SlideShare a Scribd company logo
Reliable Openstack – Designing
for Availability and Enterprise
Readiness
Ajay Gulati, Chaitanya BVK
ZeroStack, Inc.
Tokyo, October 27-30, 2015
1
Talk Outline
• High Availability: Definition and Types
• Platform HA
• Standard Approach
• ZeroStack Approach
• Demo
• VM HA
• Application HA
2
High Availability vs. Fail-Over
• High Availability
• Always on system
• Few seconds downtime in case of failures
• Auto healing with no manual intervention
• Fail Over
• Can tolerate fault temporarily
• Needs manual intervention for healing
3
Goal
• A web-scale private cloud
• Highly available
• Scale-on-demand
VM VM VM
V
M VM VM
V
M
V
M
V
M V
M
V
M
V
M
V
M
V
M
VM
V
M
VM
VM
4
Standard Approach: Stateless
• Special Controller Nodes
• Multiple instances with HAProxy with VIP
Nova Nova
HAProxy
VIP: 10.10.1.5
Compute/storage nodesControl cluster
5
Problem #1
• Controller nodes don’t scale automatically
• Can become bottleneck when compute scales
Nova Nova
HAProxy
VIP: 10.10.1.5
Compute/storage nodesControl cluster
6
Problem #2
• HAProxy becomes a single point of failure
• Every request has to go through an extra hop
Nova Nova
HAProxy
VIP: 10.10.1.5
Compute/storage nodesControl cluster
X
7
Problem #3
• If service or control node fails, no auto healing is done
• Manual intervention is needed to set up another node
Nova
Nov
a
HAProxy
VIP: 10.10.1.5
Compute/storage nodesControl cluster
X Nova
8
Standard Approach: Stateful
• Special database or AMQP nodes
• Multiple instances with replication or shared storage
• Active-active or Active-passive
MySql
VIP: 10.10.1.5
Replicated storage
Data replication (DRBD, …)
MySql
Shared reliable storage
9
MySql
MySql
Problem: #1
• Special nodes are needed
• Manual intervention is needed to heal the cluster
MySq
l
VIP: 10.10.1.5
Replicated storage
Data replication (DRBD, …)
X MySql
VIP: 10.10.1.5
10
Problem: #2
• Special nodes are needed
• Expensive shared storage silo needed
• Manual intervention is needed to heal the cluster
MySq
l
Shared reliable storage
X
MySql
VIP: 10.10.1.5
11
Standard Approach: Summary
• These methods don’t scale well
• Manual intervention needed, can be error prone
• Too many special nodes or silos
We need a more scalable approach to High
availability
12
Distributed Control Plane
• A distributed service that manages OpenStack services
• Can use any available node for fail-over
• Supports up to 97 failures in a 100 node cluster.
• Initiates auto-healing on failures (eg: fix under-replication, etc)
• Allows us to support more failures after healing
Glance MySqlRabbitMQ
KeystoneNeutron Heat
Distributed control plane
13
Cinder
Nova
Leader Election
• A fault-tolerant Leader Election algorithm picks one node as the Leader
• Multiple leader elections may be used for different responsibilities (eg: load balancing)
• Leader brings up OpenStack services on the cluster
• Monitors health of all services continuously
• Migrates services across nodes as necessary
Glance MySqlRabbitMQ
KeystoneNeutron Heat
continuous status checks
leader
14
Cinder
Nova
Service & Node Failures
• Leader detects node and service failures
• Leader computes a new service mapping
• Migrates services to match the new service map
• Initiate service specific healing if necessary
Cinder Glance
Nova
MySql
RabbitMQ
KeystoneNeutron
Heat
node failedleader
15
MySql
Heat
X
Leader Failures
• All nodes participate in leader election
• Non-leader nodes watch for leader’s health
• Initiate re-election when leader becomes unavailable
• Leader state (eg: service mapping, etc.) is stored in a Distributed WAL
• New leader restores the previous leader’s state from the Distributed WAL
Cinder
Glance
Nova
MySql
RabbitMQ
KeystoneNeutron
Heat
new leader
Xleader
16
Cinder
Nova
Detecting Node Failures
How to differentiate failed vs. disconnected node? You cannot!
• Leader issues service specific leases to each node
• An agent on each node stops service when lease is expired
• Leader refreshes the leases periodically
• Waits for previous lease to expire before starting a service
Glance
MySql
RabbitMQ
KeystoneNeutron
HeatCinder
Nova
leader
periodic lease refresh
17
Some Implementation Notes
• Use a distributed key-value store for fault-tolerance and Distributed WAL
• Zookeeper, etcd, custom raft based…
• Use service specific Virtual IPs so that host node doesn’t matter
• Expire ARP cache entries
• Use configurable timeouts to adjust responsive-ness
Glance
MySql
RabbitMQ
KeystoneNeutron
Heat
distributed key-value store
Cinder
Nova
18
Key Benefits
• No single point of failure
• High fault tolerance
• No special controller nodes
• Automatic healing
• No manual intervention required
19
Platform HA Summary
High
Availability
Fail-over
manual healing
Leader-based
self-healing
Distributed
control-plane
Reliable storage
Active-passive
with HAProxy
Replicated
storage
Not suited for Web-scale
Easier to do manually
Designed for Web-scale
Hard to debug manually
20
Demo: HA in Action
• Create a scale-out cloud
• Do some operations
• Kill a node with live services
• Do more operations
21
Users Admins
VM Level HA
Goal: Restart VM with same disks in case of failure
• Key problem: how do you identify a dead VM?
• Network disconnect?
• Not a reliable signal
• VM may still be doing IOs on its disks
• Storage IO disconnect is necessary
• Avoids data corruption in case of two VMs
22
VM Level HA: Solution
• Run agent on host for connectivity check
• Within cluster or to Internet
• In case of disconnection: kill VMs, stop IO access
• In case of host failure: do IPMI shutdown
• Restart the VMs on other hosts in priority order, specified by user
VM
V
M
V
M
V
M
V
M V
M
V
M
V
M
VM
VMVM VM
host-agent
X
X
23
VM
VM VM
Application Level HA
• Two kinds of failures
• Infrastructure failures (our focus here)
• Application bugs, hangs
24
Standard Approach
• Run application across multiple availability zones
• Run a load-balancer in front
25
VM VM
VM VM VM
VM VM VM
tier 1
tier 2
tier 3
AZ-east AZ-west
tier 1
tier 2
tier 3
LB
VM VM
VM VM VM
VM VM VM
Problem #1
• No locality within AZ
• Higher latency for inter-tier or inter-VM requests
26
VM
VM
VM
VM
VM
VM VM
VM
AZ-east
Rack 1 Rack N
Problem #2
• No failure tolerance against within AZ failure
• Rack, power or host failures can take down the app
27
VM
VM
VM
VM VM
VM
VM
VM
AZ-east
Rack 1 Rack N
X
X
Solution
• Control on placement
• Use affinity rules within VMs across tiers
• Use anti-affinity rules for VMs within a tier
28
VM VM
VM VM VM
VM VM VM
tier 1
tier 2
tier 3
AZ-east
Affinity within a group
Anti-affinity across
groups
Conclusions
• Current techniques are not sufficient for web-scale
infrastructure
• Key ideas for scalable, automated HA:
• No special nodes, symmetric design
• Automatic healing
• Distributed consensus needed for taking decisions
• VM level HA requires stronger failure detection & isolation
• Use application level HA for better performance and higher
reliability
29

