Offline-First: Architektur-Patterns & Implementierung für APIs & Backends

"Keine Internetverbindung" – ein Satz, den wir alle aus vielen Apps kennen, sei es Desktopanwendung oder Smartphone App. Oftmals sind die Clients auf dem jeweiligen Zielgerät installiert, bedienen sich aber einer externen Datenquelle, bspw. in Form eines HTTP APIs. Wie man sein Backend um eine offline-fähige Architektur erweitert, um es bspw. von einer SPA/PWA nutzen zu können, zeigen Thomas Hilzendegen und Manuel Rauber von Thinktecture in diesem Workshop. Denn bei einer offline-fähigen Implementierung sind nicht nur Lese-, sondern vor allem auch Schreibzugriffe wichtig. Sprich, die Daten müssen in beide Richtungen synchronisiert werden können, sobald die Internetverbindung wiederhergestellt wurde. Allgemein werden Themen im API Design, u.a. Synchronisationsstrategien von Anwendungs- und großen Binärdaten, Konfliktmanagement oder Datenbankarchitektur – sowohl für neue als auch bestehende Anwendungen – diskutiert. Am Beispiel einer PWA mit einem Backend in Java als auch in .NET Core werden Client und Server zur Umsetzung der besprochenen Konzepte und Patterns gezeigt.

GitHub: https://github.com/thinktecture/thinktecture-boardist

Transcript

1. Thomas Hilzendegen
   @hilzendegent
   Consultant
   Manuel Rauber
   @ManuelRauber
   Consultant
   Architektur-Patterns & Implementierung
   Offline First APIs & Backends
   

   View Slide

2. Thomas Hilzendegen
   Speakers
   Manuel Rauber
   thomas.hilzendegen@thinktecture.com
   @hilzendegent
   manuel.rauber@thinktecture.com
   @manuelrauber
   https://manuel-rauber.com
   Microsoft MVP
   

   View Slide

3. Timetable
   Time Doing
   09:00 - 10:30 Part I
   10:30 - 11:00 ☕
   11:00 - 12:30 Part II
   

   View Slide

4. • Progressive Web App
   • Offline availability
   • Offline synchronization
   • Change tracking
   • Security considerations
   • Pitfalls
   Agenda
   

   View Slide

5. Intro
   

   View Slide

6. Offline… but why?
   

   View Slide

7. • Just because the phone has a connection,
   e.g. Edge, it does not mean we are online
   • Depends on the use case which connection
   quality is required to determine, if the
   app is reliable online
   • Connection quality could be measured
   by the time an exclusive request takes
   • Duration < 150 ms: online
   • Duration !>=150 ms: bad connection
   What does “offline” mean?
   

   View Slide

8. • A lot of reasons to be offline
   • Traveling, Train, Flights (bad or no signal)
   • Server is not available
   • Routing problems
   • Roaming
   • Costs
   Motivation
   

   View Slide

9. Demo Application
   

   View Slide

10. • A little application to manage boardgames
    • Angular, .NET Core, MS SQL Server
    • Brownfield application, started as a pure online application
    • As typical for any good demo: no security
    • Available on Azure: https://tt-boardist.azurewebsites.net
    • GitHub: https://github.com/thinktecture/thinktecture-boardist
    Demo Application “Thinktecture Boardist”
    

    View Slide

11. • Backend
    • .NET Core 3.0
    • SQL Server
    • Frontend
    • Angular 8.2.13
    Demo Application Specs
    

    View Slide

12. Demo Application Database Scheme
    

    View Slide

13. Live Demo
    

    View Slide

14. Now let’s go offline!
    

    View Slide

15. Does it work in flight mode?
    NOPE!
    

    View Slide

16. Progressive Web App
    

    View Slide

17. Live Demo
    

    View Slide

18. Does it work offline now?
    STILL NOPE!
    

    View Slide

19. Real offline availability
    

    View Slide

20. • ServiceWorker is only able to take data offline which has been requested by
    the application
    • If all the URLs are known beforehand, the ServiceWorker could cache them
    all
    • Data which was not requested, is not available offline (no real offline
    synchronisation)
    • But what about … the Background Sync API?
    What’s the problem?
    

    View Slide

21. • Name is misleading
    • Does not offer any data synchronisation possibilities
    • Just raises an event periodically or after a delay (optionally based on the
    network type)
    • It’s totally up to the developer what to do in the onsync event
    • Since the sync is done in the ServiceWorker, the page can be closed, the
    sync will be fulfilled anyway
    • It does not help anything with syncing your actual data!
    Background Sync API
    

    View Slide

22. Background Sync API
    

    View Slide

23. • Cookies (not meant for large data or binaries)
    • Web Storage like session storage or local storage (not meant for large data
    or binaries)
    • IndexedDB
    • Cache Storage (based on request/response)
    Storage capabilities
    

