 10

Choosing Solaris 10


• No crashes, better system-level tools.

Choosing Solaris 10



• prstat, iostat, vmstat, smf, fault-management.

Choosing Solaris 10




• ZFS

Choosing Solaris 10




• ZFS

• snapshots (persistent), BLI backups.

Choosing Solaris 10




• ZFS


• Excellent support for enterprise storage.

Choosing Solaris 10




• ZFS



• DTrace.

Choosing Solaris 10




• ZFS



• DTrace.

• Free (too).

Oracle features we need


• Partitioning


• Partitioning

• Statistics and Aggregations


• Partitioning


• rank over partition, lead, lag, etc.


• Partitioning



• Large selects (100GB)


• Partitioning




• Autonomous transactions


• Partitioning




• Autonomous transactions

• Replication from Oracle (to Oracle)

Partitioning

For large data sets:

Partitioning

pgods=# select count(1) from ods.ods_tblpick_super;


Partitioning

pgods=# select count(1) from ods.ods_tblpick_super; count ------------ 1790994512(1 row)


Partitioning

• Next biggest tables: 850m, 650m, 590m



Partitioning


• Allows us to cluster data over specific ranges (by date in our case)



Partitioning



• Simple, cheap archiving and removal of data.



Partitioning



• Simple, cheap archiving and removal of data.

• Can put ranges used less often in different tablespaces (slower, cheaper storage)



Partitioning PostgreSQL style


• PostgreSQL doesn’t support partition...



• It supports inheritance... (what’s this?)




• some crazy object-relation paradigm.





• We can use it to implement partitioning:






• One master table with no rows.







• Child tables that have our partition constraints.







• Child tables that have our partition constraints.

• Rules on the master table for insert/update/delete.

Partitioning PostgreSQL realized


• Cheaply add new empty partitions



• Cheaply remove old partitions




• Migrate less-often-accessed partitions to slower storage





• Different indexes strategies per partition






• PostgreSQL >8.1 supports constraint checking on inherited tables.







• smarter planning







• smarter planning

• smarter executing

RANK OVER PARTITION

• In Oracle:

• In PostgreSQL:

With 8.4, we have windowing functions

RANK OVER PARTITION

• In Oracle:

• In PostgreSQL:

select userid, email from ( select u.userid, u.email, row_number() over (partition by u.email order by userid desc) as position from (...)) where position = 1


RANK OVER PARTITION

• In Oracle:

• In PostgreSQL:

select userid, email from ( select u.userid, u.email, row_number() over (partition by u.email order by userid desc) as position from (...)) where position = 1

FOR v_row IN select u.userid, u.email from (...) order by email, userid descLOOP IF v_row.email != v_last_email THEN RETURN NEXT v_row; v_last_email := v_row.email; v_rownum := v_rownum + 1; END IF;END LOOP;


Large SELECTs

• Application code does:

Large SELECTs

select u.*, b.browser, m.lastmess from ods.ods_users u, ods.ods_browsers b, ( select userid, min(senddate) as senddate from ods.ods_maillog group by userid ) m, ods.ods_maillog l where u.userid = b.userid and u.userid = m.userid and u.userid = l.userid and l.senddate = m.senddate;


Large SELECTs

• The width of these rows is about 2k



Large SELECTs


• 50 million row return set



Large SELECTs


• 50 million row return set

• > 100 GB of data



The Large SELECT Problem

• libpq will buffer the entire result in memory.

• This affects language bindings (DBD::Pg).

• This is an utterly deficient default behavior.

• This can be avoided by using cursors

• Requires the app to be PostgreSQL specific.

• You open a cursor.

• Then FETCH the row count you desire.

Big SELECTs the Postgres way

The previous “big” query becomes:


DECLARE CURSOR bigdump FORselect u.*, b.browser, m.lastmess from ods.ods_users u, ods.ods_browsers b, ( select userid, min(senddate) as senddate from ods.ods_maillog group by userid ) m, ods.ods_maillog l where u.userid = b.userid and u.userid = m.userid and u.userid = l.userid and l.senddate = m.senddate;



DECLARE CURSOR bigdump FORselect u.*, b.browser, m.lastmess from ods.ods_users u, ods.ods_browsers b, ( select userid, min(senddate) as senddate from ods.ods_maillog group by userid ) m, ods.ods_maillog l where u.userid = b.userid and u.userid = m.userid and u.userid = l.userid and l.senddate = m.senddate;

FETCH FORWARD 10000 FROM bigdump;


Then, in a loop:

Autonomous Transactions


• In Oracle we have over 2000 custom stored procedures.



