Cross-platform Data Synchronization

Cross-Platform Data Synchronization

Dan Grover

Wonder Warp Software LLC

Friday, October 16, 2009

Good morning. I’m going to talk today about how you can write your own cross-plaform data synchronization as part of your iPhone apps.

1

Outline 1 Why Syncing Is Important 2

Syncing Through The Ages and why you still might want to write your own

3 Algorithms & Architecture 4 Implementing Sync in Obj-C Friday, October 16, 2009

2

Here’s what we’re going to talk about today. - First, I want to persuade you why data synchronization is important, and why you might want to add it to your app. - Next, I’ll explain the ways data syncing has been solved by apps before, the advantages and disadvantages of various approaches, and explain why you may want to write your own. - Then we’re going to go over different algorithms that you can use to write your data synchronization code. I’m going to be very abstract and handwavy because it’s hard to talk about this kind of stuff when you’re also talking about implementation details. - Finally, we’ll dig in and talk about how to actually implement this stuff in Objective-C using the Cocoa APIs available to you.

Who I Am • Former Northeastern student • Independent software developer

Friday, October 16, 2009

3

How I Learned About Syncing

Friday, October 16, 2009

4

ShoveBox for Mac

Friday, October 16, 2009

I write an app called ShoveBox for the Mac. (describe)

5

Friday, October 16, 2009

6

Last November, I get an email from a friend of mine involved in the local Mobile Monday group here in Boston. They were going to do a fancy event at the Omni Parker House on “up and coming” mobile companies. Unfortunately, they couldn’t find enough up and coming mobile companies, so they asked me to present instead. “So do you have anything you would like to present?” At that time, I was mostly focused on Mac software -- I had a game out, but nothing much. So I said “Oh, of course, I can demo the new iPhone version of ShoveBox” Unfortunately, there was no iPhone version of ShoveBox. I didn’t really want to do one. It was kind of beyond the scope of the app. And syncing was HAAARRRD. So I made up a functional prototype of the iPhone app. I added a pretend dialog to the Mac app to show it syncing. I had a script that I used to convert the example data over so it looked the same.

? Friday, October 16, 2009

7

I actually *did* want to make the iPhone version for real, though. But I had no idea how it was going to work. I played around with a few half-way solutions -- storing the new entries and just propagating those on sync. But I realized that real, honest-to-god two-way syncing was doable if I just sat down and thought about it for a while. I studied all the ways that people are doing syncing and realized it wouldn’t be too hard to write my own from scratch. Sounds crazy.

Friday, October 16, 2009

A few months later, I finally ship the iPhone version. Sales quadruple, it gets two reviews in Macworld. Still some bugs with syncing, but eventually those get ironed out.

8

Quick Demo

Friday, October 16, 2009

9

Outline 1 Why Syncing Is Important 2

Syncing Through The Ages and why you still might want to write your own

3 Algorithms & Architecture 4 Implementing Sync in Obj-C Friday, October 16, 2009

10

1 Why Syncing is Important Friday, October 16, 2009

I’m going to get on my soapbox for a moment and explain briefly why I think this is an important topic, and how it’s applicable to more apps than you’d think.

11

Friday, October 16, 2009

Syncing has been something people have been trying to solve for a long time. If you follow the current hype, we don’t have to worry about it because...

12

the

CLOUD

Friday, October 16, 2009

...you put everything on the cloud! The cloud will solve all our problems! The popular conception of the trend of “cloud computing” is a little wrong. People think of it as a monolithic thing.

13

Friday, October 16, 2009

14

But the reality is that the huge benefit of cloud computing is that you can outsource the right things to the right people. I use one company for sending my email newsletters, because they have the best infrastructure and software for that. I use another for my regular web hosting, and yet another to host downloads. And I use a help desk app called Zendesk. So it’s not really on “the cloud” -- it’s on a lot of clouds! So we’re back to the same problem -- data is going to be in a ton of different places, and you have to build systems that can deal with that. Sync plays a big part.

A

CLOUD A

CLOUD A

CLOUD

A

CLOUD

Friday, October 16, 2009

So the future’s more complicated than it seems. It’s not “the cloud”, but lots of clouds and client apps and platforms and apparently goats. And they all have to be share data but operate independently.