More Related Content

What's hot

VMworld 2015: The Future of Network Virtualization with VMware NSX
VMworld 2015: The Future of Network Virtualization with VMware NSXVMworld 2015: The Future of Network Virtualization with VMware NSX
VMworld 2015: The Future of Network Virtualization with VMware NSX
VMworld
 
VMworld 2015: VMware NSX Deep Dive
VMworld 2015: VMware NSX Deep DiveVMworld 2015: VMware NSX Deep Dive
VMworld 2015: VMware NSX Deep Dive
VMworld
 
nsx overview with use cases 1.0
nsx overview with use cases 1.0nsx overview with use cases 1.0
nsx overview with use cases 1.0
Ploynatcha Akkaraputtipat
 
VMworld 2016: How to Deploy VMware NSX with Cisco Infrastructure
VMworld 2016: How to Deploy VMware NSX with Cisco InfrastructureVMworld 2016: How to Deploy VMware NSX with Cisco Infrastructure
VMworld 2016: How to Deploy VMware NSX with Cisco Infrastructure
VMworld
 
SDN in the Public Cloud: Windows Azure
SDN in the Public Cloud: Windows AzureSDN in the Public Cloud: Windows Azure
SDN in the Public Cloud: Windows Azure
Open Networking Summits
 
VMware 2015: Next Horizon for Cloud Networking and Security
VMware 2015: Next Horizon for Cloud Networking and SecurityVMware 2015: Next Horizon for Cloud Networking and Security
VMware 2015: Next Horizon for Cloud Networking and Security
VMworld
 
NSX: La Virtualizzazione di Rete e il Futuro della Sicurezza
NSX: La Virtualizzazione di Rete e il Futuro della SicurezzaNSX: La Virtualizzazione di Rete e il Futuro della Sicurezza
NSX: La Virtualizzazione di Rete e il Futuro della Sicurezza
VMUG IT
 
Windows Azure: Scaling SDN in the Public Cloud
Windows Azure: Scaling SDN in the Public CloudWindows Azure: Scaling SDN in the Public Cloud
Windows Azure: Scaling SDN in the Public Cloud
Open Networking Summits
 
Network Virtualization with VMware NSX
Network Virtualization with VMware NSXNetwork Virtualization with VMware NSX
Network Virtualization with VMware NSX
Scott Lowe
 
VMUG.IT Meeting Napoli - Opening
VMUG.IT Meeting Napoli - OpeningVMUG.IT Meeting Napoli - Opening
VMUG.IT Meeting Napoli - Opening
Andrea Mauro
 
VMworld 2013: VMware NSX Integration with OpenStack
VMworld 2013: VMware NSX Integration with OpenStack VMworld 2013: VMware NSX Integration with OpenStack
VMworld 2013: VMware NSX Integration with OpenStack
VMworld
 
VMworld 2013: Bringing Network Virtualization to VMware Environments with NSX
VMworld 2013: Bringing Network Virtualization to VMware Environments with NSX VMworld 2013: Bringing Network Virtualization to VMware Environments with NSX
VMworld 2013: Bringing Network Virtualization to VMware Environments with NSX
VMworld
 
VMworld 2013: Technical Deep Dive: Build a Collapsed DMZ Architecture for Opt...
VMworld 2013: Technical Deep Dive: Build a Collapsed DMZ Architecture for Opt...VMworld 2013: Technical Deep Dive: Build a Collapsed DMZ Architecture for Opt...
VMworld 2013: Technical Deep Dive: Build a Collapsed DMZ Architecture for Opt...
VMworld
 
NSX 9 Core Use Cases
NSX 9 Core Use CasesNSX 9 Core Use Cases
NSX 9 Core Use Cases
Kevin Groat
 
Does Hypervisor matter in OpenStack
Does Hypervisor matter in OpenStackDoes Hypervisor matter in OpenStack
Does Hypervisor matter in OpenStack
Nermina Miller
 
VMworld Europe 2014: Advanced Network Services with NSX
VMworld Europe 2014: Advanced Network Services with NSXVMworld Europe 2014: Advanced Network Services with NSX
VMworld Europe 2014: Advanced Network Services with NSX
VMworld
 
Cloud Computing with OpenStack
Cloud Computing with OpenStackCloud Computing with OpenStack
Cloud Computing with OpenStack
inside-BigData.com
 
CloudStackFinalProject
CloudStackFinalProjectCloudStackFinalProject
CloudStackFinalProject
Gustavo El Khoury
 
IaaS with Software Defined Networking
IaaS with Software Defined NetworkingIaaS with Software Defined Networking
IaaS with Software Defined Networking
Prasenjit Sarkar
 
VMworld 2013: Virtualized Network Services Model with VMware NSX
VMworld 2013: Virtualized Network Services Model with VMware NSX VMworld 2013: Virtualized Network Services Model with VMware NSX
VMworld 2013: Virtualized Network Services Model with VMware NSX
VMworld
 

What's hot (20)

