Download as PDF, PPTX
Uploaded byPostgresOpen
Keith Paskett - Postgres on ZFS @ Postgres Open
The document discusses using ZFS as a storage solution for PostgreSQL databases. ZFS provides advantages like low-impact snapshots, cloning snapshots for testing and validation, replication of snapshots, and continuous integrity checking and automatic repair. It presents ZFS as a way to help protect against common "human" disasters with databases like dropped tables or deleted data. The document outlines the benefits of ZFS and provides examples of how to use ZFS snapshots for tasks like testing upgrades, rolling back changes, and accessing historical data.
Keith Paskett - Postgres on ZFS @ Postgres Open
- 1. SENSING THE PRESENT.SECURING THE FUTURE PostgreSQL on ZFS ‣ Replica(on, Backup, Human-‐disaster Recovery, and More... ‣ Keith Paske> ‣ Keith@Paske>.org ‣ @klpaske> Monday, September 23, 13
- 2. Agenda ‣ Introduc-on ‣ Defining the Problem (Opportunity) ‣ ZFS Advantages for your Postgres instance ‣ What ZFS is and where you can get it ‣ ZFS snapshots—where the magic happens ‣ Cloning for Valida-on, Tes-ng and Recovery ‣ Adding ZFS to Log/Streaming Replica-on ‣ Beyond the database ‣ Summary and Demo Postgres on ZFS -‐ Keith Paske5 2 Monday, September 23, 13
- 3. Introduc-on Postgres on ZFS -‐ Keith Paske5 3 Text Monday, September 23, 13
- 4. Postgres on ZFS -‐ Keith Paske5 4 Trouble lurks Where you least expect it Monday, September 23, 13
- 5. And comes in all sizes Postgres on ZFS -‐ Keith Paske5 5 Monday, September 23, 13
- 6. Backup & Recovery Disconnect ‣ Less likely disaster scenarios • Server failure • Mul(ple drive failures in a raid array • Data center fla>ened by (choose your disaster) ‣ Common 'human' disaster scenarios • Dropped table • Deleted data • Altered data ‣ Many backup solu-ons focus on the least likely disaster ‣ ZFS helps protect against more common disasters Postgres on ZFS -‐ Keith Paske5 6 Monday, September 23, 13
- 7. I Wish I Could... ‣ Test an upgrade script on a mul--‐terabyte database without having to set up a separate server with terabytes of storage ‣ Quickly roll back from an upgrade if things go badly ‣ Have point-‐in-‐-me access to a large database without having to do a restore then replay a week’s worth of transac-on logs Postgres on ZFS -‐ Keith Paske5 7 Monday, September 23, 13
- 8. ZFS Advantages for Databases ‣ Fast efficient replica-on (one-‐way periodic update) ‣ Low/no-‐impact snapshots ‣ Read/write access to snapshots via clones ‣ Pool physical devices ‣ Send/receive snapshots ‣ Bidirec-onal incremental send/receive ‣ Solid state cache drives ‣ Upgrade to larger physical drives with zero down-me ‣ Con-nuous integrity checking and automa-c repair ‣ Built in compression—can actually improve performance Postgres on ZFS -‐ Keith Paske5 8 Monday, September 23, 13
- 9. What ‣ ZFS ZeZabyte File System. ‣ Not just a file system. It’s a storage infrastructure. ‣ Transac-onal, Copy-‐on-‐write ‣ Always consistent on disk. ‣ Fully checksummed. ‣ Loves memory and SSD, and it knows how to use them. ‣ It’s easy! There are only two commands. Postgres on ZFS -‐ Keith Paske5 9 Monday, September 23, 13
