Postgres on OpenStack

Postgres on OpenStack
• Dave Page | 18/9/2014
© 2014 EnterpriseDB Corporation. All rights reserved. 1

Introduction
• PostgreSQL:
− Core team member
− pgAdmin lead developer
− Web/sysadmin teams
− PGCAC/PGEU board member
• EDB:
− Tools
− Cloud products
− Configuration management
− Packaging/distribution
− Infrastructure

What is OpenStack?
• Cloud environment, similar to AWS:
− Private implementations
− Public implementations – Rackspace, HP Cloud
• Modular design, offering a choice of services including:
− Image Service (Glance)
− Block Storage (Cinder)
− Compute (Nova)
− Network (Neutron)
− Object Storage (Swift)

Glance Image Service
• Store virtual machine images and snapshots
• Preconfigure an image or snapshot with the desired
OS/software
• Share images with members of selected projects (AKA,
tenants).

Cinder Block Storage Service
• Analogous to Amazon Elastic Block Storage (EBS)
• Provides block storage devices to virtual machine
instances
• Can be provided from one or more service nodes via a
controller
• Service nodes may use various storage systems, with
different properties including:
− LVM
− GlusterFS
− NFS
− Various SANs

Nova Compute Service
• Analogous to Amazon Elastic Compute Service (EC2)
• Schedules and runs VM instances
• Multiple machine flavours with individual RAM, CPU
and disk requirements
• Compute node selection is highly configurable:
− Pluggable scheduler filters
− Target images to specific hosts
− Target flavours to specific hosts
− Target instances to hosts with enough free resources

Neutron Networking Service
• Provides network services to VM instances
• Attach instances to one or more virtual networks
• Map floating/public IP addresses to a VM
• Configure firewall rules for ingress and egress

Swift Object Storage
• Analogous to Amazon Simple Storage Service (S3)
• Provides highly redundant object storage
• Suitable for storage and retrieval of whole files

Wait, what about Trove?
• Still in relatively early stages of development
• Currently seems focused on MySQL/Cassandra
• No support for clustering or other non-trivial setups
• API favours psql/pgAdmin operations, not deployment/
configuration

Well that all sounds pretty cool,
but what can we do with it?

Database Server Images

Database server images
• Preconfigure a “standard” database image
• Enables consistency, which leads to ease of
management
• Preinstall commonly required software:
− Procedural languages
− WAL-E
− pgBouncer/pgPool
• Minimise launch/deployment time

Image or Snapshot?
• Technically there’s not much difference
• OCD: I like images to be, well images
• Images may be considered “blessed” or approved
• Snapshots are a convenient way to take backups,
perhaps of work in progress
• Images are “clean” configurations, whilst snapshots
may contain artifacts from previous use, e.g. shell
history or log files.

Creating an image
• Create a small root disk in QCOW2 format
• Install the base OS on the disk (I use KVM on RHEL)
• Install the required software
• Add the EPEL Yum repo, and install cloud-init
• Disable zeroconf for the network
• Shutdown the VM
• Run virt-sysprep to clean the OS installation

Database Server Machine
Configurations

Dedicated database flavours
• “Flavors” (without the u) describe the instance
configuration:
− vCPUs
− Memory
− Root disk
− Ephemeral disk
− Swap
− Other attributes
• Additional attributes allow us to target specific
hypervisor properties, e.g.
− Compute nodes with SSDs

Targeting a host aggregate
• Create a host aggregate or group containing the
desired compute nodes
• Add a property to the host aggregate to identify the
desired attribute for matching
• Create the desired flavour(s)
• Add a property to the flavour to correspond with the
property on the host aggregate
• Share the flavour with one or more tenants