VMworld 2015: The Future of Network Virtualization with VMware NSX
VMworld 2015: The Future of Network Virtualization with VMware NSXVMworld 2015: The Future of Network Virtualization with VMware NSX
VMworld 2015: The Future of Network Virtualization with VMware NSX
 
VMworld 2015: VMware NSX Deep Dive
VMworld 2015: VMware NSX Deep DiveVMworld 2015: VMware NSX Deep Dive
VMworld 2015: VMware NSX Deep Dive
 
nsx overview with use cases 1.0
nsx overview with use cases 1.0nsx overview with use cases 1.0
nsx overview with use cases 1.0
 
VMworld 2016: How to Deploy VMware NSX with Cisco Infrastructure
VMworld 2016: How to Deploy VMware NSX with Cisco InfrastructureVMworld 2016: How to Deploy VMware NSX with Cisco Infrastructure
VMworld 2016: How to Deploy VMware NSX with Cisco Infrastructure
 
SDN in the Public Cloud: Windows Azure
SDN in the Public Cloud: Windows AzureSDN in the Public Cloud: Windows Azure
SDN in the Public Cloud: Windows Azure
 
VMware 2015: Next Horizon for Cloud Networking and Security
VMware 2015: Next Horizon for Cloud Networking and SecurityVMware 2015: Next Horizon for Cloud Networking and Security
VMware 2015: Next Horizon for Cloud Networking and Security
 
NSX: La Virtualizzazione di Rete e il Futuro della Sicurezza
NSX: La Virtualizzazione di Rete e il Futuro della SicurezzaNSX: La Virtualizzazione di Rete e il Futuro della Sicurezza
NSX: La Virtualizzazione di Rete e il Futuro della Sicurezza
 
Windows Azure: Scaling SDN in the Public Cloud
Windows Azure: Scaling SDN in the Public CloudWindows Azure: Scaling SDN in the Public Cloud
Windows Azure: Scaling SDN in the Public Cloud
 
Network Virtualization with VMware NSX
Network Virtualization with VMware NSXNetwork Virtualization with VMware NSX
Network Virtualization with VMware NSX
 
VMUG.IT Meeting Napoli - Opening
VMUG.IT Meeting Napoli - OpeningVMUG.IT Meeting Napoli - Opening
VMUG.IT Meeting Napoli - Opening
 
VMworld 2013: VMware NSX Integration with OpenStack
VMworld 2013: VMware NSX Integration with OpenStack VMworld 2013: VMware NSX Integration with OpenStack
VMworld 2013: VMware NSX Integration with OpenStack
 
VMworld 2013: Bringing Network Virtualization to VMware Environments with NSX
VMworld 2013: Bringing Network Virtualization to VMware Environments with NSX VMworld 2013: Bringing Network Virtualization to VMware Environments with NSX
VMworld 2013: Bringing Network Virtualization to VMware Environments with NSX
 
VMworld 2013: Technical Deep Dive: Build a Collapsed DMZ Architecture for Opt...
VMworld 2013: Technical Deep Dive: Build a Collapsed DMZ Architecture for Opt...VMworld 2013: Technical Deep Dive: Build a Collapsed DMZ Architecture for Opt...
VMworld 2013: Technical Deep Dive: Build a Collapsed DMZ Architecture for Opt...
 
NSX 9 Core Use Cases
NSX 9 Core Use CasesNSX 9 Core Use Cases
NSX 9 Core Use Cases
 
Does Hypervisor matter in OpenStack
Does Hypervisor matter in OpenStackDoes Hypervisor matter in OpenStack
Does Hypervisor matter in OpenStack
 
VMworld Europe 2014: Advanced Network Services with NSX
VMworld Europe 2014: Advanced Network Services with NSXVMworld Europe 2014: Advanced Network Services with NSX
VMworld Europe 2014: Advanced Network Services with NSX
 
Cloud Computing with OpenStack
Cloud Computing with OpenStackCloud Computing with OpenStack
Cloud Computing with OpenStack
 
CloudStackFinalProject
CloudStackFinalProjectCloudStackFinalProject
CloudStackFinalProject
 
IaaS with Software Defined Networking
IaaS with Software Defined NetworkingIaaS with Software Defined Networking
IaaS with Software Defined Networking
 
VMworld 2013: Virtualized Network Services Model with VMware NSX
VMworld 2013: Virtualized Network Services Model with VMware NSX VMworld 2013: Virtualized Network Services Model with VMware NSX
VMworld 2013: Virtualized Network Services Model with VMware NSX
 

Viewers also liked

NIVELACIÓN GEOMÉTRICA DE CANALES
NIVELACIÓN GEOMÉTRICA DE CANALESNIVELACIÓN GEOMÉTRICA DE CANALES
NIVELACIÓN GEOMÉTRICA DE CANALES
JOHNNY JARA RAMOS
 
Finger wart removal
Finger wart removalFinger wart removal
Finger wart removal
danneeledge
 
week 7 (2)
week 7 (2)week 7 (2)
week 7 (2)
Brad Reid
 
Library report 2016
Library report 2016Library report 2016
Costume and Props
Costume and PropsCostume and Props
Costume and Props
LaurenClarke123
 
艺术品、展品航空运...PDF
艺术品、展品航空运...PDF艺术品、展品航空运...PDF
艺术品、展品航空运...PDFJoe Yang
 
Debdatta_Bakshi_Supply_Chain_Management_4yrs[1]
Debdatta_Bakshi_Supply_Chain_Management_4yrs[1]Debdatta_Bakshi_Supply_Chain_Management_4yrs[1]
Debdatta_Bakshi_Supply_Chain_Management_4yrs[1]
Debdatta Bakshi
 
Theoryofsupply
TheoryofsupplyTheoryofsupply
Theoryofsupply
Conferat Conferat
 
Education
EducationEducation
Latihan exel
Latihan exelLatihan exel
Latihan exel
sembiring969696
 
Cpl- A Practical Guide to Coming Home
Cpl- A Practical Guide to Coming HomeCpl- A Practical Guide to Coming Home
Cpl- A Practical Guide to Coming Home
Claire McCormack BBS HRM
 
