How to break apart a monolithic system safely without destroying your team - Velocity 2016 - Matthew Skelton

DevOps comes in many flavors: there is no one-size-fits-all. This is particularly true of team structures and responsibility boundaries, where what works for one organization might be completely unsuitable for another.

Since 2013 we have curated the DevOps Team Topologies (http://devopstopologies.com/), whilst helping organizations around the world to adopt and evolve effective team structures for DevOps. The changing technology landscape of IaaS, PaaS, containers, microservices, and serverless architectures all shape the communication patterns and responsibilities of technical teams, together with Conway's Law. Furthermore, as engineering organizations grow and mature different topologies become suitable.

---

From a talk at Velocity 2016

Team topologies should be mutable and responsive to the conditions around the teams. This talk will provide insights into how, why and when to evolve team topologies based on our real-world experience.

We also explain the key difference between an enabling 'capability' team and a service-providing platform team and the implications for DevOps and collaboration.

The insights are based on work with a variety of organizations including: a global publishing organisation, software vendors with blue-chip clients, a global technology company specialising in telecoms, several organisations in the financial services sector in the UK, an international commodities trading organisation, and a major UK broadcaster.

Transcript

1. How to break apart a monolithic system safely without destroying your team Matthew Skelton, Skelton Thatcher Consulting @matthewpskelton 07 November 2016, Amsterdam, NL - #velocityconf
2. Today Cognitive load for teams ‘Monolith’ Code Forensics Team-first boundaries Monolith-splitting recipe
3. For now, let’s forget: Microservices CQRS / Event Sourcing Queues (Architectural changes)
5. Team-first digital transformation 30+ organisations UK, US, EU, India, China
7. How to break apart a monolithic system safely without destroying your team
8. Safer, more rapid changes to software systems (Business Agility)
9. A ‘team-first’ approach to software subsystem boundaries
10. TEAM
12. TEAM capabilities appetite & aptitude understanding responsibilities
13. (assumption) the team is stable, slowly changing, and long-lived #NoProjects
14. Conway’s Law
15. “organizations which design systems ... are constrained to produce designs which are copies of the communication structures of these organizations” – Mel Conway, 1968 http://www.melconway.com/Home/Conways_Law.html
16. Front-end developers Back-end developers
17. ‘Reverse Conway’ Tobbe Gyllebring (@drunkcod)
19. homomorphic force (#Conway  #Yawnoc) HT @allankellynet (same) (shape)
20. Cognitive load for teams
21. Cognitive load the total amount of mental effort being used in the working memory (see Sweller, 1988)
22. Cognitive load Intrinsic Extraneous (Irrelevant ) Germane (Relevant)
23. ‘Hacking Your Head’: Jo Pearce See http://www.slideshare.net/JoPearce5/hacking-your-head-managing-information-overload-45-mix @jdpearce
24. We have SCIENCE!
25. Science since 1988 (!) • Driskell et al, 1999 ‘Does Stress Lead to a Loss of Team Perspective?’ Group Dynamics: Theory, Research, and Practice 3, no. 4 (1999): 291. • Fan et al, 2010 ‘Learning HMM-Based Cognitive Load Models for Supporting Human-Agent Teamwork’. Cognitive Systems Research 11, no. 1 (2010): 108–119. • Ilgen & Hollenbeck, 1993 ‘Effective Team Performance under Stress and Normal Conditions: An Experimental Paradigm, Theory and Data for Studying Team Decision Making in Hierarchical Teams with Distributed Expertise’. DTIC Document, 1993. • Johnston et al, 2002 ‘Application of Cognitive Load Theory to Developing a Measure of Team Decision Efficiency’. DTIC Document, 2002. • Sweller, John, 1994 ‘Cognitive Load Theory, Learning Difficulty, and Instructional Design’. Learning and Instruction 4 (1994): 295–312. • Sweller, John, 1988. ‘Cognitive Load during Problem Solving: Effects on Learning’. Cognitive Science 12, no. 2 (1988): 257–285.
26. “stress impacts team performance … by narrowing or weakening the team-level perspective required for effective team behavior.” – Driskell et al, 1999 Group Dynamics: Theory, Research, and Practice 1999, Vol. 3, No. 4,291-302
27. (not just ‘pop’ science!)
28. ‘Monolith’
29. monolith μόνο λίθος “single stone” Reiner Flassig - CC BY-SA 2.0 de - Wikipedia
30. “Don’t start with a monolith when your goal is a microservices architecture” – Stefan Tilkov, innoQ http://martinfowler.com/articles/dont-start-monolith.html
31. “Start monolithic and extract” – Tammer Saleh, Pivotal https://www.infoq.com/presentations/cloud-anti-patterns
32. A ‘team-first’ approach to software subsystem boundaries
33. Types of software monoliths •Application monolith •Joined at the DB •Monolithic build (rebuild everything) •Monolithic releases (coupled) •Monolithic thinking (standardisation)
