Rapid Upgrades With Pg_Migrator

BRUCE MOMJIAN,ENTERPRISEDB

March, 2010

AbstractPg_Migrator allows migration between major releases of Postgreswithout a data dump/reload. This presentation explains howpg_migrator works.

Creative Commons Attribution License http://momjian.us/presentations

Traditional Postgres Upgrade Options� pg_dump (logical dump)/restore

� Slony

Rapid Upgrades With Pg-Migrator 1

Why Upgrading Postgres Is Complex� New features often require system table changes

� However, the internal data format rarely changes

Rapid Upgrades With Pg-Migrator 2

Why Pg_Migrator� Very fast upgrades

� Optionally no additional disk space

pg_migrator installs new system tables while using data files from theprevious Postgres version.

Rapid Upgrades With Pg-Migrator 3

How It Works: Initial Setup

1

2

3 6

5

4 7

8

9

pg_class

User Tables

15 21 27

262014

13 19 25

241812

17 2311

10 16 22

System Tables

clog

Old Cluster

1

2

3 6

5

4 7

8

9

pg_class

User Tables

System Tables

clog

New Cluster

Rapid Upgrades With Pg-Migrator 4

Decouple New Clog Via Freezing

1

2

3 6

5

4 7

8

9

pg_class

User Tables

15 21 27

262014

13 19 25

241812

17 2311

10 16 22

System Tables

clog

Old Cluster

1

2

3 6

4 7

8

9

pg_class

User Tables

System Tables

clog

New Cluster

5Freeze

X X

Rapid Upgrades With Pg-Migrator 5

Transfer Clog and XID

1

2

3 6

5

4 7

8

9

pg_class

User Tables

15 21 27

262014

13 19 25

241812

17 2311

10 16 22

System Tables

clog

Old Cluster

1

2

3 6

5

4 7

8

9

pg_class

User Tables

System Tables

clog

New Cluster

controldata controldataxid

Rapid Upgrades With Pg-Migrator 6

Get Schema Dump

1

2

3 6

5

4 7

8

9

pg_class

User Tables

15 21 27

262014

13 19 25

241812

17 2311

10 16 22

System Tables

clog

Old Cluster

1

2

3 6

5

4 7

8

9

pg_class

User Tables

System Tables

clog

New Cluster

pg_dumpall - -schema

Rapid Upgrades With Pg-Migrator 7

Reserve TOAST OIDs Using Relfilenodes (pre-9.0)

1

2

3 6

5

4 7

8

9

pg_class

User Tables

15 21 27

262014

13 19 25

241812

17 2311

10 16 22

System Tables

clog

Old Cluster

1

2

3 6

5

4 7

8

9

pg_class

User Tables

26

18

System Tables

clog

New Cluster

This is necessary because heap references to TOAST tables contain theTOAST oids for easy lookup.Rapid Upgrades With Pg-Migrator 8

Restore Schema In New Cluster

1

2

3 6

5

4 7

8

9

pg_class

User Tables

15 21 27

262014

13 19 25

241812

17 2311

10 16 22

System Tables

clog

Old Cluster

1

2

3 6

5

4 7

8

9

pg_class

User Tables

2640

39

31

System Tables

clog

New Cluster

pg_dumpall - -schema

30

32

33

34

35

36

37

41

42

43

44

45

47

18

Rapid Upgrades With Pg-Migrator 9

Connect TOAST Placeholders To the Proper Relations(pre-9.0)

1

2

3 6

5

4 7

8

9

pg_class

User Tables

15 21 27

262014

13 19 25

241812

17 2311

10 16 22

System Tables

clog

Old Cluster

1

2

3 6

5

4 7

8

9

pg_class

User Tables

26

39

31

System Tables

clog

New Cluster

30

32

33

34

35

36

37

42

43

44

45

47

18

41

40

X

Rapid Upgrades With Pg-Migrator 10

Copy User Heap/Index Files

1

2

3 6

5

4 7

8

9

pg_class

User Tables

15 21 27

262014

13 19 25

241812

17 2311

10 16 22

System Tables

clog

Old Cluster

1

2

3 6

5

4 7

8

9

pg_class

User Tables

2640

39

31

System Tables

clog

New Cluster

30

32

33

34

35

36

37

42

43

44

45

47

18

41

Rapid Upgrades With Pg-Migrator 11

Complete

1

2

3 6

5

4 7

8

9

pg_class

User Tables

15 21 27

262014

13 19 25

241812

17 2311

10 16 22

System Tables

clog

Old Cluster

1

2

3 6

5

4 7

8

9

pg_class

User Tables

2640

39

31

System Tables

clog

New Cluster

30

32

33

34

35

36

37

42

43

44

45

47

18

41

Rapid Upgrades With Pg-Migrator 12

How It Works: In Detail� Check for cluster compatability

– locale– encoding– integer datetimes (default changed from 8.3 -> 8.4)