Ppt proyecto
Ppt proyectoPpt proyecto
Ppt proyecto
IsaacsitoZ
 
[스타워즈1]시리즈순서
[스타워즈1]시리즈순서[스타워즈1]시리즈순서
[스타워즈1]시리즈순서
MinJin Yang
 
Didáctica crítica
Didáctica críticaDidáctica crítica
Didáctica crítica
ortacc
 
Young Marketers 4 - Application 4 - Eliter
Young Marketers 4 - Application 4 - EliterYoung Marketers 4 - Application 4 - Eliter
Young Marketers 4 - Application 4 - Eliter
kieuthienvan
 
Yulong garment catalog
Yulong garment catalogYulong garment catalog
Yulong garment catalog
lisa zhu
 
Vanessa ortega escobar
Vanessa ortega escobarVanessa ortega escobar
Vanessa ortega escobar
vane1793
 
Social Media
Social MediaSocial Media
Social Media
JAIDEEP CHATTERJEE
 
Codes and Conventions of Front Covers
Codes and Conventions of Front CoversCodes and Conventions of Front Covers
Codes and Conventions of Front Covers
EvijaKapeljuha
 
Macroeconomía
MacroeconomíaMacroeconomía
Macroeconomía
aemorocho
 

Viewers also liked (20)

NIVELACIÓN GEOMÉTRICA DE CANALES
NIVELACIÓN GEOMÉTRICA DE CANALESNIVELACIÓN GEOMÉTRICA DE CANALES
NIVELACIÓN GEOMÉTRICA DE CANALES
 
Finger wart removal
Finger wart removalFinger wart removal
Finger wart removal
 
week 7 (2)
week 7 (2)week 7 (2)
week 7 (2)
 
Library report 2016
Library report 2016Library report 2016
Library report 2016
 
Costume and Props
Costume and PropsCostume and Props
Costume and Props
 
艺术品、展品航空运...PDF
艺术品、展品航空运...PDF艺术品、展品航空运...PDF
艺术品、展品航空运...PDF
 
Debdatta_Bakshi_Supply_Chain_Management_4yrs[1]
Debdatta_Bakshi_Supply_Chain_Management_4yrs[1]Debdatta_Bakshi_Supply_Chain_Management_4yrs[1]
Debdatta_Bakshi_Supply_Chain_Management_4yrs[1]
 
Theoryofsupply
TheoryofsupplyTheoryofsupply
Theoryofsupply
 
Education
EducationEducation
Education
 
Latihan exel
Latihan exelLatihan exel
Latihan exel
 
Cpl- A Practical Guide to Coming Home
Cpl- A Practical Guide to Coming HomeCpl- A Practical Guide to Coming Home
Cpl- A Practical Guide to Coming Home
 
Ppt proyecto
Ppt proyectoPpt proyecto
Ppt proyecto
 
[스타워즈1]시리즈순서
[스타워즈1]시리즈순서[스타워즈1]시리즈순서
[스타워즈1]시리즈순서
 
Didáctica crítica
Didáctica críticaDidáctica crítica
Didáctica crítica
 
Young Marketers 4 - Application 4 - Eliter
Young Marketers 4 - Application 4 - EliterYoung Marketers 4 - Application 4 - Eliter
Young Marketers 4 - Application 4 - Eliter
 
Yulong garment catalog
Yulong garment catalogYulong garment catalog
Yulong garment catalog
 
Vanessa ortega escobar
Vanessa ortega escobarVanessa ortega escobar
Vanessa ortega escobar
 
Social Media
Social MediaSocial Media
Social Media
 
Codes and Conventions of Front Covers
Codes and Conventions of Front CoversCodes and Conventions of Front Covers
Codes and Conventions of Front Covers
 
Macroeconomía
MacroeconomíaMacroeconomía
Macroeconomía
 

Similar to Zerostack reliable openstack

Ame 2269 ibm mq high availability
Ame 2269 ibm mq high availabilityAme 2269 ibm mq high availability
Ame 2269 ibm mq high availability
Andrew Schofield
 
Why Nutanix for Enterprise Workloads
Why Nutanix for Enterprise WorkloadsWhy Nutanix for Enterprise Workloads
Why Nutanix for Enterprise Workloads
NEXTtour
 
Hyun goo oVirt study - Presentation
Hyun goo oVirt study - PresentationHyun goo oVirt study - Presentation
Hyun goo oVirt study - Presentation
Johnny Hyun Goo
 
IBM MQ High Availabillity and Disaster Recovery (2017 version)
IBM MQ High Availabillity and Disaster Recovery (2017 version)IBM MQ High Availabillity and Disaster Recovery (2017 version)
IBM MQ High Availabillity and Disaster Recovery (2017 version)
MarkTaylorIBM
 
VMworld 2015: Extreme Performance Series - vSphere Compute & Memory
VMworld 2015: Extreme Performance Series - vSphere Compute & MemoryVMworld 2015: Extreme Performance Series - vSphere Compute & Memory
VMworld 2015: Extreme Performance Series - vSphere Compute & Memory
VMworld
 
Does Hypervisor Matter in OpenStack?
Does Hypervisor Matter in OpenStack?Does Hypervisor Matter in OpenStack?
Does Hypervisor Matter in OpenStack?
Mirantis
 
Network and Service Virtualization tutorial at ONUG Spring 2015
Network and Service Virtualization tutorial at ONUG Spring 2015Network and Service Virtualization tutorial at ONUG Spring 2015
Network and Service Virtualization tutorial at ONUG Spring 2015
SDN Hub
 
Neutron scaling
Neutron scalingNeutron scaling
Neutron scaling
Vinay Bannai
 
CloudStack Overview
CloudStack OverviewCloudStack Overview
CloudStack Overview
sedukull
 
Simplifying Hyper-V Management for VMware Administrators
Simplifying Hyper-V Management for VMware AdministratorsSimplifying Hyper-V Management for VMware Administrators
Simplifying Hyper-V Management for VMware Administrators
5nine
 
Introduction to failover clustering with sql server
Introduction to failover clustering with sql serverIntroduction to failover clustering with sql server
Introduction to failover clustering with sql server
Eduardo Castro
 