15

Does your app pass the Green Line Test?

Friday, October 16, 2009

And if you don’t think data synchronization applies to your app, I’d like you all to try this while you’re in the city. I call it the Green Line Test.

16

Friday, October 16, 2009

17

I used to live near Lechmere in East Cambridge, and I’d commute in to classes at Northeastern using the Green Line. The Green Line touches a lot of areas of Boston and goes above ground and below. Some of the stations underground are dead, some have reception. Inevitably, the ones that the train stops inexplicably for 20 minutes in will be those that don’t. You see, they’ve upgraded all the trains and haven’t quite got all the kinks worked out. If your app is one of those “thin” or “hybrid” apps that needs to make an HTTP request to do anything, you should try running your app for the entirety of a Green Line ride. How does it handle it when you lose connectivity for a minute? Pop up an error? Or stall indefinitely? How good an experience is it? Do you cache things well, or does it always need a connection? If you find that it’s not very good in this situation, you should consider making more of your application operate on the device itself, and then sync its state back to the cloud. It will be more responsive and usable more of the time. You’ve probably avoided something like this because, well, syncing is a pain. But what I’m going to talk about in this presentation will help.

2 Syncing Through The Ages and

Why You Still Might Want To Write Your Own Friday, October 16, 2009

18

Friday, October 16, 2009

I thought this tweet from Steven Frank was funny. It’s true. It never works. I think that’s because there’s not a lot of knowledge about syncing out there. There are a lot of companies that have written (bad) syncing, and a few academic papers on it. But not a lot of talk about syncing as a subject. If more people didn’t have to waste all this time learning the basics for themselves, we could have better syncing as more people work out the kinks and integrate it in more systems.

19

Set-Reconciliation Problem

Friday, October 16, 2009

20

Academics call syncing the “set reconciliation problem”. You’ve got two sets, and you want to reconcile their differences. The literature on it is pretty limited though.

rsync Friday, October 16, 2009

21

Subversion

Friday, October 16, 2009

Subversion is a kind of syncing a lot of us probably use every day. Like most version control systems, the idea is that your whole team can have the most current copy of the code.

22

Data ≠ Files

Friday, October 16, 2009

But it’s important to note that there’s a big difference between syncing *data* and syncing *files*. Syncing data is a LOT harder!

23

DropBox Friday, October 16, 2009

Dropbox is a consumer file syncing solution. But it actually ends up working a lot more like Subversion than you’d think. It keeps revisions and actually handles conflicts in a neat way.

24

HotSync Friday, October 16, 2009

Palm was one of the first companies to try to make a comprehensive syncing solution for consumers. The way HotSync works is that, once you’ve done the first sync, the Palm would set these status flags on any piece of data that you changed. That would make it really fast to sync back up with your PC, because the PC had an old copy of the data that both devices had the last time you synced.

25

Mac OS X

Sync Services Friday, October 16, 2009

Sync Services is Apple’s syncing framework. It’s pretty neat, and if you were like me and trying to write a Mac app that synced with an iPhone app, it would *almost* work.

26

Sync Services

Your App

Truth Database Macs

Friday, October 16, 2009

27

Sync Services has this concept of a “Truth Database” -- where you replicate all your data so that it can sync it elsewhere. It gives you lots of goodies to sync your app to the Truth database -- pushing and pulling changes. They give you tools to define the schema you want the Truth to keep for your data. But then it gets magically put on MobileMe and synced to other Macs. You don’t have any control over that. The iPhone supports MobileMe, but only for syncing contacts, appointments, and notes. It doesn’t read in the truth database from Sync Services, it’s totally separate. There is no Sync Services for the iPhone. So that’s kind of a bummer.