34. Application monolith Single block of code Deployed as a unit
35. Joined at the DB Difficult to change separately (but not impossible) Risk is (probably) elevated Chris Collyer, http://www.stone-circles.org.uk/stone/pentreifan.htm
36. Monolithic builds One gigantic CI build just to get a new version of any component
37. Monolithic releases Smaller components bundled together into a ‘release’
38. Monolithic thinking ‘One-size-fits-all’ for teams Assumption that minimising variation is A Good Thing
39. Dangers of splitting a monolith •Reduced domain consistency •Data duplication (unintentional) •Additional operational complexity due to distributed system and async messaging •Degraded UX across the product
43. Splitting a monolith Reiner Flassig - CC BY-SA 2.0 de - Wikipedia Choose the right technique for splitting Understand the nature of the monolith
45. ‘Fracture planes’ for code •Business domain bounded context •Regulatory compliance •Change cadence •Risk •Performance isolation •User personas •Team location
46. ‘Fracture planes’ for code Ask: Could we consume this thing as a service?
47. ‘Fracture planes’ for code Ask: Could we provide this thing as a service?
48. Code Forensics
49. Forensics Your Code as a Crime Scene Adam Tornhill
50. Adam Tornhill Code, Crime, Complexity: Analyzing software with forensic psychology Adam Tornhill TEDxTrondheim youtube.com/watch?v=qJ_hplxTYJw
51. ‘Code Maat’ tool
52. Adam Tornhill, http://www.adamtornhill.com/articles/crimescene/codeascrimescene.htm Code City plus Code Maat forensics
53. Beware of badly-named subsystems
54. "information-poor abstract names are magnets for extra [unwanted] responsibilities" – Adam Tornhill p.185, Your Code as a Crime Scene
55. Team-first boundaries
56. DevOpsTopologies.com
57. Team types Component team Platform / ’substrate’ team Supporting / ‘productivity’ team Product/Feature team
58. Team types devopstopologies.com Platform / ’substrate’ team Product/Feature team
59. Team types devopstopologies.com Component team Platform / ’substrate’ team Product/Feature team
60. Team types devopstopologies.com Component team Platform / ’substrate’ team Product/Feature team Supporting / ‘productivity’ team
61. Code repositories Align repositories to subsystem boundaries Avoid monolithic-y repos like TFS* * Don’t get me started on TFS, grrr…
62. Code repositories Repo 1 Build Test Deploy Run Repo 2 Build Test Deploy Run Repo 3 Build Test Deploy Run
63. “You can use a monorepo only if your organisation has published a scientific paper on Computer Science. Otherwise, use one repo per deployable runnable thing.” – Matthew Skelton LondonCD meetup group, 11 Oct 2016 
64. Find natural or available ‘fracture planes’ for splitting a monolith (with the team in mind)
65. Industry experience Examples from recent work with clients Different monoliths Different ‘fracture planes’
66. Software for R&D teams Customers: over 80% of top 20 global pharma companies
67. 2014: Monolithic: builds, releases Pharma GxP: traceability Changes slow and tricky
68. Fracture plane: business domain
71. Fracture plane: technology
72. Compose with packages Logging for insights Technology splits Improved traceability
73. Major UK broadcaster (TV + online) Increasing digital delivery of content
74. 2014: Wide mix of technologies Need for visibility of changes Need for rapid turnaround
75. Fracture planes: read-only, location & cadence
76. Split off Reporting (read-only) Split on change cadence Increased change throughput
77. UK’s second largest credit reference agency Customers: major banks + large orgs
78. 2014: 6-weekly monolithic release Outage-sensitive customers
79. Fracture planes: risk & business domain
80. Split releases based on risk Align teams to value streams More rapid release cadence
81. Software for Internal Comms Customers: FTSE 100 & Fortune 1000
82. 2015: Successful system (monolithic) ISO 27001 compliance Aim to expand the org (x N)
83. Fracture plane: business domain
84. Optimise split for team engagement & personas Minimise ISO 27001 footprint Aided organisation expansion
85. When not to split a monolith •‘Heritage’ ERP system (‘cloudified’) •No native Unit Test framework •20-30 min startup times •VMs need 56GB RAM (yes) •CI builds take 50 mins
86. Monolith-splitting recipe Tried and tested!
87. Monolith-splitting recipe 1. Instrument the monolith – logging 2. Grok data flows and fault responses 3. Align teams to available segments 4. Split off segments one-by-one
88. Instrument the monolith
89. Instrument the monolith
90. Instrument the monolith
91. search by event Event ID {Delivered, InTransit, Arrived}
92. transaction trace Correlation ID 612999958…
93. Technical Domain public enum EventID { // Badly-initialised logging data NotSet = 0, // An unrecognised event has occurred UnexpectedError = 10000, ApplicationStarted = 20000, ApplicationShutdownNoticeReceived = 20001, PageGenerationStarted = 30000, PageGenerationCompleted = 30001, MessageQueued = 40000, MessagePeeked = 40001, BasketItemAdded = 60001, BasketItemRemoved = 60002, CreditCardDetailsSubmitted = 70001, // ... }