VMworld 2013: Three Advantages of Running Cloud Foundry in a VMware Private C...
VMworld 2013: Three Advantages of Running Cloud Foundry in a VMware Private C...VMworld 2013: Three Advantages of Running Cloud Foundry in a VMware Private C...
VMworld 2013: Three Advantages of Running Cloud Foundry in a VMware Private C...
VMworld
 
Breaking Down the Barriers through Virtualization - Frank Feldman, Red Hat
Breaking Down the Barriers through Virtualization - Frank Feldman, Red HatBreaking Down the Barriers through Virtualization - Frank Feldman, Red Hat
Breaking Down the Barriers through Virtualization - Frank Feldman, Red Hat
Linuxmalaysia Malaysia
 
Cocktail of Environments. How to Mix Test and Development Environments and St...
Cocktail of Environments. How to Mix Test and Development Environments and St...Cocktail of Environments. How to Mix Test and Development Environments and St...
Cocktail of Environments. How to Mix Test and Development Environments and St...
Aleksandr Tarasov
 
Module 1b - Hyper-v Configuration.pptx
Module 1b - Hyper-v Configuration.pptxModule 1b - Hyper-v Configuration.pptx
Module 1b - Hyper-v Configuration.pptx
JoshLagan
 
Presentation drs advanced concepts, best practices and future directions
Presentation   drs advanced concepts, best practices and future directionsPresentation   drs advanced concepts, best practices and future directions
Presentation drs advanced concepts, best practices and future directions
solarisyourep
 
XPDDS18: Performance tuning on Xen platform - Bo Zhang & Yifei Jiang, Huawei
XPDDS18: Performance tuning on Xen platform - Bo Zhang & Yifei Jiang, HuaweiXPDDS18: Performance tuning on Xen platform - Bo Zhang & Yifei Jiang, Huawei
XPDDS18: Performance tuning on Xen platform - Bo Zhang & Yifei Jiang, Huawei
The Linux Foundation
 
17-virtualization.pptx
17-virtualization.pptx17-virtualization.pptx
17-virtualization.pptx
KowsalyaJayakumar2
 
VMworld 2014: vSphere Distributed Switch
VMworld 2014: vSphere Distributed SwitchVMworld 2014: vSphere Distributed Switch
VMworld 2014: vSphere Distributed Switch
VMworld
 
Server virtualization 2
Server virtualization   2Server virtualization   2
Server virtualization 2
Eqhball Ghazizadeh
 

Similar to Zerostack reliable openstack (20)

Ame 2269 ibm mq high availability
Ame 2269 ibm mq high availabilityAme 2269 ibm mq high availability
Ame 2269 ibm mq high availability
 
Why Nutanix for Enterprise Workloads
Why Nutanix for Enterprise WorkloadsWhy Nutanix for Enterprise Workloads
Why Nutanix for Enterprise Workloads
 
Hyun goo oVirt study - Presentation
Hyun goo oVirt study - PresentationHyun goo oVirt study - Presentation
Hyun goo oVirt study - Presentation
 
IBM MQ High Availabillity and Disaster Recovery (2017 version)
IBM MQ High Availabillity and Disaster Recovery (2017 version)IBM MQ High Availabillity and Disaster Recovery (2017 version)
IBM MQ High Availabillity and Disaster Recovery (2017 version)
 
VMworld 2015: Extreme Performance Series - vSphere Compute & Memory
VMworld 2015: Extreme Performance Series - vSphere Compute & MemoryVMworld 2015: Extreme Performance Series - vSphere Compute & Memory
VMworld 2015: Extreme Performance Series - vSphere Compute & Memory
 
Does Hypervisor Matter in OpenStack?
Does Hypervisor Matter in OpenStack?Does Hypervisor Matter in OpenStack?
Does Hypervisor Matter in OpenStack?
 
Network and Service Virtualization tutorial at ONUG Spring 2015
Network and Service Virtualization tutorial at ONUG Spring 2015Network and Service Virtualization tutorial at ONUG Spring 2015
Network and Service Virtualization tutorial at ONUG Spring 2015
 
Neutron scaling
Neutron scalingNeutron scaling
Neutron scaling
 
CloudStack Overview
CloudStack OverviewCloudStack Overview
CloudStack Overview
 
Simplifying Hyper-V Management for VMware Administrators
Simplifying Hyper-V Management for VMware AdministratorsSimplifying Hyper-V Management for VMware Administrators
Simplifying Hyper-V Management for VMware Administrators
 
Introduction to failover clustering with sql server
Introduction to failover clustering with sql serverIntroduction to failover clustering with sql server
Introduction to failover clustering with sql server
 
VMworld 2013: Three Advantages of Running Cloud Foundry in a VMware Private C...
VMworld 2013: Three Advantages of Running Cloud Foundry in a VMware Private C...VMworld 2013: Three Advantages of Running Cloud Foundry in a VMware Private C...
VMworld 2013: Three Advantages of Running Cloud Foundry in a VMware Private C...
 
Breaking Down the Barriers through Virtualization - Frank Feldman, Red Hat
Breaking Down the Barriers through Virtualization - Frank Feldman, Red HatBreaking Down the Barriers through Virtualization - Frank Feldman, Red Hat
Breaking Down the Barriers through Virtualization - Frank Feldman, Red Hat
 
Cocktail of Environments. How to Mix Test and Development Environments and St...
Cocktail of Environments. How to Mix Test and Development Environments and St...Cocktail of Environments. How to Mix Test and Development Environments and St...
Cocktail of Environments. How to Mix Test and Development Environments and St...
 
