Introduction to
Distributed
Architecture
What is Architecture?
SEI @ Carnegie Mellon
A software architecture
is an abstract view of a
software system
distinct from the details
of implementation,
algorithms, and data
representation.
Why Architecture Matters
Accidental Architecture
Grady Booch
Every… system has an
architecture. While
some of these
architectures are
intentional, most
appear to be accidental
Waterfall Process
Requirements
Analysis
Design
Code and Test
Integration
System Test
Waterfall Process
Requirements
Analysis
Analysis
Paralysis
Design
Design divorced
from reality
Code and Test
Massive
Integration
Integration
System Test
Manifesto for Agile Software Development
We are uncovering better ways of developing
software by doing it and helping others do it.
Through this work we have come to value:
Individuals and interactions over processes and tools
Working software over comprehensive documentation
Customer collaboration over contract negotiation
Responding to change over following a plan
That is, while there is value in the items on
the right, we value the items on the left more.
Scrum: An agile Process
Techies crave Extremes
Architect
Code
Functional requirements
What the system should do.
• Use cases, User stories, acceptance criteria
• Defines passing QA
• Day to day you focus on this
Nonfunctional requirements
What the system should be.
• Extremely expensive or impossible to “fix”
later on
• Defined passing Production
• Assumed by your customers and users
Nonfunctional Requirements
•
•
•
•
•
•
Availability
Stability
Efficiency
Reliability
Maintainability
Extensibility
•
•
•
•
•
•
Fault Tolerance
Security
Capacity
Latency
Flexibility
Scalability
Functional:
Passed QA
Beautiful site (in my opinion)
Easy to use
Nonfunctional:
Failed Production
Availability
Stability
Latency
Capacity
Throughput
Architecture
There is no best architecture
Pacemakers and Guided Missiles
The Flying Buttress
Patterns
Example: Layered Architecture
Three Tier Architecture
The classic 3 tier architecture
N-Tier Architecture
Because three tiers wasn’t enough!
N-Tier Architecture
SMS
SHAREPOINT
IIS
BIZTALK
Windows Server 2008
SQL SERVER
License
management
Questions about Architecture?
Distributed Architecture
Michael T. Nygard
Designs appropriate for
small brochureware
websites fail outrageously
when applied to thousanduser, transactional,
distributed systems...
Key Constraints on Distributed Systems
•
•
•
•
•
Stability
Reliability and fault tolerance
Consistency
Capacity and Scalability
Security
Scenario: Stability
You are the lead developer on a national health
care site that will register millions of users a week.
You are responsible for the signup process.
Your team must verify the identity of users with a
third party API. You call the service and the third
party system will return a boolean true if the
identity is valid or false otherwise.
private string Register(RegistrationInfo registrationInfo)
{
try
{
bool validIdentity = identityService.verifyIdentity(registrationInfo);
if (validIdentity)
{
database.save(registrationInfo);
return “register_success.htm";
}
return “register_failure.htm";
}
catch (Exception)
{
return "errorpage.htm";
}
}
Stability
1. Register
6. Registered
2. Verify
Site Server
3. Verified
4. Save User
5. User Saved
Database
Identity
Verification
Unbalanced Capacities
1. Register
6. Registered
2. Verify
Site Server
3. Verified
4. Save User
5. User Saved
Database
Identity
Verification
Blocked Threads
Verify joe
Verify joe
joe verified
joe verified
Verify sally
Site Server
Verify sally
Verify bob
Verify bob
Verify bob
Verify bob
Verify bob
Verify bob
Identity
Verification
Cascading Failures
Site
Server
Site Server
Site
Server
Site
Server
Pattern 1: Timeouts
private string Register(RegistrationInfo registrationInfo)
{
try
{
identityService.Timeout = 10000;
bool validIdentity = identityService.verifyIdentity(registrationInfo);
if (verified)
{
db.save(registrationInfo);
return “register_success.htm";
}
return “register_failure.htm";
}
catch (Exception)
{
return "errorpage.htm";
}
}
Pattern 1: Timeouts
Site Server
Identity
Verification
Pattern 2: Non-blocking I/O
private void BeginRegister(RegistrationInfo registrationInfo, Function<string> callback)
{
try
{
identityService.EndIdentity += identityService_EndIdentity(callback)
identityService.BeginIdentity(registrationInfo);
}
catch (Exception)
{
callback("errorpage");
}
}
private void identityService_EndIdentity(bool success, Function<string> callback)
{
if (success)
{
callback("register_success.htm");
}
callback("register_failure.htm");
}
Pattern 2: Non-blocking I/O
Site Server
Identity
Verification
For Users…
==
The site is stable. We’re still failing.