• During these procedures, we like to:




• COMMIT incrementallyUseful for long transactions (update/delete) that need not be atomic -- incremental COMMITs.




• COMMIT incrementallyUseful for long transactions (update/delete) that need not be atomic -- incremental COMMITs.

• start a new top-level txn that can COMMITUseful for logging progress in a stored procedure so that you know how far you progessed and how long each step took even if it rolls back.

PostgreSQL shortcoming


• PostgreSQL simply does not support Autonomous transactions and to quote core developers “that would be hard.”



• When in doubt, use brute force.



• When in doubt, use brute force.

• Use pl/perl to use DBD::Pg to connect to ourselves (a new backend) and execute a new top-level transaction.

Replication

Replication

• Cross vendor database replication isn’t too difficult.

Replication


• Helps a lot when you can do it inside the database.

Replication



• Using dbi-link (based on pl/perl and DBI) we can.

Replication




• We can connect to any remote database.

Replication





• INSERT into local tables directly from remote SELECT statements.[snapshots]

Replication





• INSERT into local tables directly from remote SELECT statements.[snapshots]

• LOOP over remote SELECT statements and process them row-by-row.[replaying remote DML logs]

Replication (really)


• Through a combination of snapshotting and DML replay we:

• replicate over into over 2000 tables in PostgreSQL from Oracle

• snapshot replication of 200

• DML replay logs for 1800


• Through a combination of snapshotting and DML replay we:

• replicate over into over 2000 tables in PostgreSQL from Oracle

• snapshot replication of 200

• DML replay logs for 1800

• PostgreSQL to Oracle is a bit harder

• out-of-band export and imports

New Architecture

• Master: Sun v890 and Hitachi AMS + warm standby running Oracle (1TB)

• Logs: several customs running MySQL instances (2TB each)

• ODS BI: 2x Sun v40 running PostgreSQL 8.3 (6TB on Sun JBODs on ZFS each)

• ODS archive: 2x custom running PostgreSQL 8.3 (14TB internal storage on ZFS each)

PostgreSQL is Lacking

• No upgrades (AYFKM).

• pg_dump is too intrusive.

• Poor system-level instrumentation.

• Poor methods to determine specific contention.

• It relies on the operating system’s filesystem cache.(which make PostgreSQL inconsistent across it’s supported OS base)

Enter Solaris

• Solaris is a UNIX from Sun Microsystems.

• Is it different than other UNIX/UNIX-like systems?

• Mostly it isn’t different (hence the term UNIX)

• It does have extremely strong ABI backward compatibility.

• It’s stable and works well on large machines.

• Solaris 10 shakes things up a bit:

• DTrace

• ZFS

• Zones

Solaris / ZFS

• ZFS: Zettaback Filesystem.

• 264 snapshots, 248 files/directory, 264 bytes/filesystem,278 (256 ZiB) bytes in a pool, 264 devices/pool, 264 pools/system

• Extremely cheap differential backups.

• I have a 5 TB database, I need a backup!

• No rollback in your database? What is this? MySQL?

• No rollback in your filesystem?

• ZFS has snapshots, rollback, clone and promote.

• OMG! Life altering features.

• Caveat: ZFS is slower than alternatives, by about 10% with tuning.

Solaris / Zones

• Zones: Virtual Environments.

• Shared kernel.

• Can share filesystems.

• Segregated processes and privileges.

• No big deal for databases, right?

But Wait!

Solaris / ZFS + Zones = Magic Juju

• ZFS snapshot, clone, delegate to zone, boot and run.

• When done, halt zone, destroy clone.

• We get a point-in-time copy of our PostgreSQL database:

• read-write,

• low disk-space requirements,

• NO LOCKS! Welcome back pg_dump,you don’t suck (as much) anymore.