� Use pg_dumpall to dump old cluster schema (no data)

� Freeze all new cluster rows (remove reference to clog entries)

� Rename tablespaces to *_old (pre-9.0)

� New cluster uses old xid counter value (see freeze above)

– Set system table frozen xids to match the current xid

� Create new users/databasesRapid Upgrades With Pg-Migrator 13

� Collect cluster information

� Install support functions that call internal backend functions

� Create placeholder files to reserve relfilenode file names (pre-9.0)

� Create schema in new cluster

� Adjust new cluster to use reserved relfilenode names (pre-9.0)

– Delete placeholder toast relfilenode files– Remove new cluster toast tables– Create new cluster toast table using reserved relfilenode– Assign new toast tables with proper relfilenodes to relations

� Copy or link files from old cluster to new cluster

– Toast tables have the same relfilenodes as in the old cluster

� Warn about any remaining issues, like REINDEX requirements

Rapid Upgrades With Pg-Migrator 14

Sample Run: Performing Consistency Checks

Performing Consistency Checks

-----------------------------

Checking old data directory (/u/pgsql.old/data) ok

Checking new data directory (/u/pgsql/data) ok

Checking for columns with user-defined composite types ok

Checking for columns with user-defined array types ok

Checking for columns with user-defined enum types ok

Checking for /contrib/isn with bigint-passing mismatch ok

Checking for invalid ’name’ user columns ok

Checking for tsquery user columns ok

Creating script to adjust sequences ok

Creating catalog dump ok

Checking for presence of required libraries ok

| If pg_migrator fails after this point, you must

| re-initdb the new cluster before continuing.

| You will also need to remove the ".old" suffix

| from /u/pgsql.old/data/global/pg_control.old and old

| tablespace directory names.

Many of these checks are unnecessary when migrating to Postgres 9.0.Rapid Upgrades With Pg-Migrator 15

Sample Run: Performing Migration

Performing Migration

--------------------

Adding ".old" suffix to tablespace directory names ok

Adding ".old" suffix to old global/pg_control ok

Analyzing all rows in the new cluster ok

Freezing all rows on the new cluster ok

Deleting new commit clogs ok

Copying old commit clogs to new server ok

Setting next transaction id for new cluster ok

Resetting WAL archives ok

Setting frozenxid counters in new cluster ok

Creating databases in the new cluster ok

Creating placeholder relfiles for toast relations

/u/pgsql/data/base/16397/16586

Adding support functions to new cluster ok

Restoring database schema to new cluster ok

Restoring relations to use fixed toast file names

/u/pgsql/data/base/16397/16586

Removing support functions from new cluster ok

Restoring user relation files

/u/pgsql.old/data/base/11511/2613

Setting next oid for new cluster ok

Creating script to delete old cluster ok

Adjusting sequences okRapid Upgrades With Pg-Migrator 16

Sample Run: Completion

Upgrade complete

----------------

| Optimizer statistics and free space information

| are not transferred by pg_migrator so consider

| running:

| vacuumdb --all --analyze

| on the newly-upgraded cluster.

| Running this script will delete the old cluster’s data files:

| /u/postgres/pg_migrator_output/delete_old_cluster.sh

Rapid Upgrades With Pg-Migrator 17

Possible DataFormat Changes

Change Conversion Method

clog none

heap page header, including bitmask convert to new page format on read

tuple header, including bitmask convert to new page format on read

data value format create old data type in new cluster

index page format reindex, or recreate index methods

TOAST page format convert to new page format on read

Rapid Upgrades With Pg-Migrator 18

Migration Timings

Migration Method Minutesdump/restore 300.0dump with parallel restore 180.0pg_migrator in copy mode 44.0pg_migrator in link mode 0.7

Database size: 150GB, 850 tables

The last duration is 44 seconds.

Timings courtesy of

Stefan Kaltenbrunner

(mastermind on IRC)

Rapid Upgrades With Pg-Migrator 19

Conclusion

Rapid Upgrades With Pg-Migrator 20