Two Approaches: History-Based Ex-Post-Facto Friday, October 16, 2009

28

History-Based

Ex-Post-Facto

PROS

PROS

- Efficient and accurate

CONS

- All client software must maintain status flags/history - Does not scale as well - Complicated

Friday, October 16, 2009

- Easy to bolt onto an existing system - Hot swappable: arbitrary configurations of devices in any state can be synced CONS

- Syncing can be slower - Requires accurate date/time info

29

History-Based

Ex-Post-Facto

Subversion Dropbox HotSync (Fast)

Rsync Sync Services HotSync (Slow)

Friday, October 16, 2009

30

When To Write Your Own • When your schema demands custom handling • Dependencies • Ordering

• When data needs to be specially converted and prepared for different clients/devices • iTunes and iPod Shuffles

• When it’s a core function Friday, October 16, 2009

31

3 Algorithms and

Architecture Friday, October 16, 2009

32

A

Friday, October 16, 2009

A∩B

B

33

So in these algorithms, we’re going to be a little abstract and think of this as two sets of data. - A is all the data that’s on your first device, B is all the data that’s on the second device. - Here’s all the data that’s *only on A*. That needs to be put on B if it was added, deleted from A if not. - Here’s all the data that’s *only on B*. That needs to be put on A if it was added, deleted from B if not. - Here’s the data that’s on both. This is the trickiest part. We need to sift through this data and figure out if any of it has been modified since the last sync. We need to merge modifications when we can, and otherwise, ask the user to resolve the conflict.

Friday, October 16, 2009

34

Goal of a Sync Algorithm Make Two Sets The Same (duh!) ... in a way consistent with user expectations ... as quickly as possible

Friday, October 16, 2009

So what is the goal of any sync algorithm? To make both sets of data the same. Well, that part is pretty easy. I could just erase what’s on your server account and erase what’s on your iPhone. Done! Turns out it’s more complicated. There are a lot of *correct* ways to make this happen, but only some of them are what the user is expecting to see. The sync also has to be fast. This usually means a minimum of data being transferred.

35

Three Algorithms Copy

Friday, October 16, 2009

Sync

Merge

36

But there are a few ways to skin a cat. Let’s look at each of these. They all meet the definition we discussed, but go about it differently.

Copy A

B

Good Will Hunting

Good Will Hunting

The Departed

Spenser: For Hire

21

The Boondock Saints With Honors

Friday, October 16, 2009

37

Copy A

B

Good Will Hunting

Good Will Hunting

The Departed

The Departed

21

21

Friday, October 16, 2009

38

Merge A

B

Good Will Hunting

Good Will Hunting

The Departed

Spenser: For Hire

21

The Boondock Saints With Honors

Friday, October 16, 2009

39

Merge A

B

Good Will Hunting

Good Will Hunting

Spenser: For Hire

Spenser: For Hire

The Departed

The Departed

The Boondock Saints

The Boondock Saints

21

21

With Honors

With Honors

Friday, October 16, 2009

40

Sync A

last sync = 12PM now = 3PM

Good Will…

created: 11AM modified: 11AM

The Departed

created: 2PM modified: 2PM

21

created: 11AM modified: 11AM

Friday, October 16, 2009

B Good Will…

created: 11AM 2PM created: modified: 11AM 2PM modified:

Boondock…

created: 1PM modified: 1PM

With Honors

created: 2PM modified: 2PM

41

Sync A

last sync = 12PM now = 3PM

Good Will…

created: 11AM modified: 11AM

The Departed

created: 2PM modified: 2PM

Boondock…

created: 1PM modified: 1PM

With Honors

created: 2PM modified: 2PM

Friday, October 16, 2009

B Good Will… The Departed

created: 11AM 2PM created: modified: 11AM 2PM modified:

created: 2PM modified: 2PM

Boondock…

created: 1PM modified: 1PM

With Honors

created: 2PM modified: 2PM

42

Three Algorithms Copy

Sync

Merge

Friday, October 16, 2009

So let’s go back here and talk about when to use each of these algorithms: SYNC: This is what you’re going to want to do 95% of the time. The other two algorithms are for when you’re first setting two devices up to sync. COPY: Some people doing sync like to offer you a choice of data on either device to become the “one true” set of data. MERGE: What I do with ShoveBox is just do a merge the first time -- because there might be data on both devices they want to keep. It avoids any confusion over the choice, and nobody’s going to be pissed with the initial result.