Module 1b - Hyper-v Configuration.pptx
Module 1b - Hyper-v Configuration.pptxModule 1b - Hyper-v Configuration.pptx
Module 1b - Hyper-v Configuration.pptx
 
Presentation drs advanced concepts, best practices and future directions
Presentation   drs advanced concepts, best practices and future directionsPresentation   drs advanced concepts, best practices and future directions
Presentation drs advanced concepts, best practices and future directions
 
XPDDS18: Performance tuning on Xen platform - Bo Zhang & Yifei Jiang, Huawei
XPDDS18: Performance tuning on Xen platform - Bo Zhang & Yifei Jiang, HuaweiXPDDS18: Performance tuning on Xen platform - Bo Zhang & Yifei Jiang, Huawei
XPDDS18: Performance tuning on Xen platform - Bo Zhang & Yifei Jiang, Huawei
 
17-virtualization.pptx
17-virtualization.pptx17-virtualization.pptx
17-virtualization.pptx
 
VMworld 2014: vSphere Distributed Switch
VMworld 2014: vSphere Distributed SwitchVMworld 2014: vSphere Distributed Switch
VMworld 2014: vSphere Distributed Switch
 
Server virtualization 2
Server virtualization   2Server virtualization   2
Server virtualization 2
 

Recently uploaded

HCL Notes and Domino License Cost Reduction in the World of DLAU
HCL Notes and Domino License Cost Reduction in the World of DLAUHCL Notes and Domino License Cost Reduction in the World of DLAU
HCL Notes and Domino License Cost Reduction in the World of DLAU
panagenda
 
Essentials of Automations: Exploring Attributes & Automation Parameters
Essentials of Automations: Exploring Attributes & Automation ParametersEssentials of Automations: Exploring Attributes & Automation Parameters
Essentials of Automations: Exploring Attributes & Automation Parameters
Safe Software
 
Columbus Data & Analytics Wednesdays - June 2024
Columbus Data & Analytics Wednesdays - June 2024Columbus Data & Analytics Wednesdays - June 2024
Columbus Data & Analytics Wednesdays - June 2024
Jason Packer
 
Apps Break Data
Apps Break DataApps Break Data
Apps Break Data
Ivo Velitchkov
 
Taking AI to the Next Level in Manufacturing.pdf
Taking AI to the Next Level in Manufacturing.pdfTaking AI to the Next Level in Manufacturing.pdf
Taking AI to the Next Level in Manufacturing.pdf
ssuserfac0301
 
GraphRAG for LifeSciences Hands-On with the Clinical Knowledge Graph
GraphRAG for LifeSciences Hands-On with the Clinical Knowledge GraphGraphRAG for LifeSciences Hands-On with the Clinical Knowledge Graph
GraphRAG for LifeSciences Hands-On with the Clinical Knowledge Graph
Neo4j
 
Programming Foundation Models with DSPy - Meetup Slides
Programming Foundation Models with DSPy - Meetup SlidesProgramming Foundation Models with DSPy - Meetup Slides
Programming Foundation Models with DSPy - Meetup Slides
Zilliz
 
HCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAU
HCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAUHCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAU
HCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAU
panagenda
 
Introduction of Cybersecurity with OSS at Code Europe 2024
Introduction of Cybersecurity with OSS  at Code Europe 2024Introduction of Cybersecurity with OSS  at Code Europe 2024
Introduction of Cybersecurity with OSS at Code Europe 2024
Hiroshi SHIBATA
 
Skybuffer SAM4U tool for SAP license adoption
Skybuffer SAM4U tool for SAP license adoptionSkybuffer SAM4U tool for SAP license adoption
Skybuffer SAM4U tool for SAP license adoption
Tatiana Kojar
 
Mutation Testing for Task-Oriented Chatbots
Mutation Testing for Task-Oriented ChatbotsMutation Testing for Task-Oriented Chatbots
Mutation Testing for Task-Oriented Chatbots
Pablo Gómez Abajo
 
Generating privacy-protected synthetic data using Secludy and Milvus
Generating privacy-protected synthetic data using Secludy and MilvusGenerating privacy-protected synthetic data using Secludy and Milvus
Generating privacy-protected synthetic data using Secludy and Milvus
Zilliz
 
Deep Dive: AI-Powered Marketing to Get More Leads and Customers with HyperGro...
Deep Dive: AI-Powered Marketing to Get More Leads and Customers with HyperGro...Deep Dive: AI-Powered Marketing to Get More Leads and Customers with HyperGro...
Deep Dive: AI-Powered Marketing to Get More Leads and Customers with HyperGro...
saastr
 
Freshworks Rethinks NoSQL for Rapid Scaling & Cost-Efficiency
Freshworks Rethinks NoSQL for Rapid Scaling & Cost-EfficiencyFreshworks Rethinks NoSQL for Rapid Scaling & Cost-Efficiency
Freshworks Rethinks NoSQL for Rapid Scaling & Cost-Efficiency
ScyllaDB
 
Choosing The Best AWS Service For Your Website + API.pptx
Choosing The Best AWS Service For Your Website + API.pptxChoosing The Best AWS Service For Your Website + API.pptx
Choosing The Best AWS Service For Your Website + API.pptx
Brandon Minnick, MBA
 
9 CEO's who hit $100m ARR Share Their Top Growth Tactics Nathan Latka, Founde...
9 CEO's who hit $100m ARR Share Their Top Growth Tactics Nathan Latka, Founde...9 CEO's who hit $100m ARR Share Their Top Growth Tactics Nathan Latka, Founde...
9 CEO's who hit $100m ARR Share Their Top Growth Tactics Nathan Latka, Founde...
saastr
 
Main news related to the CCS TSI 2023 (2023/1695)
Main news related to the CCS TSI 2023 (2023/1695)Main news related to the CCS TSI 2023 (2023/1695)
Main news related to the CCS TSI 2023 (2023/1695)
Jakub Marek
 
How to Interpret Trends in the Kalyan Rajdhani Mix Chart.pdf
How to Interpret Trends in the Kalyan Rajdhani Mix Chart.pdfHow to Interpret Trends in the Kalyan Rajdhani Mix Chart.pdf
How to Interpret Trends in the Kalyan Rajdhani Mix Chart.pdf
Chart Kalyan
 
"Frontline Battles with DDoS: Best practices and Lessons Learned", Igor Ivaniuk
"Frontline Battles with DDoS: Best practices and Lessons Learned",  Igor Ivaniuk"Frontline Battles with DDoS: Best practices and Lessons Learned",  Igor Ivaniuk
"Frontline Battles with DDoS: Best practices and Lessons Learned", Igor Ivaniuk
Fwdays
 
Principle of conventional tomography-Bibash Shahi ppt..pptx
Principle of conventional tomography-Bibash Shahi ppt..pptxPrinciple of conventional tomography-Bibash Shahi ppt..pptx
Principle of conventional tomography-Bibash Shahi ppt..pptx
BibashShahi
 

Recently uploaded (20)

