shiny-data-sync

name: shiny-data-sync description: Guide for generating code that uses Shiny.Data.Sync for reliable, background-capable bidirectional JSON sync over HTTP on iOS, Android, Windows, Linux, macOS, and Blazor WASM auto_invoke: true triggers: - data sync - background sync - outbox - inbox - delta sync - DataSyncManager - IDataSyncManager - IDataSyncDelegate - SyncOperation - SyncEndpoint - ISyncEntity - Shiny.Data.Sync - offline-first - eventual consistency - conflict resolution - PullNow

Shiny Data Sync

Reliable, background-capable bidirectional JSON sync between a mobile/desktop app and an HTTP backend. Mirrors the platform-tier guarantees of Shiny.Net.Http (NSURLSession on Apple, Foreground Service on Android, HttpClient fallback elsewhere), but for structured records rather than files.

Provides an outbox for queued local changes (Create/Update/Delete with JSON payloads), an inbox for delta pulls from the server, retry with exponential backoff, pluggable conflict resolution, per-endpoint batching, and AOT-compatible JSON via JsonTypeInfo.

When to Use This Skill

Use this skill when the user needs to:

• Sync app entities to a REST backend reliably, even when offline or backgrounded
• Build an offline-first feature that eventually consistent-syncs to the server
• Queue Create/Update/Delete operations that must survive app kill / device reboot
• Pull server-side changes on a schedule (delta sync)
• Resolve sync conflicts (HTTP 409 / 412) with custom merge logic
• Batch multiple queued operations into a single server round-trip
• Persist a per-endpoint cursor for delta pulls

Do not use this skill for: large file uploads/downloads (use shiny-http-transfers), realtime data streams (use SignalR/MQTT), or push-driven sync (use shiny-push to trigger PullNow).

Library Overview

Setup

1. Define an Entity

Implement ISyncEntity on any record you want to sync:

using Shiny.Data.Sync;

public record TodoItem(string Identifier, string Title, bool Completed) : ISyncEntity;

2. Register Services

In MauiProgram.cs (reflection-friendly path):

using Shiny;

builder.Services.AddDataSync<MyDataSyncDelegate>(opts =>
{
    opts.RegisterEndpoint<TodoItem>("https://api.example.com/todos");

    opts.RegisterEndpoint<Project>("https://api.example.com/projects", ep =>
    {
        // -- Direction --
        ep.Direction = SyncDirection.Both;           // or PullOnly / PushOnly

        // -- Network policy --
        ep.UseMeteredConnection = false;             // wait for WiFi
        ep.Batch = true;                             // coalesce ops per round-trip
        ep.MaxAttempts = 8;                          // retry transient failures up to 8x
        ep.RetryBaseDelay = TimeSpan.FromSeconds(3);

        // -- Conflicts --
        ep.DefaultConflictPolicy = ConflictPolicy.ServerWins;

        // -- Inbox throttle --
        ep.MinPullInterval = TimeSpan.FromMinutes(5); // SyncJob/PullAll skip; PullNow bypasses

        // -- Per-verb URL overrides (optional) --
        ep.PullUrl = "https://api.example.com/projects/feed";  // GET different URL on pull
        ep.BatchUrl = "https://api.example.com/projects/bulk"; // POST batched ops elsewhere
        ep.CursorParameter = "updatedSince";          // default "since"

        // -- Tombstones (server-side delete stream) --
        ep.TombstoneUrl = "https://api.example.com/projects/deleted";
        ep.TombstoneCursorParameter = "since";

        // -- Soft-delete / expiry predicates --
        ep.SoftDeletePredicate = entity => entity is Project p && p.IsArchived;
        ep.ExpiryPredicate     = entity => entity is Project p && p.OwnerId == null;

        // -- Per-endpoint request hook (after ISyncInterceptor) --
        ep.OnBeforeSend = req =>
        {
            req.Headers.Add("X-Trace-Id", Guid.NewGuid().ToString("N"));
            return Task.CompletedTask;
        };
    });
});

// Global auth — runs before every endpoint's OnBeforeSend
builder.Services.AddSyncInterceptor<MyAuthInterceptor>();

