SimpleWorkflow in the
Cloud
POA June 17, 2010
Alan Robbins and Rick Sears
Todays Talk
Motivations and Goals
AWS SimpleWorkflow (SWF)
Building Workflows
Workflow Patterns
Synchronous Web Services
A Workflow We All Know
About
Can you recognize a
Workflow?
Most developers dont know of workflows
Im just building my application!
This is simple, I just need a couple of
states
100s of workflows at Amazon right now!
Indicators?
Tasks longer than ~5 seconds
Sweep a database looking for state
Sequence of tasks connected by queues
Polling for activity completion
Distribute tasks to daemons
Background or batch jobs
What is a Workflow?
Multi-step activity orchestration
Business logic relating those activities
Address failures and exceptions
May be long-lived business activities
Asynchronous or event-driven execution
Lasting minutes or months (or longer)
Involve human decision making
Span organizational boundaries
BPM Market Segment
BPM Market Segment
Recognized ubiquitous need
Many standard workflow modeling
languages
BPML, WSFL, WSCL, BPSS, XLANG, WSCI, WSCDL, XAML, jPDL, xPDL, BPEL, YAWL,
Many Sophisticated BPM Software Products
IBM, Tibco, Oracle, Microsoft, JBoss, ...
Most extend a standard or require a
proprietary workflow modeling language
Vast majority of developers still build ad hoc
Huge market opportunity!
BPM Product Architecture
XML-based Workflow Definition
BPM Workflow Engine
Database
Downside of BPM Products
Large up-front costs (besides the $ for
product)
Require users to host the engine and database
Start with an esoteric XML modeling
language
Steep learning curve barrier for developers
Develop within a proprietary framework
Expose external functionality as SOAP endpoints
Add-on ESB technology and build adaptors
Approach is neither flexible nor scalable
Not designed for the cloud
Workflows in the Cloud?
Unique Musts of Cloud
Services
Service, not software product!
Low barrier to entry
Wide appeal
Suitable for mission critical apps
Support any programming language
Inexpensive relative to software
products
No practical scalability limitations
Secure
SimpleWorkflow Design
Goals
Open Architecture for Developers
Low-level Primitives with a Simple API
Provide all the Runtime Infrastructure
Enable users to define workflow
capabilities that are:
Compositional
Customizable
Sharable
Monetizable
Marketplace of Cloud
Capability
Marketplace of Cloud
Capability
Anyone can offer their activities and
workflows
Document API and extension points
Include pricing plans
BUY enables access
SimpleWorkflow brokers all interactions
Authentication, permissions, metering & billing
Amazon to offer capabilities
Utilizing other AWS services (S3 , EMR, MTurk, )
Amazons eCommerce capabilities
How to Accomplish these
Goals?
We had an Insight!
We had an Insight!
For any workflow
Regardless of representation language
Regardless of simplicity or complexity
Even in exceptional situations
At any point in time the decisions
about what to do next can be coded
as a stateless function of what
happened before!
A Decider Represents a
Workflow
Workflow
History
- Events
- Inputs
- Result
Decider
(stateless function)
Decisions
Dont need to define a modeling language!
SimpleWorkflow isnt
SimpleWorkflow
Architecture
SimpleWorkflo
w
Activity Workers
External worker processes
Deployed on multiple hosts
Receive Activity Tasks [Input]
Respond Completed [Result] or Failed
while true {
Poll for Activity Task (long poll)
<execute business activity, script, or
program>
Respond Completed (with Result) or Failed
}
Activity Workers
External worker processes
Deployed on multiple hosts
Receive Activity Tasks [Input]
Respond Completed [Result] or Failed
while true {
Poll for Activity Task (long poll)
<hand off task to a website or other app>
}
Respond Completed (with Result) or Failed
Decider
External worker processes
Deployed on multiple hosts
Receive Workflow Tasks [History]
Respond Completed [Decisions] or Failed
while true {
Poll for Workflow Task (long poll)
<apply decider logic: History Decisions>
Respond Completed (with Decisions) or
Failed
}
Decider is Open
Many innovative
implementation approaches:
Decider Framework [available now]
Annotations Framework
xPDL Interpreter
Asynchrony
?
<domain specific languages>
Worker Protocols
Advantages of polling
Secure: no endpoint exposed to the Internet!