HCL Notes and Domino License Cost Reduction in the World of DLAU
HCL Notes and Domino License Cost Reduction in the World of DLAUHCL Notes and Domino License Cost Reduction in the World of DLAU
HCL Notes and Domino License Cost Reduction in the World of DLAU
 
Essentials of Automations: Exploring Attributes & Automation Parameters
Essentials of Automations: Exploring Attributes & Automation ParametersEssentials of Automations: Exploring Attributes & Automation Parameters
Essentials of Automations: Exploring Attributes & Automation Parameters
 
Columbus Data & Analytics Wednesdays - June 2024
Columbus Data & Analytics Wednesdays - June 2024Columbus Data & Analytics Wednesdays - June 2024
Columbus Data & Analytics Wednesdays - June 2024
 
Apps Break Data
Apps Break DataApps Break Data
Apps Break Data
 
Taking AI to the Next Level in Manufacturing.pdf
Taking AI to the Next Level in Manufacturing.pdfTaking AI to the Next Level in Manufacturing.pdf
Taking AI to the Next Level in Manufacturing.pdf
 
GraphRAG for LifeSciences Hands-On with the Clinical Knowledge Graph
GraphRAG for LifeSciences Hands-On with the Clinical Knowledge GraphGraphRAG for LifeSciences Hands-On with the Clinical Knowledge Graph
GraphRAG for LifeSciences Hands-On with the Clinical Knowledge Graph
 
Programming Foundation Models with DSPy - Meetup Slides
Programming Foundation Models with DSPy - Meetup SlidesProgramming Foundation Models with DSPy - Meetup Slides
Programming Foundation Models with DSPy - Meetup Slides
 
HCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAU
HCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAUHCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAU
HCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAU
 
Introduction of Cybersecurity with OSS at Code Europe 2024
Introduction of Cybersecurity with OSS  at Code Europe 2024Introduction of Cybersecurity with OSS  at Code Europe 2024
Introduction of Cybersecurity with OSS at Code Europe 2024
 
Skybuffer SAM4U tool for SAP license adoption
Skybuffer SAM4U tool for SAP license adoptionSkybuffer SAM4U tool for SAP license adoption
Skybuffer SAM4U tool for SAP license adoption
 
Mutation Testing for Task-Oriented Chatbots
Mutation Testing for Task-Oriented ChatbotsMutation Testing for Task-Oriented Chatbots
Mutation Testing for Task-Oriented Chatbots
 
Generating privacy-protected synthetic data using Secludy and Milvus
Generating privacy-protected synthetic data using Secludy and MilvusGenerating privacy-protected synthetic data using Secludy and Milvus
Generating privacy-protected synthetic data using Secludy and Milvus
 
Deep Dive: AI-Powered Marketing to Get More Leads and Customers with HyperGro...
Deep Dive: AI-Powered Marketing to Get More Leads and Customers with HyperGro...Deep Dive: AI-Powered Marketing to Get More Leads and Customers with HyperGro...
Deep Dive: AI-Powered Marketing to Get More Leads and Customers with HyperGro...
 
Freshworks Rethinks NoSQL for Rapid Scaling & Cost-Efficiency
Freshworks Rethinks NoSQL for Rapid Scaling & Cost-EfficiencyFreshworks Rethinks NoSQL for Rapid Scaling & Cost-Efficiency
Freshworks Rethinks NoSQL for Rapid Scaling & Cost-Efficiency
 
Choosing The Best AWS Service For Your Website + API.pptx
Choosing The Best AWS Service For Your Website + API.pptxChoosing The Best AWS Service For Your Website + API.pptx
Choosing The Best AWS Service For Your Website + API.pptx
 
9 CEO's who hit $100m ARR Share Their Top Growth Tactics Nathan Latka, Founde...
9 CEO's who hit $100m ARR Share Their Top Growth Tactics Nathan Latka, Founde...9 CEO's who hit $100m ARR Share Their Top Growth Tactics Nathan Latka, Founde...
9 CEO's who hit $100m ARR Share Their Top Growth Tactics Nathan Latka, Founde...
 
Main news related to the CCS TSI 2023 (2023/1695)
Main news related to the CCS TSI 2023 (2023/1695)Main news related to the CCS TSI 2023 (2023/1695)
Main news related to the CCS TSI 2023 (2023/1695)
 
How to Interpret Trends in the Kalyan Rajdhani Mix Chart.pdf
How to Interpret Trends in the Kalyan Rajdhani Mix Chart.pdfHow to Interpret Trends in the Kalyan Rajdhani Mix Chart.pdf
How to Interpret Trends in the Kalyan Rajdhani Mix Chart.pdf
 
"Frontline Battles with DDoS: Best practices and Lessons Learned", Igor Ivaniuk
"Frontline Battles with DDoS: Best practices and Lessons Learned",  Igor Ivaniuk"Frontline Battles with DDoS: Best practices and Lessons Learned",  Igor Ivaniuk
"Frontline Battles with DDoS: Best practices and Lessons Learned", Igor Ivaniuk
 
Principle of conventional tomography-Bibash Shahi ppt..pptx
Principle of conventional tomography-Bibash Shahi ppt..pptxPrinciple of conventional tomography-Bibash Shahi ppt..pptx
Principle of conventional tomography-Bibash Shahi ppt..pptx
 