43

Needed for Sync • On each device, each object needs: • Creation Date • Modification Date • UDID Friday, October 16, 2009

44

Sync: In Depth PREPARE SYNC OBJECTS IN ONLY A SYNC OBJECTS IN ONLY B SYNC INTERSECTION CLEAN UP Friday, October 16, 2009

45

Sync: In Depth PREPARE SYNC OBJECTS IN ONLY A SYNC OBJECTS IN ONLY B SYNC INTERSECTION CLEAN UP Friday, October 16, 2009

46

PREPARE Establish Communication With Sources

•

Grab summaries from A and B UUIDs, creation, modification

•

•

Sort into sets

•

Friday, October 16, 2009

47

Sync: In Depth PREPARE SYNC OBJECTS IN ONLY A SYNC OBJECTS IN ONLY B SYNC INTERSECTION CLEAN UP Friday, October 16, 2009

48

SYNC OBJECTS IN ONLY A •

For each object o in a: if o.creation > last sync then tell b to copy o over else tell a to delete o end if next •

•

•

•

•

•

Friday, October 16, 2009

49

Sync: In Depth PREPARE SYNC OBJECTS IN ONLY A SYNC OBJECTS IN ONLY B SYNC INTERSECTION CLEAN UP Friday, October 16, 2009

50

SYNC OBJECTS IN ONLY B •

For each object o in b: if o.creation > last sync then tell a to copy o over else tell b to delete o end if next •

•

•

•

•

•

Friday, October 16, 2009

51

Sync: In Depth PREPARE SYNC OBJECTS IN ONLY A SYNC OBJECTS IN ONLY B SYNC INTERSECTION CLEAN UP Friday, October 16, 2009

52

SYNC INTERSECTION •

For each object o in both a and b: if o.modification < last sync then skip it else if only a’s mod > last sync then propogate a’s version to b else if only b’s mod > last sync then propogate b’s version to a else if both a and b’s mod > last sync then present conflict end next •

•

•

•

•

•

•

•

•

•

•

Friday, October 16, 2009

53

Sync: In Depth PREPARE SYNC OBJECTS IN ONLY A SYNC OBJECTS IN ONLY B SYNC INTERSECTION CLEAN UP Friday, October 16, 2009

54

CLEAN UP • •

Friday, October 16, 2009

tell a and b we’re finished store current time as last sync

55

What’s wrong with this? 1. Single last-sync date can cause problems with partial syncs. Sync engine keeps per-item last-sync dates SOLUTION

2. Single modification date makes merging hard Keep per-attribute modification dates on each source SOLUTION

Friday, October 16, 2009

56

INTERSECTION REVISITED •

else if both a and b’s mod > last sync then let c = new list of conflicting keys let e = new entry record • •

• •

for each key k on o if a[o].k == b[o].k then e.k = a[o].k else if only a[o].k.mod > o.last sync then e.k = a[o].k else if only b[o].k.mod > o.last sync then e.k = b[o].k else c += k end if end if next •

•

•

•

•

•

•

•

•

•

Friday, October 16, 2009

•

• • • •

if c.count > 0 then present conflict to user e=a|b end if

push e to a and b next entry

57

Going Further

• On textual keys, if the same key on the

same entry was modified on both entries, then use diff to do a text merge and

• only ask the user to select one

version or the other if there is a text merge conflict

Friday, October 16, 2009

58

Architecture

Friday, October 16, 2009

59

Architecture

Syncer

Friday, October 16, 2009

60

Architecture

Syncer

A

B

Source

Source

Friday, October 16, 2009

61

Architecture

Syncer

A

B

LocalSource SQLLite DB

Source Web Service

Friday, October 16, 2009

62

Architecture

iPhone App

Web Service Friday, October 16, 2009

63

Architecture

The Cloud

iPhone App

Web Service Friday, October 16, 2009

64

Architecture Mac App

Friday, October 16, 2009

iPhone App

65

Architecture

Syncer

A

B

Source

Source

Friday, October 16, 2009

66

Sync Source Abstraction

• A sync source supports: • Create/Overwrite Object • Delete Object • Get Object • Get summary Friday, October 16, 2009