Workers can be located behind corporate firewalls
Automatic throttling and load balancing
Easier to implement for many app types
Downsides of polling
A lot of work for nothing
To reduce latency, clients aggressively poll
Long polling: from Jabber and Comet
Eliminates the downsides of polling
Push protocols will be supported soon
SOAP and REST web services (+ BSF for Amazon
internal)
SNS task delivery
Runtime Infrastructure
Runtime Infrastructure
WF State
Persistence
Task
Synchronization
Scheduling &
Queuing
Timeout
Management
Task Routing
Versioning
Access/Permissio
ns
Monitoring/Metri
cs
Metering and
Billing
Audit Trail
Archival
Console
Simple API
Workflow Instances
Start
Signal
TryCancel
Terminate
Visibility of Instances
List, Describe,
GetHistory
Access/Permissions
Get/SetPolicy, Associate
Activity & Decider
Tasks
PollForTask
Respond
Completed or Failed
Decisions
Schedule Activity
TryCancel Activity
Record Split/Join
Complete Signal
Complete Workflow
Takeaways
Workflows are multi-step asynchronous
capabilities which may be very long lived
Workflows are ubiquitous
You probably have them if you know it or not!
Requires special techniques and infrastructure
SimpleWorkflow is a cloud-centric AWS service
Providing all the infrastructure you need
Marketplace enables sharing and monetization
Building Workflows using
SWF
Two Approaches
Choreography
Orchestration
Choreographies
Participants
coordinate actions
between one another
Everyone must know:
Where to look?
Whom to tell?
How to handle the
unexpected?
How to report status?
Item Processing
Pipeline
Orchestrations
Actions coordinated
by centralized
decision maker
Workers only need
to know:
How to do their job
How to communicate
Centralized status
and failure handling
Apollo deployment
SWFs Role
Both are valid approaches
Recognized need for support for
orchestration approach
SWF acts as auditable coordinator
between:
Decider acting as orchestrator
Activity workers performing specific jobs
Getting Started
First Things First
Break your workflow into pieces
What activities?
What data do they need?
How do they relate to one another?
What do I do when things go wrong?
How do I know whats happening?
Understand how the workflow will
evolve
Next Steps
Tell us about those pieces
Connect Activity Workers
SimpleWorkflo
w
Connect Decider
SimpleWorkflo
w
Decider
(stateless function)
Workflow
History
Decisions
Kick it off!
Designing Complex Systems
Design Patterns
Famous GOF patterns for designing software
Covers common practices in a language
agnostic way
Building blocks of application logic
Some examples:
Creational
Singleton, Factory Method
Structural
Faade
Behavioral
Visitor
Workflow Patterns
Common practices for building multi-step
applications
Building blocks for orchestrating applications
Some examples:
Control
Sequence, Parallel Split, Synchronization,
Data
Data-based Routing, Scope Data,
Exception Handling
Deadline Expiry,
Resource
Retain Familiar,
More Information
Patterns collected and documented
by a team lead by Professor Wil Van
Der Aalst and a team lead by
Professor Arthur ter Hofstede.
More details at:
www.workflowpatterns.com
Patterns and SWF
Control Patterns
Sequence
Structured Loop
Exclusive Choice with Simple Merge
Parallel Split with Synchronization
Persistent Trigger
Sequence
Sequence
Start
A1
A2
A3
Close
Simplest of all patterns
Used with other patterns to form
complex workflows
Also cover Task Data pattern
SWF primitives supporting this pattern:
Scheduling of activities with data
History of events
Start
A1
SimpleWorkfl
ow
A2
A3
Close
History
WorkflowInstanceStarted
WorkflowTaskScheduled
StartWorkflowInsta
nce
Start
A1
A2
Close
History
SimpleWorkfl
ow
PollForWorkflowTas
k
A3
WorkflowInstanceStarted
WorkflowTaskScheduled
Respons
e
Start
A1
A2
A3
Close
Decisions
SimpleWorkfl
ow
RespondWorkflowTaskCom
pleted
Schedule A1
History
WorkflowTaskCompleted
ActivityTaskScheduled
A1
Start
A1
A2
Close
Activity
Task
SimpleWorkfl
ow
PollForActivityTa
sk
A3
Input_1 = FOO
Respons