Zerostack reliable openstack

  • 1. Reliable Openstack – Designing for Availability and Enterprise Readiness Ajay Gulati, Chaitanya BVK ZeroStack, Inc. Tokyo, October 27-30, 2015 1
  • 2. Talk Outline • High Availability: Definition and Types • Platform HA • Standard Approach • ZeroStack Approach • Demo • VM HA • Application HA 2
  • 3. High Availability vs. Fail-Over • High Availability • Always on system • Few seconds downtime in case of failures • Auto healing with no manual intervention • Fail Over • Can tolerate fault temporarily • Needs manual intervention for healing 3
  • 4. Goal • A web-scale private cloud • Highly available • Scale-on-demand VM VM VM V M VM VM V M V M V M V M V M V M V M V M VM V M VM VM 4
  • 5. Standard Approach: Stateless • Special Controller Nodes • Multiple instances with HAProxy with VIP Nova Nova HAProxy VIP: 10.10.1.5 Compute/storage nodesControl cluster 5
  • 6. Problem #1 • Controller nodes don’t scale automatically • Can become bottleneck when compute scales Nova Nova HAProxy VIP: 10.10.1.5 Compute/storage nodesControl cluster 6
  • 7. Problem #2 • HAProxy becomes a single point of failure • Every request has to go through an extra hop Nova Nova HAProxy VIP: 10.10.1.5 Compute/storage nodesControl cluster X 7
  • 8. Problem #3 • If service or control node fails, no auto healing is done • Manual intervention is needed to set up another node Nova Nov a HAProxy VIP: 10.10.1.5 Compute/storage nodesControl cluster X Nova 8
  • 9. Standard Approach: Stateful • Special database or AMQP nodes • Multiple instances with replication or shared storage • Active-active or Active-passive MySql VIP: 10.10.1.5 Replicated storage Data replication (DRBD, …) MySql Shared reliable storage 9 MySql MySql
  • 10. Problem: #1 • Special nodes are needed • Manual intervention is needed to heal the cluster MySq l VIP: 10.10.1.5 Replicated storage Data replication (DRBD, …) X MySql VIP: 10.10.1.5 10
  • 11. Problem: #2 • Special nodes are needed • Expensive shared storage silo needed • Manual intervention is needed to heal the cluster MySq l Shared reliable storage X MySql VIP: 10.10.1.5 11
  • 12. Standard Approach: Summary • These methods don’t scale well • Manual intervention needed, can be error prone • Too many special nodes or silos We need a more scalable approach to High availability 12
  • 13. Distributed Control Plane • A distributed service that manages OpenStack services • Can use any available node for fail-over • Supports up to 97 failures in a 100 node cluster. • Initiates auto-healing on failures (eg: fix under-replication, etc) • Allows us to support more failures after healing Glance MySqlRabbitMQ KeystoneNeutron Heat Distributed control plane 13 Cinder Nova
  • 14. Leader Election • A fault-tolerant Leader Election algorithm picks one node as the Leader • Multiple leader elections may be used for different responsibilities (eg: load balancing) • Leader brings up OpenStack services on the cluster • Monitors health of all services continuously • Migrates services across nodes as necessary Glance MySqlRabbitMQ KeystoneNeutron Heat continuous status checks leader 14 Cinder Nova
  • 15. Service & Node Failures • Leader detects node and service failures • Leader computes a new service mapping • Migrates services to match the new service map • Initiate service specific healing if necessary Cinder Glance Nova MySql RabbitMQ KeystoneNeutron Heat node failedleader 15 MySql Heat X
  • 16. Leader Failures • All nodes participate in leader election • Non-leader nodes watch for leader’s health • Initiate re-election when leader becomes unavailable • Leader state (eg: service mapping, etc.) is stored in a Distributed WAL • New leader restores the previous leader’s state from the Distributed WAL Cinder Glance Nova MySql RabbitMQ KeystoneNeutron Heat new leader Xleader 16 Cinder Nova
  • 17. Detecting Node Failures How to differentiate failed vs. disconnected node? You cannot! • Leader issues service specific leases to each node • An agent on each node stops service when lease is expired • Leader refreshes the leases periodically • Waits for previous lease to expire before starting a service Glance MySql RabbitMQ KeystoneNeutron HeatCinder Nova leader periodic lease refresh 17
  • 18. Some Implementation Notes • Use a distributed key-value store for fault-tolerance and Distributed WAL • Zookeeper, etcd, custom raft based… • Use service specific Virtual IPs so that host node doesn’t matter • Expire ARP cache entries • Use configurable timeouts to adjust responsive-ness Glance MySql RabbitMQ KeystoneNeutron Heat distributed key-value store Cinder Nova 18
  • 19. Key Benefits • No single point of failure • High fault tolerance • No special controller nodes • Automatic healing • No manual intervention required 19
  • 20. Platform HA Summary High Availability Fail-over manual healing Leader-based self-healing Distributed control-plane Reliable storage Active-passive with HAProxy Replicated storage Not suited for Web-scale Easier to do manually Designed for Web-scale Hard to debug manually 20
  • 21. Demo: HA in Action • Create a scale-out cloud • Do some operations • Kill a node with live services • Do more operations 21 Users Admins
  • 22. VM Level HA Goal: Restart VM with same disks in case of failure • Key problem: how do you identify a dead VM? • Network disconnect? • Not a reliable signal • VM may still be doing IOs on its disks • Storage IO disconnect is necessary • Avoids data corruption in case of two VMs 22
  • 23. VM Level HA: Solution • Run agent on host for connectivity check • Within cluster or to Internet • In case of disconnection: kill VMs, stop IO access • In case of host failure: do IPMI shutdown • Restart the VMs on other hosts in priority order, specified by user VM V M V M V M V M V M V M V M VM VMVM VM host-agent X X 23 VM VM VM
  • 24. Application Level HA • Two kinds of failures • Infrastructure failures (our focus here) • Application bugs, hangs 24
  • 25. Standard Approach • Run application across multiple availability zones • Run a load-balancer in front 25 VM VM VM VM VM VM VM VM tier 1 tier 2 tier 3 AZ-east AZ-west tier 1 tier 2 tier 3 LB VM VM VM VM VM VM VM VM
  • 26. Problem #1 • No locality within AZ • Higher latency for inter-tier or inter-VM requests 26 VM VM VM VM VM VM VM VM AZ-east Rack 1 Rack N
  • 27. Problem #2 • No failure tolerance against within AZ failure • Rack, power or host failures can take down the app 27 VM VM VM VM VM VM VM VM AZ-east Rack 1 Rack N X X
  • 28. Solution • Control on placement • Use affinity rules within VMs across tiers • Use anti-affinity rules for VMs within a tier 28 VM VM VM VM VM VM VM VM tier 1 tier 2 tier 3 AZ-east Affinity within a group Anti-affinity across groups
  • 29. Conclusions • Current techniques are not sufficient for web-scale infrastructure • Key ideas for scalable, automated HA: • No special nodes, symmetric design • Automatic healing • Distributed consensus needed for taking decisions • VM level HA requires stronger failure detection & isolation • Use application level HA for better performance and higher reliability 29