- 10. Where ‣ IllumOS • OmniOS •OpenIndiana • SmartOS (No OS drive. Really.) ‣ FreeBSD ‣ Linux -‐ Yes it is produc-on ready ‣ OS X ‣ That other OS that helps fund the purchase of tropical islands Postgres on ZFS -‐ Keith Paske5 10 Monday, September 23, 13
- 11. ZFS -‐ It's all about the snapshots Postgres on ZFS -‐ Keith Paske5 11 •Snapshot1 •Snapshot2 •Snapshot3 •Snapshot4 •Snapshot5 •Snapshot6 •Snapshot7 FileSystem1 FileSystem2 •Snapshot1 •Snapshot2 • • •Snapshot5 •Snapshot6 •Snapshot7 FSclone3 FSclone4 FSclone1 FSclone2 Monday, September 23, 13
- 12. Send & Receive Postgres on ZFS -‐ Keith Paske5 12 •Snapshot1 DataDir1 DataDir2 •Snapshot1 •Snapshot3 DB Updates •Snapshot2 DB Updates •Snapshot2 •Snapshot3 Monday, September 23, 13
- 13. Accessing Snapshot Contents ‣.zfs directory • 'cd .zfs/snapshot/snapname' from base ZFS directory • 'ls -‐a' in base directory won't show the .zfs directory • Snapshot directories are read only • Tape backup of snapshot directory will be consistent ‣ Create a clone from a snapshot • zfs clone fsname@snapshotname clonename • Clones are writable • Only take addi-onal disk space as data is changed • Simply awesome for replica-ng a TB database in seconds Postgres on ZFS -‐ Keith Paske5 13 Monday, September 23, 13
- 14. Clone & Recover Postgres on ZFS -‐ Keith Paske5 14 •Snapshot1 •Snapshot2 •Snapshot3 DataDir1 DataDir1 clone pg_dump of table and data Drop Table Monday, September 23, 13
- 15. Snapshots & Transac-on Logs Postgres on ZFS -‐ Keith Paske5 15 421 3 xlog1 xlog2 ... xlog1 xlog2 ... xlog1 xlog2 ... xlog1 xlog2 ... Transaction log archive Monday, September 23, 13
- 16. Snapshot Frequency ‣ Tune to your needs ‣ My preferences for transac-onal databases • Every 10-‐20 minutes, keep for 9 hours • Daily, keep for 10 days • Weekly, keep for 8 weeks ‣ cronable scripts (examples in the VM) • snapshots.sh creates snapshots with hour, day, or week name embeded • snapshot-‐cleanup.sh removes snapshots older than a specified -me Postgres on ZFS -‐ Keith Paske5 16 Monday, September 23, 13
- 17. ZFS Compression ‣ Can improve performance especially for streaming data sets ‣ Turning on only compresses subsequent writes ‣ Check out test results from citusdata • hZp://citusdata.com/blog/64-‐zfs-‐compression ‣ Commands • zfs set compression=on fs/name (LZJB) • zfs set compression=gzip fs/name • zfs set compression=gzip-‐9 fs/name Postgres on ZFS -‐ Keith Paske5 17 Monday, September 23, 13
- 18. Beyond the Database ‣For Solaris and its deriva-ves • Replicate the OS drive • Simple bare-‐metal restore ‣ OS files with metadata in the DB • Snapshot both for a full consistent test/recovery environment ‣ Using Postgres FDW with non-‐DB files • Same story—snapshots, replica-on, and clones Postgres on ZFS -‐ Keith Paske5 18 Monday, September 23, 13
- 19. Demo Postgres on ZFS -‐ Keith Paske5 19 Monday, September 23, 13
- 20. Summary & Caveats ‣Great for quick duplica-on of very large databases ‣ Very efficient for periodic updates of replicas ‣ Combine snapshots with transac-on log archives for faster access to point-‐in-‐-me data ‣ Disk space is only freed when files and snapshots are deleted ‣ Have to shut down the ‘receiving’ DB during the update ‣ If your backup sooware finds the .zfs directory you will experience an--‐deduplica-on ‣ Recursive snapshots vs snapshots on separate file systems Postgres on ZFS -‐ Keith Paske5 20 Monday, September 23, 13