e
History
ActivityTaskScheduled
A1
ActivityTaskStarted A1
A1
Start
A1
A2
A3
Close
Result
SimpleWorkfl
ow
RespondActivityTaskCompl
eted
Datum_1 = FOO
History
A1
ActivityTaskStarted A1
ActivityTaskCompleted
A1
WorkflowTaskScheduled
Structured Loop
Structured Loop
n
Start
A1
A2
A3
Close
Deals with bounded re-executions of
steps or collections of steps
Simple example entails retries of single step
Also cover Deadline Expiry exception
handling pattern
SWF features supporting these patterns:
Activity timeouts
Rescheduling activities
Start
A1
A2
Respons
e
A2
Close
History
SimpleWorkfl
ow
PollForActivityTa
sk
A3
ActivityTaskScheduled
A2
ActivityTaskStarted A2
Start
A1
SimpleWorkfl
ow
RespondActivityTaskCompl
eted
A2
A2
A3
Close
Start
A1
A2
Close
History
SimpleWorkfl
ow
PollForWorkflowTas
k
A3
Respons
e
ActivityTaskStarted A2
ActivityTaskTimedOut A2
WorkflowTaskScheduled
Start
A1
A2 (2)
A3
Close
Decisions
SimpleWorkfl
ow
RespondWorkflowTaskCom
pleted
Schedule A2 (2)
History
ActivityTaskStarted A2
ActivityTaskTimedOut A2
ActivityTaskScheduled
A2
Start
A1
A2 (2)
A3
Close
History
SimpleWorkfl
ow
PollForActivityTa
sk
Respons
e
A2
ActivityTaskStarted A2
ActivityTaskTimedOut A2
ActivityTaskScheduled
A2
ActivityTaskStarted A2
Start
A1
A2 (2)
A3
Close
Result
SimpleWorkfl
ow
RespondActivityTaskCompl
eted
Datum_1 = FOO
History
A2
ActivityTaskStarted A2
ActivityTaskCompleted
A2
WorkflowTaskScheduled
Start
A1
History
ActivityTaskStarted A2
ActivityTaskTimedOut A2
ActivityTaskScheduled A2
ActivityTaskStarted A2
ActivityTaskCompleted A2
WorkflowTaskScheduled
A2 (2)
A3
Decisions
Schedule A3
Close
Exclusive Choice with Simple
Merge
Exclusive Choice with Simple
Merge
A4
a
Start
A1
A2
A3
Close
Choose one step (or set of steps) over another
Choice can be based on results of activity execution
Alternative path merged back with normal control flow
at later point
Simple example entails handling of a failure condition
Also cover Data-based Routing pattern
SWF features supporting these patterns:
Explicit failure notification
Data passing and persistence
A4
Start
A1
A2
Respons
e
A2
Close
History
SimpleWorkfl
ow
PollForActivityTa
sk
A3
ActivityTaskScheduled
A2
ActivityTaskStarted A2
A4
Start
A1
A2
A3
Close
Result
SimpleWorkfl
ow
RespondActivityTaskFail
ed
FAULT_DETAILS = FOO
History
ActivityTaskStarted A2
ActivityTaskFailed A2
WorkflowTaskScheduled
A2
A4
Start
A1
A2
Close
History
SimpleWorkfl
ow
PollForWorkflowTas
k
A3
Respons
e
ActivityTaskStarted A2
ActivityTaskFailed A2
WorkflowTaskScheduled
A4
Start
A1
A2
A3
Close
Decisions
SimpleWorkfl
ow
RespondWorkflowTaskCom
pleted
Schedule A4
History
ActivityTaskStarted A2
ActivityTaskFailed A2
ActivityTaskScheduled
A4
Parallel Split with
Synchronization
Parallel Split with
Synchronization
A1
Start
A3
Close
A2
Execute two or more steps (or sequences of
steps) at once
Continue forward with only one branch once
parallel activities are done
Also cover Scoped Data
SWF features supporting these patterns:
Recording split/join
Scheduling with split/join
Branch-scoped data
A1
Start
A3
Close
A2
History
SimpleWorkfl
ow
PollForWorkflowTas
k
WorkflowInstanceStarted
WorkflowTaskScheduled
Respons
e
A1
Start
A3
Close
A2
Decisions
SimpleWorkfl
ow
RespondWorkflowTaskCom
pleted
Record Split
Schedule A1
Schedule A2
History
WorkflowTaskCompleted
SplitRecorded
ActivityTaskScheduled A1
ActivityTaskScheduled A2
A1
Start
A3
Close
A2
History
SimpleWorkfl
ow
PollForActivityTa
sk
Respons
e
A1
SplitRecorded
ActivityTaskScheduled A1
ActivityTaskScheduled A2
ActivityTaskStarted A1
A1
Start
A3
Close
A2
History
SimpleWorkfl
ow
PollForActivityTa
sk
Respons
e
A2
ActivityTaskScheduled A1
ActivityTaskScheduled A2
ActivityTaskStarted A1
ActivityTaskStarted A2
A1
Start
A3
Close
A2
Result
SimpleWorkfl
ow
RespondActivityTaskCompl
eted
Datum_1 = FOO
History
A1
ActivityTaskStarted A1
ActivityTaskCompleted
A1
WorkflowTaskScheduled
A1
Start
A3
Close
A2
History
SimpleWorkfl
ow
PollForWorkflowTas
k
Respons
e
ActivityTaskScheduled A1
ActivityTaskScheduled A2