• Fast snapshot to usable copy time:

• On our 20 GB database: 1 minute.

• On our 1.2 TB database: 2 minutes.

https://labs.omniti.com/trac/pgsoltools/browser/trunk/pitr_clone/clonedb_startclone.sh



ZFS: how I saved my soul.

• Database crash. Bad. 1.2 TB of data... busted.The reason Robert Treat looks a bit older than he should.

• xlogs corrupted. catalog indexes corrupted.

• Fault? PostgreSQL bug? Bad memory? Who knows?

• Trial & error on a 1.2 TB data set is a cruel experience.

• In real-life, most recovery actions are destructive actions.

• PostgreSQL is no different.

• Rollback to last checkpoint (ZFS), hack postgres code, try, fail, repeat.

Let DTrace open your eyes

• DTrace: Dynamic Tracing

• Dynamically instrument “stuff” in the system:

• system calls (like strace/truss/ktrace).

• process/scheduler activity (on/off cpu, semaphores, conditions).

• see signals sent and received.

• trace kernel functions, networking.

• watch I/O down to the disk.

• user-space processes, each function... each machine instruction!

• Add probes into apps where it makes sense to you.

Can you see what I see?

• There is EXPLAIN... when that isn’t enough...

• There is EXPLAIN ANALYZE... when that isn’t enough.

• There is DTrace.

; dtrace -q -n ‘postgresql*:::statement-start{ self->query = copyinstr(arg0); self->ok=1;}io:::start/self->ok/{ @[self->query, args[0]->b_flags & B_READ ? "read" : "write", args[1]->dev_statname] = sum(args[0]->b_bcount);}’dtrace: description 'postgres*:::statement-start' matched 14 probes^C

select count(1) from c2w_ods.tblusers where zipcode between 10000 and 11000; read sd1 16384select division, sum(amount), avg(amount) from ods.billings where txn_timestamp between ‘2006-01-01 00:00:00’ and ‘2006-04-01 00:00:00’ group by division; read sd2 71647232

OmniTI Labs / pgsoltools

• https://labs.omniti.com/trac/pgsoltools

• Where we stick out PostgreSQL on Solaris goodies...

• like pg_file_stress

FILENAME/DBOBJECT READS WRITES # min avg max # min avg maxalldata1__idx_remove_domain_external 1 12 12 12 398 0 0 0slowdata1__pg_rewrite 1 12 12 12 0 0 0 0slowdata1__pg_class_oid_index 1 0 0 0 0 0 0 0slowdata1__pg_attribute 2 0 0 0 0 0 0 0alldata1__mv_users 0 0 0 0 4 0 0 0slowdata1__pg_statistic 1 0 0 0 0 0 0 0slowdata1__pg_index 1 0 0 0 0 0 0 0slowdata1__pg_index_indexrelid_index 1 0 0 0 0 0 0 0alldata1__remove_domain_external 0 0 0 0 502 0 0 0alldata1__promo_15_tb_full_2 19 0 0 0 11 0 0 0slowdata1__pg_class_relname_nsp_index 2 0 0 0 0 0 0 0alldata1__promo_177intaoltest_tb 0 0 0 0 1053 0 0 0slowdata1__pg_attribute_relid_attnum_index 2 0 0 0 0 0 0 0alldata1__promo_15_tb_full_2_pk 2 0 0 0 0 0 0 0alldata1__all_mailable_2 1403 0 0 423 0 0 0 0alldata1__mv_users_pkey 0 0 0 0 4 0 0 0

https://labs.omniti.com/trac/pgsoltools

https://labs.omniti.com/trac/pgsoltools

Results

Results

• Move ODS Oracle licenses to OLTP

Results


• Run PostgreSQL on ODS

Results



• Save $800k in license costs.

Results




• Spend $100k in labor costs.

Results




• Spend $100k in labor costs.

• Learn a lot.

Thanks!

• Thank you.

• http://omniti.com/does/postgresql

• We’re hiring, but only if you love:

• lots of data on lots of disks on lots of big boxes

• smart people

• hard problems

• more than one database technology (including PostgreSQL)

• responsibility

moving ,” and - percona · moving a “large,” “ ... relatively low barrier to entry, almost...

Documents