What’s the president recommending?
What might be happening
Thank you. We’ll email
you when you are
verified.
What might be happening
Identity
Verification
“The” Registration Process
Udi Dahan
When two principles
are pushing in opposite
directions, some
underlying assumption
is wrong. Often the
word the is the culprit
Read this again
You are the lead developer on a national health
care site that will register millions of users a week.
You are responsible for the signup process.
Your team must verify the identity of users with a
third party API. You call the service and the third
party system will return a boolean true if the
identity is valid or false otherwise.
Pattern 3: Decoupling
Thank you. We’ll email
you when you are
verified
Site Server
Pending
Registration
Database
Pattern 3: Decoupling
Pending
Registration
Database
Verification
Application
Identity
Verification
Scenario 2: Reliability
You are the lead developer on a hospital’s
prescription filling service. Your RX wholesaler has
provided you with an HTTPS endpoint to integrate
with.
It is critical a prescription is not accidently
prescribed twice and that prescriptions are not
lost.
Reliability
1. Prescribe
5. Success
2. Fill RX
Site Server
3. RX ID
4. RX ID and Fill Info
5. Record Updated
Database
RX Service
Invoking the Service
private string Prescribe(PrescriptionInfo prescriptionInfo)
{
try
{
RxService rxService = new RxService();
int rxID = rxService.Prescribe(prescriptionInfo);
database.Save(rxId, prescriptionInfo);
return "success.htm";
}
catch (Exception)
{
return "errorpage.htm";
}
}
What if the network goes down?
2. Prescribe
1. Prescribe
Site Server
404 Timeout
RX Service
Invoking the Service
private string Prescribe(PrescriptionInfo prescriptionInfo)
{
try
{
RxService rxService = new RxService();
int rxID = rxService.Prescribe(prescriptionInfo);
database.Save(rxId, prescriptionInfo);
return "success.htm";
}
catch (Exception)
{
return "errorpage.htm";
}
}
Can I retry?
Site Server
Prescribe
RX Service
404 Timeout
Got it, but I couldn’t get Back to you.
Site Server
Prescribe
RX Service
Pattern 1: Idempotency
Site Server
Prescribe
RX Service
404 Timeout
Got it, but I couldn’t get Back to you.
Site Server
Prescribe
RX Service
Sheesh… I already got it!
What if the database is down?
1. Prescribe
5. Success
2. Fill RX
Site Server
3. RX ID
4. Update Patient Record
Database
RX Service
Invoking the Service
private string Prescribe(PrescriptionInfo prescriptionInfo)
{
try
{
RxService rxService = new RxService();
int rxID = rxService.Prescribe(prescriptionInfo);
database.Save(rxId, prescriptionInfo);
return "success.htm";
}
catch (Exception)
{
return "errorpage.htm";
}
}
Can We Guarantee this code?
private string Prescribe(PrescriptionInfo prescriptionInfo)
{
try
{
RxService rxService = new RxService();
int rxID = rxService.Prescribe(prescriptionInfo);
database.Save(rxId, prescriptionInfo);
return "success.htm";
}
catch (Exception)
{
return "errorpage.htm";
}
}
Pattern 2: Transactional Queues
Transactional Queue
1. Prescribe
3. Pending
Site Server
Fill RX Message
Queue
Inserting into a Queue
private string Prescribe(PrescriptionInfo prescriptionInfo)
{
try
{
Queue.save(new RXPrescribeMessage(prescriptionInfo));
return "rxPending.htm";
}
catch (Exception)
{
return "errorpage.htm";
}
}
Transactional Queue
rxPrescribeMessage
Queue
rxDBUpdateMessage
Fill rx
Handler
rx ID
RX Service
Transactional Queue
Queue
rxDBUpdateMessage
Handler
Save rx
Database
Message Handlers
public void HandleRXPrescribeMessage(RXPrescribeMessage message)
{
var prescriptionInfo = message.prescriptionInfo;
RxService rxService = new RxService();
int rxID = rxService.Prescribe(prescriptionInfo);
Queue.save(new RXDBUpdateMessage(rxID, prescriptionInfo));
}
public void HandleDBUpdateMessage(RXDBUpdateMessage message)
{
var rxId = message.rxID;
var prescriptionInfo = message.prescriptionInfo
database.Update(message.rxID, message.prescriptionInfo)
}
Nygard, Michael T.
Cynical software expects bad
things to happen and is never
surprised when they do.
Cynical software doesn’t even
trust itself.. It refuses to get
too intimate with other
systems, because it could get
hurt.