67

4 Implementing Sync in Objective-C

Friday, October 16, 2009

68

UDIDs example:

DBCE017A-AF95-11DE-98BE-228156D89593 how to generate: CFUUIDRef uuid = CFUUIDCreate(kCFAllocatorDefault); CFUUIDCreateString(kCFAllocatorDefault,uuid);

Friday, October 16, 2009

69

Dates • NSDate contains time zone info • You can compare two NSDate objects or two timestamps

• UNIX Timestamp (1970) • NSDate Timestamp (2001) Friday, October 16, 2009

70

Syncing with CoreData • Set modification date in -willSave • Check -isUpdated and -  changedValues

• Don’t update the mod date if it’s just the mod date that changed.

• Set creation date, mod date, and GUID in -awakeFromInsert

Friday, October 16, 2009

71

Networking • Protocol choices: • HTTP • GameKit • BEEP/BLIP-based protocol • Roll your own (not recommended) • Using Bonjour/ZeroConf Friday, October 16, 2009

You have a few choices for your protocol. If you’re communicating with a server, you can make yourself a web service API. Your sync source is just wrapping code that makes NSURLRequests. I made the unfortunate choice of using it locally over the network. Writing an HTTP server that just has to talk with one other device isn’t too hard, but it was a really dumb architectural decision. Routers like to screw with it, even when it’s on a non-standard port.

72

Some (Bad) Syncing Code from My App

Friday, October 16, 2009

73

ShoveBox Mac App SBSyncEngine SBSyncSource

SBIPhoneSyncSource

Friday, October 16, 2009

SBLocalDBSyncSource

74

- (id) initWithLastSyncDate:(NSDate *)lastSync sourceA:(NSObject<SBSyncSource> *)a sourceB:(NSObject<SBSyncSource> *)b operation:(SBSyncEngineOperation)newOperation; - (IBAction) start:(id)sender - (IBAction) cancel:(id)sender; - (NSDate *) lastSyncDate; - (NSString *) currentlySyncingObjectName; - (SBSyncEngineOperation) operation; - (NSObject<SBSyncSource> *) sourceA; - (NSObject<SBSyncSource> *) sourceB; - (NSObject<SBSyncEngineDelegate> *) delegate; - (void) setDelegate:(NSObject<SBSyncEngineDelegate> *)theDelegate;

Friday, October 16, 2009

75

typedef enum SBSyncEngineOperation { ! SBSyncEngineOperationSync = 0, // Time-based sync A and B ! SBSyncEngineOperationMerge = 1, // Non-destructive merge between A and B ! SBSyncEngineOperationCopy = 2, // Replace B’s contents with A’s } SBSyncEngineOperation;

Friday, October 16, 2009

76

@protocol SBSyncEngineDelegate - (void) syncEngineFinishedSyncingSuccesfully:(SBSyncEngine *)syncEngine; - (void) syncEngineDidCancel:(SBSyncEngine *)syncEngine; - (void) syncEngine:(SBSyncEngine *)syncEngine abortedWithError:(NSError *)err; - (BOOL) syncEngine:(SBSyncEngine *)syncEngine pausedWithRecoverableError:(NSError *)err; // return YES to continue, NO to cancel - (void) syncEngine:(SBSyncEngine *)syncEngine syncedObjects:(NSUInteger)objects ofTotal:(NSUInteger)total; // return the index of the correct choice - (NSUInteger) syncEngine:(SBSyncEngine *)syncEngine encounteredEntryConflictWithA:(NSDictionary *)aEntryInfo b:(NSDictionary *)bEntryInfo; - (NSUInteger) syncEngine:(SBSyncEngine *)syncEngine encounteredFolderConflictWithA:(NSDictionary *)aFolderInfo b:(NSDictionary *)bFolderInfo; - (NSUInteger) syncEngine:(SBSyncEngine *)syncEngine encounteredSimpleEntityConflictWithKeyPath:(NSString *)keyPath aValue:(id)aValue bValue:(id)bValue; @end Friday, October 16, 2009

77

Questions/Discussion

Friday, October 16, 2009

78