Create host aggregate
$ nova aggregate-create pg-servers
+----+------------+-------------------+-------+----------+
| Id | Name | Availability Zone | Hosts | Metadata |
+----+------------+-------------------+-------+----------+
| 14 | pg-servers | - | | |
+----+------------+-------------------+-------+----------+
$ nova aggregate-add-host pg-servers nova5.ox.uk
Host nova5.ox.uk has been successfully added for aggregate 14
+----+------------+-------------------+---------------+----------+
+----+------------+-------------------+---------------+----------+
| 14 | pg-servers | - | 'nova5.ox.uk' | |
+----+------------+-------------------+---------------+----------+
$ nova aggregate-set-metadata pg-servers pgsvr=true
Metadata has been successfully updated for aggregate 14.
+----+------------+-----------------------------------+--------------+
+----+------------+-------------------+------- -------+--------------+
| 14 | pg-servers | - | 'nova5.ox.uk' | 'pgsvr=true' |
+----+------------+-------------------+---------------+--------------+

Create flavour
$ nova flavor-create --ephemeral 0 --swap 4096 --is-public true
pgsvr.small auto 4096 40 4
+--------------------------------------+-------------+-----------+...
| ID | Name | Memory_MB |...
+--------------------------------------+-------------+-----------+...
| 44221572-a7e0-4660-97fc-352467743fce | pgsvr.small | 4096 |...
+--------------------------------------+-------------+-----------+...
...------+-----------+------+-------+-------------+-----------+
... Disk | Ephemeral | Swap | VCPUs | RXTX_Factor | Is_Public |
...------+-----------+------+-------+-------------+-----------+
... 40 | 0 | 4096 | 4 | 1.0 | True |
...------+-----------+------+-------+-------------+-----------+

Create flavour
$ nova flavor-key pgsvr.small set aggregate_instance_extra_specs:pgsvr=true
$ nova flavor-show pgsvr.small
+----------------------------+--------------------------------------------------+
| Property | Value |
+----------------------------+--------------------------------------------------+
| OS-FLV-DISABLED:disabled | False |
| OS-FLV-EXT-DATA:ephemeral | 0 |
| disk | 40 |
| extra_specs | {"aggregate_instance_extra_specs:pgsvr": "true"} |
| id | 44221572-a7e0-4660-97fc-352467743fce |
| name | pgsvr.small |
| os-flavor-access:is_public | True |
| ram | 4096 |
| rxtx_factor | 1.0 |
| swap | 4096 |
| vcpus | 4 |
+----------------------------+--------------------------------------------------+

A note on Scheduler Filters
• Scheduler filters determine which compute node a new
instance will be created on
• Filters are chained, and can reject hosts based on:
− Capabilities
− Current utilisation
− Physical location
− Affinity
− Other factors such as tenant isolation requirements
• Our example used the
aggregate_instance_extra_specs filter
• Custom filters can be defined in JSON (json_filter), or
written from scratch in Python

Instance Storage
-or-
Where do my VMs live?

Local storage
• Known as Instance Store on AWS
• Images are copied to local storage on the compute
node and started
• Pros:
− Easy to setup
− Performant
− Minimal hardware requirements
• Cons:
− Moving instances requires downtime
− Potentially less reliable

Shared instance storage
• Same as Instance Store, but uses storage shared
between compute nodes using NFS, iSCSI, etc.
• Pros
− Allows easy migration of instances between compute nodes
− Potentially more reliable storage (if a SAN or similar is used)
• Cons
− “Local” I/O is over the network
− More complex to setup and maintain
− Can be far more expensive (SAN, 10GBe or fibre channel
network etc.)

Volume storage
• Known as Elastic Block Store (EBS) on AWS
• Root volume is stored in Cinder
• Pros:
− Available as a launch-time option, if Cinder is configured
− Allows easy migration of instances between compute nodes
− Potentially more reliable storage (if a SAN or similar is used)
• Cons:
− “Local” I/O is over the network
− More complex to setup and maintain
− Can be far more expensive (SAN, 10GBe or fibre channel
network etc.)