// Centralize base address / Polly handlers on the named client
builder.Services.AddHttpClient(RestSyncTransport.HttpClientName, c =>
    c.BaseAddress = new Uri("https://api.example.com")
);

Native AOT / trimmed builds

All serialization runs through Shiny.Json.Default (the shared ISerializer from Shiny.Extensions.Serialization). Add the entity type to a context decorated with [ShinyJsonContext] — a source-generated module initializer wires it into the shared chain before any code runs:

[ShinyJsonContext]
[JsonSerializable(typeof(TodoItem))]
[JsonSerializable(typeof(Project))]
public partial class AppJsonContext : JsonSerializerContext;

The same context covers every endpoint that uses these types — no per-endpoint plumbing. To customize options globally use services.ConfigureJsonSerializer(opts => ...) from Shiny.Extensions.Serialization.

3. Implement the Delegate

using Shiny.Data.Sync;

public class MyDataSyncDelegate : IDataSyncDelegate
{
    public Task OnSent(SyncOperation op, string? responseBody)
    {
        // Server accepted the op. responseBody may contain a server-assigned id, new ETag, etc.
        return Task.CompletedTask;
    }

    public Task OnError(SyncOperation op, int statusCode, Exception ex)
    {
        // Op exhausted its retry budget. Persist to a dead-letter store or notify the user.
        return Task.CompletedTask;
    }

    public Task OnReceived(SyncReceivedItem item)
    {
        // The engine already deserialized item.Entity using the endpoint's JsonTypeInfo
        // or JsonOptions. RawPayload is also available for custom handling.
        if (item.Entity is TodoItem todo)
            myLocalStore.Apply(todo, item.Verb);
        return Task.CompletedTask;
    }

    public Task<ConflictResolution> OnConflict(SyncOperation op, string remotePayload)
    {
        // The server already has a newer version. Decide what to do.
        return Task.FromResult(ConflictResolution.AcceptRemote);
    }
}

Common Tasks

Queue an outbox operation

public class TodosService(IDataSyncManager sync)
{
    public Task CreateTodo(TodoItem item) => sync.Queue(SyncVerb.Create, item);
    public Task UpdateTodo(TodoItem item) => sync.Queue(SyncVerb.Update, item);
    public Task DeleteTodo(TodoItem item) => sync.Queue(SyncVerb.Delete, item);
}

Force a pull (pull-to-refresh)

public Task RefreshTodos(IDataSyncManager sync, CancellationToken ct)
    => sync.PullNow<TodoItem>(ct);

Observe outbox + inbox progress

// Typed events for fine-grained subscribers
sync.PendingCountChanged += (s, count) => StatusLabel.Text = $"{count} pending";
sync.UpdateReceived += (s, result) =>
{
    if (result.State == SyncOperationState.Error)
        ShowToast($"Sync failed: {result.Exception?.Message}");
};
sync.PullCompleted += (s, c) =>
{
    if (c.Error != null) ShowToast($"Pull failed for {c.EndpointKey}: {c.Error.Message}");
    else if (c.ItemsReceived > 0) ShowToast($"{c.ItemsReceived} new {c.EndpointKey} items");
};

// Unified Activity stream — covers every lifecycle moment
sync.Activity += (s, evt) =>
{
    Console.WriteLine($"{evt.Type} {evt.EndpointKey} items={evt.ItemCount} status={evt.StatusCode}");
};

Activity fires SyncEvent records for: OutboxQueued, OutboxStarted, OutboxSent, OutboxFailed, OutboxConflict, OutboxRetryScheduled, OutboxCanceled, InboxPullStarted, InboxItemReceived, InboxPullCompleted, InboxPullFailed, TombstonesApplied.

Cancel queued work

await sync.Cancel(operationId);              // one operation
await sync.CancelAll<TodoItem>();            // everything for one endpoint
await sync.CancelAll();                       // entire outbox (in-flight inbox pulls are left alone)

Platform Behavior

On iOS the JSON payload is serialized to a temp file (background NSURLSessions require file-backed uploads); the temp file is cleaned up when the operation completes or fails. The same background NSUrlSession carries both upload and download tasks (HttpMaximumConnectionsPerHost = 4).