ActivityTaskStarted A1
ActivityTaskStarted A2
ActivityTaskCompleted A1
WorkflowTaskScheduled
A1
Start
A3
Close
A2
Decisions
SimpleWorkfl
ow
RespondWorkflowTaskCom
pleted
History
ActivityTaskStarted A1
ActivityTaskStarted A2
ActivityTaskCompleted A1
WorkflowTaskScheduled
WorkflowTaskStarted
WorkflowTaskCompleted
A1
Start
A3
Close
A2
Result
SimpleWorkfl
ow
RespondActivityTaskCompl
eted
Datum_1 = BAR
History
A2
ActivityTaskStarted A2
ActivityTaskCompleted A2
WorkflowTaskScheduled
A1
Start
A3
Close
A2
History
SimpleWorkfl
ow
PollForWorkflowTas
k
Respons
e
ActivityTaskCompleted A1
ActivityTaskCompleted A2
WorkflowTaskScheduled
A1
Start
A3
Close
A2
Decisions
SimpleWorkfl
ow
RespondWorkflowTaskCom
pleted
Record Join
Schedule A3
History
ActivityTaskCompleted A1
ActivityTaskCompleted A2
WorkflowTaskCompleted
JoinRecorded
ActivityTaskScheduled A3
Persistent Trigger
Persistent Trigger
Start
A1
Wait
A2
Close
Trigger from external process affecting the
workflow execution
Triggers come in the form of signals
Workflow can wait for a period of time for signal
Also cover Event-based Task Trigger
SWF features supporting these patterns:
Signal workflow
Complete signal
Schedule activity with delay
Cancel activity
Start
A1
Wait
Close
History
SimpleWorkfl
ow
PollForWorkflowTas
k
A2
Respons
e
ActivityTaskStarted A1
ActivityTaskCompleted A1
WorkflowTaskScheduled
Start
A1
Wait
A2
Close
Decisions
SimpleWorkfl
ow
RespondWorkflowTaskCom
pleted
Schedule Wait (Empty, Delayed)
History
WorkflowTaskCompleted
ActivityTaskScheduled
Wait
Start
A1
SimpleWorkfl
ow
SignalWorkflowInsta
nce
Wait
A2
Close
History
ActivityTaskScheduled
Wait
SignalRecorded
WorkflowTaskScheduled
Start
A1
Wait
Close
History
SimpleWorkfl
ow
PollForWorkflowTas
k
A2
Respons
e
ActivityTaskScheduled Wait
SignalRecorded
WorkflowTaskScheduled
Start
A1
Wait
A2
Close
Decisions
SimpleWorkfl
ow
RespondWorkflowTaskCom
pleted
Complete Signal
Cancel Wait
Schedule A2
History
WorkflowTaskCompleted
SignalClosed
ActivityTaskCancelTried
Wait
ActivityTaskCanceled Wait
ActivityTaskScheduled A2
Start
A1
History
ActivityTaskStarted Wait
ActivityTaskCompleted
Wait
WorkflowTaskScheduled
Wait
A2
Close
Decisions
Many more
Multi-Choice
Workflow Data
Interleaved Parallel Routing
Direct Distribution
Suspension-Resumption
Cancelling Discriminator
Cancel Region
Explicit Termination
Data Interaction - to Multiple Instance Task
Blocking Partial Join
Selection Autonomy
Capability-Based Distribution
Multiple Instances with a Priori Run-Time
Knowledge
Task Postcondition - Data Existence
Decider Framework
Provides direct support for many
patterns
Build workflows by composing
scoped deciders
Simple set of components combine
to express complex workflows
Composing Deciders
Workflow
Sequence
Split/Join
Activity
A1
Start
A1
A2
Activity
A2
A3
Activity
A3
Close
Provided Decider
Implementations
WorkflowDecider closes/fails workflow and input/result
ActivityDecider controls single activity execution and
retries
SequenceDecider sequence of deciders
SwitchDecider one of the set based on condition
SplitJoinDecider parallel execution of deciders
WhileDecider loop with precondition
DoWhileDecider loop with postcondition
TryCatchFinallyDecider exception based error handling
WaitForSignalDecider blocks execution until signal
received
DelayDecider delays execution of a decider
RetryDecider retries execution of a decider in case of
failures
ConditionDecider conditionally executes a decider
Links
Patterns:
www.workflowpatterns.com
Wiki:
w.amazon.com/?SWF
Email:
[email protected]
[email protected]
[email protected]
Contact-us
Takeaways
Steps to build and execute workflows
using SimpleWorkflow
Workflow patterns are useful
conceptual design tools
SimpleWorkflow has the capabilities
necessary to express many workflow
patterns
Frameworks are available to make your
job easier
Thanks!
SWF Team
Rodrigo De Castro
Drew Gochenaur
Maxim Fateev
Jacques Thomas
Andy Kotlinski
Fatih Boyaci
Pat Maguire
Srikanth Satyanarayana
POA Talk Advisors
Users