Data Storage

Data storage
• In the instance:
− Easy to setup and manage – click and go
− Resizing will pause the instance
− Not very “elastic”
− Slower snapshots – data has to be moved to Glance
• In Cinder:
− More complex to setup
− More expensive for a high performance setup
− Add storage on the fly by adding volumes and utilising LVM
− Can choose storage types – SSD vs. rust, low vs. high IOPs
− Fast snapshots – Cinder can use filesystem/SAN snapshots

Firewalling

Instance firewalling
• Neutron provides Security Groups to define firewall
rules
• No ingress access by default
• Each security group has one or more ingress or egress
rules defined
• All instances have one or more security groups
associated with them
• Allows easy configuration of a standard set of firewall
rules for multiple servers

Security group configuration

Networking

Private cluster network
• Create one or more virtual networks to contain a
cluster of database servers, entire application or some
other group of instances
• Analogous to Virtual Private Cloud (VPC) in Amazon
• Optionally attach virtual routers to networks
• Attach one or more networks to each instance
• Use Floating IPs (Elastic IPs) to access instances from
outside the virtual network
• Each network can have a DHCP server and custom
routes

Network configuration

Network topology

WAL Archiving

WAL archiving to Swift
• Object storage is ideal for storing WAL segments and
backups
• Highly redundant storage offers reliability and scaling
• Easy to use with WAL-E (https://github.com/wal-e/wal-e)
− WAL archival and retrieval
− Base backup creation and restoration
− WAL cleanup/pruning

Installing WAL-E
# pip install wal-e
<stuff happens>
# vi /var/lib/pgsql/9.3/data/postgresql.conf
wal_level = archive
archive_mode = on
archive_command = '/var/lib/pgsql/wal-e.sh %p’
archive_timeout = 60 # For testing
# vi /var/lib/pgsql/wal-e-creds.sh
export SWIFT_AUTHURL=http://auth.enterprisedb.com:35357/v2.0
export SWIFT_USER=wal-e
export SWIFT_PASSWORD=ReallyReallySecret
export SWIFT_TENANT=admin
export WALE_SWIFT_PREFIX=swift://wal-e
# chown postgres:postgres /var/lib/pgsql/wal-e-creds.sh
# chmod 600 /var/lib/pgsql/wal-e-creds.sh

Installing WAL-E
# vi /var/lib/pgsql/wal-e.sh
#!/bin/sh
source /var/lib/pgsql/wal-e-creds.sh
/usr/bin/wal-e wal-push $1
# chmod +x /var/lib/pgsql/wal-e-creds.sh
# service postgresql-9.3 restart
# tail –f /var/lib/pgsql/9.3/data/pg_log/postgresql.log
wal_e.main INFO MSG: starting WAL-E
DETAIL: The subcommand is "wal-push".
STRUCTURED: time=2014-08-29T16:55:56.236552-00 pid=18301
wal_e.worker.upload INFO MSG: begin archiving a file
DETAIL: Uploading "pg_xlog/00000001000000000000000E" to "swift://wal-e/
wal_005/00000001000000000000000E.lzo".
STRUCTURED: time=2014-08-29T16:55:56.360962-00 pid=18301 action=push-wal
key=swift://wal-e/wal_005/00000001000000000000000E.lzo prefix= seg=00000001000000000000000E
state=begin
urllib3.connectionpool INFO Starting new HTTP connection (1): auth.enterprisedb.com
urllib3.connectionpool INFO Starting new HTTP connection (1): swift.enterprisedb.com
wal_e.worker.upload INFO MSG: completed archiving to a file
DETAIL: Archiving to "swift://wal-e/wal_005/00000001000000000000000E.lzo" complete
at 9.29666KiB/s.
STRUCTURED: time=2014-08-29T16:56:03.467501-00 pid=18301 action=push-wal
key=swift://wal-e/wal_005/00000001000000000000000E.lzo prefix= rate=9.29666
seg=00000001000000000000000E state=complete

WAL Archives

Summary

Summary
• Deploy standardised, maintainable database servers
from images in Glance
• Configure database-optimised compute nodes
• Configure database-optimised machine flavours in
Nova to target desired nodes
• Select instance storage based on performance and
flexibility requirements vs. budget.
• Select data storage based on complexity vs. flexibility

Summary contd.
• Use Neutron’s security groups to provide standardised
security configuration outside of the instance
• Utilise virtual networks to isolate logical clusters or
groups of machines
• Archive base backups and WAL to Swift with WAL-E
for ease of backup management, redundancy and
scalability

Questions?

Postgres on OpenStack

Recommended

More Related Content

What's hot (20)

Similar to Postgres on OpenStack (20)

More from EDB (20)

Recently uploaded (20)

Postgres on OpenStack