    View Slide

24. User can clear anything
    at any time by
    “Remove temporary internet files”
    or “Clear website data”
    

    View Slide

25. • Key-value database of the browser
    • Stores data permanently
    • ServiceWorker and Web App share access to the same IndexedDB
    • Possibility of scenarios, where the ServiceWorker (or Web App) stores
    synchronised data in the IndexedDB and the Web App reads the data
    • Like everything based on the origin
    IndexedDB
    

    View Slide

26. IndexedDB Availability
    

    View Slide

27. • The standard API of IndexedDB is inconvenient to use (lots of callback)
    • Dexie.js is a minimalistic wrapper for IndexedDB
    • Operations are promise-based
    • Near native performance (even for bulk inserts)
    • Open Source @ GitHub: https://github.com/dfahlander/Dexie.js
    IndexedDB API
    

    View Slide

28. Offline Sync
    

    View Slide

29. • Offline Sync means to download all data available to client into a persistent
    offline storage, without the user having to explicitly request the data
    • Depending on the scenario, client can do CRUD on the offline data
    • Data will be synced back to the server, whenever a connection is possible
    Offline Sync Basics
    

    View Slide

30. • Online/offline recognition
    • Conflict management
    • Binary data
    • Local changes
    • Deleting data
    • Update interval (incoming new data)
    • Error handling
    • Primary key generation
    Offline Sync Challenges
    

    View Slide

31. • Having Edge may not mean that we are online
    • Being online could depend on several factors
    • Connection quality (Edge, 3G, 4G, 5G)
    • Connection speed (latency to your backend)
    • Reachability of backend systems (database, 3rd party system, storages)
    Offline Sync Challenges - Online/offline recognition
    

    View Slide

32. • Client needs to be online for write operations
    • Locks the data, so no other client can overwrite it
    • Data stays locked, until the client either saves or discards changes
    • Last One (Write) Wins
    • Visual conflict management (diffing like in Git, SVN, etc.)
    Offline Sync Challenges - Conflict management
    

    View Slide

33. • Downloading large binary data may not be possible on all platforms
    • A browser does not have a big enough storage for storing the data
    • Native apps (Cordova, Electron) could be a solution to directly access the
    platform’s file system
    • Uploading large binary data
    • Server needs to support chunk based upload with connection interruptions
    • Client needs permanent access to the binary file being uploaded, which
    could be hard in a browser-only scenario
    Offline Sync Challenges - Binary data
    

    View Slide

34. • User opens an edit form and makes changes to data
    • Incoming sync from server would change the data the user is editing
    • Depending on your use-case, decide what to do
    • Inform the user about the data change?
    • Show a visual diff?
    • Override the local changed data?
    • “Do nothing” and override the server data on next sync
    Offline Sync Challenges - Local changes
    

    View Slide

35. • Mark “deleted” data with a flag only
    • Use a trigger to keep the deleted ID in a separate table
    • Client needs some info about “deleted” data
    • Simple list of IDs “deleted” since the last sync
    Offline Sync Challenges - Deleting data
    

    View Slide

36. • Decide when you need to sync which data
    • Not all data needs to be synced in the same interval
    • An additional real time connection (WebSocket) could be established
    • The server can send a signal when some data needs to be synchronised
    Offline Sync Challenges - Update interval
    

    View Slide

37. • Show it, don’t hide them!
    • The user needs to know that something has occurred which shouldn’t
    • If you can’t sync to the client anymore, you may need to re-sync everything
    • Depending on your use case, you may want to write all data in one
    transaction on the server to not have partial data written into the database
    Offline Sync Challenges - Error handling
    

    View Slide

38. Primary key generation
    

    View Slide

39. • Inserted offline data needs to be available in the client immediately (e.g. the
    users reloads the application)
    • Behaves like “real synced data”, even if it was not written to server yet
    • Client is able to route to offline inserted data
    • Therefor it needs some kind of primary key/id to route to
    • Primary key can be generated on server-side and on client-side
    Primary key generation - challenge
    

    View Slide

40. • Server is responsible for generating the primary key
    • Client will generate a temporary offline primary key
    • After syncing, the client needs to replace the temporary offline key with the
    server generated primary key
    • Allows to use sequential GUIDs on the server side (MSSQL)
    Primary key generation - server-side
    

    View Slide

41. • Client is responsible for generating the primary key (GUIDs only)
    • Server inserts the client generated key into the database as the primary key
    • Server additionally needs to have a clustered key
    • Otherwise inserting a client non-sequential GUID will lead to a
    reclustering of the database, every time a client syncs new data
    Primary key generation - client-side
    

    View Slide

42. Change Tracking
    

    View Slide

43. • Possibilities to determine the client’s current state
    • Use highest rowversion within result
    • Determine by result set or additional query
    • Result of @@DBTS
    • Returns the last used rowversion of the database
    • Result of MIN_ACTIVE_ROWVERSION()
    • Returns the lowest (active) rowversion of the database
    Change Tracking - MS SQL Server’s ROWVERSION
    

    View Slide

44. • If you do a million updates a second, every second, the timestamp will wrap
    around in about 585000 years
    • Could be used for concurrency checks as well (see Conflict Management)
    • Before writing data into a row, you could check if it still as the client’s
    saved row version
    • If yes: data was not modified, saving is easily possible
    • If no: data was modified, decide per use case what to do!
    Change Tracking - MS SQL Server’s ROWVERSION
    

    View Slide

45. Change Tracking - Highest ROWVERSION within result
    Transaction #1
    ID ROWVERSION
    1 0x01
    2 0x02
    … …
    41 0x41
    Transaction #2
    ID ROWVERSION
    1 0x01
    2 0x02
    … …
    41 0x41
    42 0x42 43 0x43
    SELECT MAX(ROWVERSION) FROM TABLE
    Client’s state
    Last known rowversion 41
    43
    NOT RELIABLE!
    

