Everything about backups (Firebird), by Dmitry Kuzmenko, CEO IBSurgeon

www.FirebirdDevelopersDay.com.br © 2009 – Dmitry Kuzmenko, IBSurgeon1

Everything about backups

Dmitry Kuzmenko

IBSurgeon


Backup methods

• gbak

• nbackup/online dump

• scripts

• stop server and copy

• no backups


Rescue plan

• what is a failure

• how many data will be lost

• how long company can wait


Failure

• Can happen at any time by any cause

• Database corruption– light – can be corrected by gfix,

IBFirstAid/FBFirstAid– hard – can be only restored from backup

• Anyway, it takes time. How much?– gfix or FBFirstAid – depends on the database

size, up to several hours– backup, gbak – several hours– nbackup – several minutes


Resque plan

• Choose appropriate backup type

• Check restore time, multiply it by 2, since always things can go wrong

• Try to understand how many data can be lost since last backup


GBAK in depth

• backup speed

• restore speed

• flaws

• goods

• usage

• tools


Backup speed

Ya FB 1.5 FB 2.0 IB 7.5 IB 2007

0:000:140:28

0:430:571:12

1:261:401:552:09

2:242:382:52

3:073:213:36

3:504:044:19

4:334:485:02

5:165:315:45

6:006:146:28

6:436:577:127:26

7:407:558:09

localhostlocal

servapi

se same hd


single processor (core)

FB 1.5

0:00

0:14

0:28

0:43

0:57

1:12

1:26

1:40

1:55

2:09

2:24

2:38

2:52

3:07

3:21

3:36

3:50

4:04

4:19

4:33

4:48

5:02

5:16

5:31

5:45

6:00

6:14

6:28

6:43

localhostlocal

servapi

se same hd


2 cores

FB 1.5 FB2.1 FB2.1cs IB 2007

0:00

0:36

1:12

1:48

2:24

3:00

3:36

4:12

4:48

5:24

6:00

6:36

7:12

7:48

8:24

9:00

9:36

10:12

10:48

11:24

12:00

12:36

13:12

13:48

14:24

15:00

15:36

16:12

16:48

17:24

18:00

localhostlocal

servapi

se same hd


• Firebird 1.5, Interbase - do not use local protocol where processors or cores are more than one– use localhost or Services API instead, it's 2

times faster

• Firebird 2.x local protocol does not have that problem


Restore speed

• database 3.8gb– backup – 4 minutes, 15 minutes worst case– restore - ~1 hour.


Non-restorable backups

• why?

• how to fix it?

• best script


Non-restorable backup

• broken primary keys– duplicate records

• broken foreign keys or lost records in the reference tables– need to delete referencing records or to create

records that were lost

• broken backup file– IBBackupSurgeon


Best backup/restore script

• Check metadata– gbak –b –m, gbak –c– compare with dbcompare

• backup, without garbage collection– gbak –b –g

• test restore

• always save metadata script– isql database.fdb –x >script.ddl


NBACKUP in depth

• backup speed

• restore speed

• flaws

• goods

• usage

• tools

• InterBase dump


NBACKUP

• LOCK

• Changes goes to delta

• Copying

• Applying delta to DB

DATABASE

Delta


backup speed

• each nbackup run scans (reads) the whole database – same as database file copy speed

• amount of writes depends on number of pages, changed from previous nbackup level


restore speed

• restore needs to– copy level 0 database (full file) to the original

place– read all specified level backup files and write

them to the database

• so, restore speed is equal to copying all specified levels of nbackup from one place to another


NBackup flaws

• backup level 0 is a database, but you can't use it as database, until you "fix" it

• if you "fix" database backup level 0, you can't add any levels (1-2-3) to it

• you can't update database with level backups• nbackup of any level always scans (reads) the whole

database• nbackup is a page copy. database damage is not

controlled• ! on Classic, nbackup can freeze until some process

commits changes to the database– this result longer time of nbackup than expected, and can cause

delta file live for hours. During this time you can't make any nbackups


NBackup is good

• It is fast, as copying file– 6 minutes for 13gb database

• you may organize levels for the needed rate

• restore is also fast


Common practice

• level 0 – each month– level 1 – each week

• level 2 – each day– level 3 – each 2 hours

• monthly - kept the whole year

• weekly – last 1-2 months

• daily – last 1-2 weeks

• hourly – 1 week

• usually archives are compressed– not good, because decompression takes time


NBackup sample 1

• 50 users, 10 hours per day, 68k transactions per day, accounting• hourly nbackup – 1-2 minutes, total nbackups – 40gb (year)

5700

220

120

25

0 2000 4000 6000

Size

level 3 hour

level 2 day

level 1 week

Database


NBackup sample 2

• 100 users, 12 hours per day, 224k transactions per day, stock• hourly nbackup – 1-2 minutes

10000

1000

800

300

0 5000 10000 15000

Size

level 3 hour

level 2 day

level 1 week

Database


InterBase dump

• extention of gbak (-d option)• dump always increments "level 0" backup• database dump is a "read-only" database,

ready to use• no dump "levels"• dumps can be made contigiously

– dump db to db_dump each hour– dump db_dump to db_dump2 each day– …


Scripts, replication, datapump

• alternate methods of copying data

• extracting scripts– check how long scripts will be "returned back to

the database"

• replication– may be the best, but needs additional hardware

and expenses for the tools, triggers, etc.

• IBPump, or your own tools


• Questions?

• www.ib-aid.com

• [email protected]

Technology

Everything about backups (Firebird), by Dmitry Kuzmenko, CEO IBSurgeon