Cloud-Native in Azure Zoran B. Djordjevic’s CSCI E-175 Cloud Computing and Software as a Service class at Harvard University 19-November-2010 Copyright.
Post on 05-Jan-2016
217 Views
Preview:
Transcript
Cloud-Native in Azure
Zoran B. Djordjevic’s CSCI E-175 Cloud Computing and Software as a Service class at Harvard University
19-November-2010
Copyright (c) 2010, Bill Wilder
Boston Azure User Grouphttp://bostonazure.org@bostonazure
Bill Wilderhttp://blog.codingoutloud.com@codingoutloud
Boston West Toastmasters http://bwtoastmasters.com
Not here with my day jobOnly Bill’s personal views
Services & Patterns for Building Cloud-Native Applications on Microsoft’s Windows Azure Stack
Agenda
Azure provides Affordances that:-Nudge you towards scalable,-Cost-effective, -Durable application architectures
We will look at those services and “natural” patterns…
[Azure] Cloud Platform• The rise of focused, specialized services– Really good at one thing– True of other Cloud Platforms
• “Scalable” Persistent Storage abstractions– Set of Services– Illusion of infinite scale out– Same performance with x clients as with 10x as with
100x as with 1000x as with …
• Billing model: pay only for what you use– Friction-free access to add’l or less storage
Part A
Storage services
Part A – Executive Summary
Use Azure Blobs and Tables and your DATA will be resilient to failure and will scale like crazy…
Persistent Storage Services – Options
Type of Data Traditional Azure Way
Relational SQL Server SQL Azure
Blob File System, or SQL Server
Azure Blobs
File File System Azure Drives, or Azure Blobs
Reliable Queue MSMQ (maybe) Azure Queues
Non-Relational Azure Tables
Azure Storage and Services
• Family of specialized, complementary storage and messaging services– Clean abstractions– Specialized services
• Queues • Blobs, Drives, CDN• Tables • SQL Azure
Azure API
• RESTful interfaces for interacting with Queue, Table Storage, and Blob Storage
• SQL Azure is same as SQL Server• Client agnostic
• [Service Management API also exists and is RESTful]
All Azure Storage is Durable
• Resilient in case of failure– “All data replicated multiple times” – Sriram Krishnan,
Programming Windows Azure, p130+– 3 copies
• SQL Azure supports full relational semantics– ACID = Atomicity, Consistency, Isolation, and Durability– Part of the PaaS-ness
• Blob Storage, Table Storage, Queues– Which of these support ACID?
Azure Blob Storage• The place to put ANY LARGE OBJECT• Private or Public• Public blobs can be anonymously accessed– Images, videos, CSS files– Need not be binary
• Public blobs can be cached in Azure CDN– 20+ locations around the world– Different locations than data centers
• Time-limited signed-access available– “You have until tomorrow to download this video”
Azure Blob Storage
• Two types of Azure Blobs1.“Block Blob”– Parallel upload scenarios– Resumable download/streaming scenarios– Up to 200 GB
2.“Page Blob”– Random read/write access scenarios– Azure Drives– Up to 1000 GB
Azure Blob Storage
Storage Account
Container[*]
Blob[*]Unique URL
Properties [several]Metadata [255 entries]
Name/Value/Type
Azure Table Storage
• Best place for granular, semi-structured data– No rigid database schema
• Fast and easy to instantiate– Strongly Consistent– No performance lag
• Programming model is WCF Data Services– All data access and data updates– LINQ
Azure Table Storage
Storage Account
Table [*]
Entity [*, 1 MB data] PartitionKey + RowKey
Property [255x]Name/Value/Type
Up to 64k
Azure Table Storage
• Partition Key– Along a “logical grouping” – a “shard”– PartitionKey value of up to 64 KB
• Row Key– Identify specific row within a partition– RowKey value is String of up to 64 KB
• Table access requires Partition Key + Row Key– Not to mention (cryptographic) Access Key (for the
digital signing of the http header)
Relational Data vs. Azure TablesApproach SQL Azure Azure Tables
Normalization Normalized Denormalized(Duplication) (No duplication) (Lots of duplication)
Structure Schema Flexible
Transactions Distributed Limited scope
Responsibility Database Developer
Knobs Many Few
Scale Up (or Sharding) Out
Cost Reasonable Less expensive
Pay As You Go – Storage
• Storage– $0.15 per GB stored per month– $0.01 per 10,000 storage transactions
• Applies equally to:– Queues– Tables– Blobs– Drives
Pay As You Go – Data Transfer
• North America and Europe regions– $0.10 per GB in– $0.15 per GB out
• Asia Pacific Region– $0.30 per GB in– $0.45 per GB out
• No charge for transfer within a data center
What is Cost of 1 Byte?