    View Slide

46. • A later started transaction but committed first results in a queryable
    rowversion higher than the pending ones
    • Data will be lost (never seen by the client’s delta request)
    Change Tracking - Highest ROWVERSION within result
    

    View Slide

47. Change Tracking - Query @@DBTS
    Transaction #1
    ID ROWVERSION
    1 0x01
    2 0x02
    … …
    41 0x41
    Transaction #2
    ID ROWVERSION
    1 0x01
    2 0x02
    … …
    41 0x41
    42 0x42 43 0x43
    SELECT @@DBTS
    Client’s state
    Last known rowversion 41
    42
    43
    NOT RELIABLE!
    

    View Slide

48. • Value is database global
    • Running transaction updates the value instantly (before commit or rollback)
    • Data will be lost (never seen by the client’s delta request)
    Change Tracking - Query @@DBTS
    

    View Slide

49. Change Tracking - Query MIN_ACTIVE_ROWVERSION()
    Transaction #1
    ID ROWVERSION
    1 0x01
    2 0x02
    … …
    41 0x41
    Transaction #2
    ID ROWVERSION
    1 0x01
    2 0x02
    … …
    41 0x41
    42 0x42 43 0x43
    SELECT MIN_ACTIVE_ROWVERSION()
    Client’s state
    Last known rowversion 41
    42
    44
    RELIABLE!
    

    View Slide

50. • Value is database global
    • Running transaction updates the value instantly (before commit or rollback)
    • Data will be safe
    • Needs to be compared with greater or equal
    • May query data multiple times
    Change Tracking - Query MIN_ACTIVE_ROWVERSION()
    

    View Slide

51. Backend Preparation
    

    View Slide

52. • All syncable entities need to have rowversion column
    • rowversion is updated by MS SQL Server automatically whenever the row
    is changed (created & updated)
    • For deleted entities
    • Either set a IsDeleted flag to true (never delete any rows physically)
    • Or save the deleted IDs of the entities somewhere else (by trigger)
    Backend Preparation - MS SQL Server
    

    View Slide

53. • Use equivalents of rowversion and triggers
    • Manual implement mechanism in business logic (error-prone!)
    • Update tracking column manually by incrementing a database global
    number (during one transaction!)
    • Will be very hard für multi-row updates/inserts
    • Manual implement mechanism in triggers (if available)
    Backend Preparation - Other Database Systems
    

    View Slide

54. Frontend Preparation
    

    View Slide

55. • Choose storage area for data (e.g. IndexedDB)
    • Write all the code
    • Periodic data synchronization
    • Binary synchronization when data changes
    • Tracking of timestamps
    Frontend Preparation
    

    View Slide

56. Security
    

    View Slide

57. Typical token-based security
    Security - API level
    Browser
    Identity
    Provider
    Web API
    

    View Slide

58. • User is only able to see data based on his security level
    • Permissions
    • Roles
    • Policies
    • What happens, if the user permission change is leading to different data
    visible to the user?
    • What about data, which he does not see due to rights, but is connected to
    other data?
    • Remove frontend data after logout
    Security - Data Level
    

    View Slide

59. Pitfalls
    

    View Slide

60. • De-normalize the relational data (document style)
    • Results in multiple rowversion for one entry (use most recent one)
    • Multiplies the data (more traffic)
    • Consistent data for one entry
    • Keep the relations up to the frontend
    • Explicit rowversion for each entry
    • Partial consistency (related data may not be synced yet)
    Relational Data
    

    View Slide

61. • Keep some relational data and de-normalize some of it
    • De-normalize many-to-many relations
    • Needs trigger or business logic to change main entry’s rowversion when
    relation changes
    • Partial consistency (related data may not be synced yet)
    Relational Data
    

    View Slide

62. • PWA helps taking offline the application, but not the data
    • Online != Edge is available
    • Offline data may be “temporary” (“Remove temporary internet files”)
    • Use MIN_ACTIVE_ROWVERSION() and “greater or equal than” operator
    • Think about data level security (permissions, roles, etc.)
    • Think about conflict management
    • Depends heavily on your use case
    Summary
    

    View Slide

63. And… now?
    • Slides: https://speakerdeck.com/manuelrauber
    • Repository: https://github.com/thinktecture/thinktecture-boardist
    

    View Slide