94. BasketItemAdded = 60001 BasketItemRemoved = 60002
95. Instrument the monolith Correlation ID Logs Event ID
97. use logging as a channel/vector to make distributed systems more testable use logging as a channel/vector to make distributed systems more testable
98. Grok data flows and fault responses
99. Grok data flows and fault responses Correlation ID Event ID Unexpected collaborating subsystems Undetected fault condition
100. Grok data flows and fault responses Correlation ID Event ID Adjust subsystem boundaries Fix poor fault responses
101. runbooktemplate.info Run Book dialogue sheets
102. Align teams to available segments
103. Align teams to available segments
104. Align teams to available segments Map to business domain
105. Align teams to available segments Identify likely components or ‘platform’ elements
106. Split off segments one-by-one
107. Split off segments one-by-one
108. Split off segments one-by-one Separate: - Builds - Infrastructure - Deployments - Versions - Lifecycle
109. Team needs / responsibilities / capabilities come first
110. use logging as a channel/vector to make distributed systems more testable Invest in Build & Release Engineering
112. Monolith-splitting recipe * 1. Instrument the monolith – logging 2. Grok data flows and fault responses 3. Align teams to available segments 4. Split off segments one-by-one (5. Invest in Build & Release Engineering) * The simplistic version
113. Monolith-splitting recipe * 1. Instrument 2. Grok behaviour 3. Align teams 4. Split off segments 5. Invest in Build & Release * The simplistic version
114. Monolith-splitting recipe * * The simplistic version 1. Instrument 2. Grok behaviour 3. Align teams 4. Split off segments 5. Invest in Build & Release
115. Determine monolith type (Apply ‘Code Forensics’) Find ‘fracture planes’ Assess cognitive load for teams Use the monolith-splitting recipe How to break apart a monolithic system safely without destroying your team
116. Further material
118. Books & articles • Working Effectively with Legacy Code, by Michael Feathers • Building Microservices by Sam Newman (O’Reilly, 2015) • Managing Cognitive Load for Team Learning by Jo Pearce http://12devsofxmas.co.uk/2015/12/day-3-managing-cognitive-load- for-team-learning/ • Continuous Delivery with Windows and .NET by Matthew Skelton and Chris O’Dell (O’Reilly, 2016) http://cdwithwindows.net/ • Team Guide to Software Operability by Matthew Skelton and Rob Thatcher (Skelton Thatcher Publications, 2016) http://operabilitybook.com/ • Ye Olde Monolith by Pter Pilgrim https://dzone.com/articles/ye-olde- monolith
119. Training • From Monolith to Microservices (online training) – Sam Newman, author of Building Microservices http://www.oreilly.com/live-training/from-monolith-to- microservices.html • Run Book template & Run Book dialogue sheets http://runbooktemplate.info/
120. Talks & slides • Hacking Your Head – Managing Information Overload, by Jo Pearce - http://www.slideshare.net/JoPearce5/hacking-your- head-managing-information-overload-45-mix / http://conferences.oreilly.com/oscon/open-source- eu/public/schedule/detail/53013 • Principles of Microservices by Sam Newman @ Devoxx 2015 https://www.youtube.com/watch?v=PFQnNFe27kU
121. Research papers • Driskell, James E., Eduardo Salas, and Joan Johnston. ‘Does Stress Lead to a Loss of Team Perspective?’ Group Dynamics: Theory, Research, and Practice 3, no. 4 (1999): 291. • Fan, Xiaocong, Po-Chun Chen, and John Yen. ‘Learning HMM-Based Cognitive Load Models for Supporting Human-Agent Teamwork’. Cognitive Systems Research 11, no. 1 (2010): 108–119. • Ilgen, Daniel R., and John R. Hollenbeck. ‘Effective Team Performance under Stress and Normal Conditions: An Experimental Paradigm, Theory and Data for Studying Team Decision Making in Hierarchical Teams with Distributed Expertise’. DTIC Document, 1993. http://oai.dtic.mil/oai/oai?verb=getRecord&metadataPrefix=html&identifier=ADA284683. • Johnston, Joan H., Stephen M. Fiore, Carol Paris, and C. A. Smith. ‘Application of Cognitive Load Theory to Developing a Measure of Team Decision Efficiency’. DTIC Document, 2002. http://oai.dtic.mil/oai/oai?verb=getRecord&metadataPrefix=html&identifier=ADA525820. • Sweller, John. ‘Cognitive Load Theory, Learning Difficulty, and Instructional Design’. Learning and Instruction 4 (1994): 295–312. • Sweller, John. ‘Cognitive Load during Problem Solving: Effects on Learning’. Cognitive Science 12, no. 2 (1988): 257–285.
122. Determine monolith type (Apply ‘Code Forensics’) Find ‘fracture planes’ Assess cognitive load for teams Use the monolith-splitting recipe How to break apart a monolithic system safely without destroying your team
123. thank you Matthew Skelton @matthewpskelton skeltonthatcher.com