• Azure Storage cost/byte = x• SQL Azure cost/byte = 66x• Practical multiplier might be 100x
Part B
Decoupling patterns
Part B – Executive Summary
Scale out, not up, and your COMPUTE will be resilient to failure and will scale like crazy…
Azure Does Compute
Compute Services• Web Roles• Worker Roles
• VM Role (Nov 2010)
• Your Code Runs Here
“Out” is the New “Up”
• Scaling Out has hard limits at CPU, Memory– Architecturally more limiting
Azure Storage Services in Concert
Azure Queue Storage
• In Azure’s architecture, part of Storage Family of Services– Builds on Blob Storage
• In Your architecture, it is your Messaging between loosely-coupled system components– Compute + Storage + Messaging
Key Pattern: Roles + Queues
WebRole(IIS)
WebRole(IIS)
WorkerRole
WorkerRole
Queues
BlobsTables
Canonical Example: Thumbnails
WebRole(IIS)
WebRole(IIS)
WorkerRole
WorkerRole
Queues
BlobsTables
Adding to Queue - Conceptual
Azure Blob Storage
Adding to Queue - Actual
314159 265358 979323
Roles + Queues: API
WebRole(IIS)
WebRole(IIS)
WorkerRole
WorkerRole
Queues
queue.AddMessage( new CloudQueueMessage( payloadStringOrByteArray));
CloudQueueMessage statusUpdateMessage = queue.GetMessage( TimeSpan.FromSeconds(10));
… queue.DeleteMessage(statusUpdateMessage);
General Case: Rolesn + Queuesn
RnQnRn
WebRole(IIS)
WebRole(IIS)
WorkerRole
WorkerRole
Queues
WebRole(IIS)
WebRole(IIS)
WebRole(IIS)
WebRole(IIS)
WebRole(IIS)
WebRole(IIS)
WorkerRole
WorkerRoleWorker
RoleWorker
RoleWorker
Role Type 1
WorkerRole
Type 1
WorkerRole
WorkerRoleWorker
RoleWorker
RoleWorkerRole
WorkerRoleWorker
Role Type 2
WorkerRole
Type 2
Look @ WorkerRole.csThumbnailing Code
public class WorkerRole : RoleEntryPoint {
public override void Run() { … }}
Azure Queues – by the numbers
• Service for RELIABLE message delivery• 7 days = default TTL for item to stay in queue• 30 seconds = default “invisibility window”• 8 KB = max size of a queued item• 500 = approx number of transactions a queue
can handle per second– Beware of “spinning” – may get throttled, disabled
• What happens if you need more than 500 tps?
Key Metric
• Queue length (and trend) is key data point for tuning Role deployment numbers–Available programmatically for
monitoring–May vary across queue types
RnQnRn enables Responsive
• Response to interactive users is as fast as a work request can be persisted
• Time consuming work done off-line• Same total resource consumption, better
subjective experience• UX challenge – how to express Async to users?– Communicate Progress– Display Final results
RnQnRn enables Scalable
• Loosely coupled, concern-independent scaling• Blocking is Bane of Scalability– Decoupled front/back ends insulate from other
system issues if…– Twitter down– Email server unreachable– Order processing partner doing maintenance– Internet connectivity interruption
RnQnRn enables Distribution• Scale out systems better
suited for geographic distribution
– More efficient and flexible because more granular
– Hard for a mega-machine to be in more than one place
– Failure need not be binary
Optimization is optional• Individual role utilization may be low– Role is a VM – lots of resources– You pay by instance, not resource use within
• Make sure VM instances are “right sized”– XS ($0.05/hr), Small ($0.12/hr), Medium, Large, XL
• Make sure enough roles for uptime– SLA requires minimum of 2 instances
• Business Trade-Off for further optimizations– Optimize for CPU utilization (multiple threads)– Combine types of processing into fewer role types
RnQnRn requires Idempotent
• If we do a task twice, end result same as if we did it once
• App-specific concerns dictate approaches– Compensating transactions– Last in wins– Many others possible – hard to say
• Example with Thumnailing
RnQnRn requires Poison Message Strategy
• A Poison Message is not able to be processed– Error condition– Non-transient reason– CloudQueueMessage.DequeueCount property
• Strategy One: – Fall off the queue (TTL)– Message stays in queue for 7 days (default)
• Strategy Two:– Specify retry threshold– Remove poison messages
RnQnRn enables Resilient
• And Requires that you “Plan for failure”• There will be role restarts• Bake in handling of restarts– Not an exception case! Expect it!– Restarts are routine, system “just keeps working”
• If you follow the pattern, the payoff is substantial…
What’s Up?Aspirin-free Reliability as EMERGENT PROPERTY
Typical Site Any Azure Role Overall SystemOperating System UpgradeApplication Update / DeployChange TopologyHardware FailureSoftware Bug / Crash / FailureSecurity Patch
Questions?
?
BostonAzure.org
• Boston Azure cloud user group• Focused on Microsoft’s cloud solution• Next meeting: 6-8:30 PM Mon Dec 13sh 2010– Hacking on “Boston Azure Project”– Silverlight and Azure – better together
• Meetings usually 4th Thursday of month– No cost; food; great topics; growing community; wifi
• Join email list: http://bostonazure.org• Follow on Twitter: @bostonazure
Slides available from Bill’s blog
http://blog.codingoutloud.com-or-
http://hmbl.me/3KTBOE
Note: hmbl.me is a URL shortening service running on Azure.
Bill Wilder@codingoutloudhttp://blog.codingoutloud.com
top related