- 21. PostgreSQL and ZFS Postgres on ZFS -‐ Keith Paske5 21 They may not have been designed to work together, but they do nonetheless Monday, September 23, 13
- 22. Postgres on ZFS -‐ Keith Paske5 Go ahead. It doesn’t hurt 22 Monday, September 23, 13
- 23. VM Setup ‣ Before you import the VM into VirtualBox • VirtualBox 4.2.12 with VirtualBox Extension Pack • VT-‐x/AMD-‐V accelera-on (may need to enable in BIOS) • Enough memory to allocate 2GB to the VM • PuTTY or other terminal to access the VM's zones via ssh • Add a Host-‐Only Ethernet adapter with an IPv4 address of 192.168.68.1 and a netmask of 255.255.255.0 ‣ Import the OmniOS VM and start it up ‣ More detailed instruc-ons and the VM are at: hZp://sta-c4.usurf.usu.edu/vms ‣ You may get a message about modifying the network seqngs on import ‣ From your host’s shell or PuTTY connect to 192.168.68.10 ‣ log in as user: admin password: nimda Postgres on ZFS -‐ Keith Paske5 23 Monday, September 23, 13
- 24. VM Explora-on ‣ Root privileges • pfexec executes the following command with root privileges • prash assumes root status ‣ List, start, access and stop zones • zoneadm list -‐cv • pfexec bin/start-‐zones.sh • ssh to admin@192.168.68.(11,12, 13, or 14) password nimda • pfexec bin/stop-‐zones.sh ‣ List and explore zfs filesystems • zfs list • zfs list -‐t all • zfs list -‐t all -‐r rpool/space • zfs get all rpool/space Postgres on ZFS -‐ Keith Paske5 24 Monday, September 23, 13
- 25. ZFS Prac-ce ‣ Create a filesystem and add some content • zfs create rpool/myfs; cp -‐r Downloads /rpool/myfs/ ‣ Create a snapshot and replicate from it • zfs snapshot rpool/myfs@rep1 • zfs send rpool/myfs@rep1 | zfs recv rpool/myfs-‐copy@rep1 ‣ Add more content • cp PGrep* /rpool/myfs/ ‣ Explore .zfs directories • cd /rpool/myfs/.zfs/snapshot/rep1 ‣ Destroy the file systems • zfs destroy -‐r rpool/myfs • zfs destroy -‐r rpool/myfs-‐copy Postgres on ZFS -‐ Keith Paske5 25 Monday, September 23, 13
- 26. The OmniOS Virtual Machine Postgres on ZFS -‐ Keith Paske5 26 Global zone Zone2 File-based Slave • The /rpool/space/shared1 and /rpool/space/shared2 directories in the global zone are mounted as /shared1 and /shared2 in zones 1-4 • Each zone is its own isolated OS environment which can be rebooted independently. There is a postgres instance in each zone configured for replication with another zone. • Zones are connected via a host-only network using the 192.168.68.x subnet Zone1 File-based Master Zone3 Streaming Master Zone4 Streaming Slave Monday, September 23, 13
- 27. PostgreSQL Built-‐in Replica-on Demos ‣ Replica-on based on log-‐file shipping • bin/log-‐shipping-‐launch.sh starts replica-on from zone1 to zone2 • Add tables to PG in zone1 and veiw in zone2’s PG • Monitor server log on zone2 to see log file ingest less +F /var/postgres/server.log • bin/log-‐shipping-‐reset.sh to stop and reset ‣ Streaming replica-on • bin/stream-‐shipping-‐launch.sh starts replica-on from zone3 to zone4 • Add tables to PG in zone3 and veiw in zone4’s PG • bin/log-‐shipping-‐reset.sh to stop and reset Postgres on ZFS -‐ Keith Paske5 27 Monday, September 23, 13