A regained network connection on the HttpClient platforms (Windows / Linux / desktop) automatically triggers an outbox drain and a full PullAll via the IConnectivity.Changed event.

Retry policy

Every endpoint has MaxAttempts (default 5) and RetryBaseDelay (default 2s). Transient failures — HttpStatusCode 0 (network down), 5xx, 408, 429 — schedule a retry at baseDelay * 2^(attempts-1) capped at 60s. The retry timestamp is persisted on the SyncOperation as NextAttemptAt, so a process restart resumes the wait window correctly. After MaxAttempts, the op is handed to IDataSyncDelegate.OnError and removed from the outbox (the delegate can re-queue if it wants).

Conflict Resolution

When the server returns 409 Conflict or 412 Precondition Failed, the engine consults the endpoint's DefaultConflictPolicy:

• AskDelegate (default) — calls IDataSyncDelegate.OnConflict
• ServerWins — drops the local op, dispatches the remote payload through OnReceived as an Update
• ClientWins — re-queues the local op as-is

The delegate's OnConflict returns:

• ConflictResolution.AcceptRemote — same as ServerWins
• ConflictResolution.KeepLocal — re-queue the local op
• ConflictResolution.UseMerged(string mergedPayload) — replace the op's payload with a merge result and retry

Tombstones (separate delete stream)

Some servers can't merge deletes into the main pull. Set endpoint.TombstoneUrl and the engine follows every successful pull with a GET against that URL, expecting one of two shapes:

["id1", "id2", ...]

or, when the server paginates / cursors deletes separately:

{ "cursor": "<opaque next cursor>", "ids": ["id1","id2",...] }

Each ID dispatched to IDataSyncDelegate.OnReceived with Verb = Delete and Entity = null. A separate SyncTombstoneCursor record persists the tombstone cursor independently from the main SyncCursor. On iOS / Mac Catalyst the tombstone fetch also rides the background NSURLSession (tombstone:{endpointKey} task description), so it survives suspension just like the main pull.

Soft-delete and Expiry predicates

When a server signals deletes via a flag on the entity (IsDeleted = true) or via a state change that the client should treat as eviction (AssignedTo = null), point the engine at it:

ep.SoftDeletePredicate = e => e is Project p && p.IsArchived;
ep.ExpiryPredicate     = e => e is Project p && p.OwnerId == null;

Both run on the deserialized entity inside the inbox dispatch loop, before delegates fire. When either returns true for a Create/Update item, the verb is rewritten to Delete and Entity stays populated (so consumers can read the final state on the way out the door).

Direction (PullOnly / PushOnly / Both)

Set ep.Direction to restrict what's allowed:

• Both (default) — Queue + PullNow + PullAll all work
• PullOnly — Queue throws; the server is the source of truth
• PushOnly — PullNow throws, PullAll silently skips this endpoint; useful for telemetry / SyncUp queues

ISyncInterceptor (global request hook)

Per-endpoint OnBeforeSend handles endpoint-specific tweaks, but cross-cutting auth is better as a single ISyncInterceptor:

public class AuthInterceptor(ITokenService tokens) : ISyncInterceptor
{
    public Task BeforePull(SyncEndpoint endpoint, string? cursor, HttpRequestMessage req)
    {
        req.Headers.Authorization = new("Bearer", tokens.Current());
        return Task.CompletedTask;
    }

    public Task BeforePush(SyncEndpoint endpoint, IReadOnlyList<SyncOperation> ops, HttpRequestMessage req)
    {
        req.Headers.Authorization = new("Bearer", tokens.Current());
        return Task.CompletedTask;
    }

    // BeforeTombstoneFetch's default forwards to BeforePull — override only if your
    // tombstone URL is on a different auth domain.
}

builder.Services.AddSyncInterceptor<AuthInterceptor>();

All registered interceptors run before the per-endpoint OnBeforeSend, so endpoint hooks still win on header conflicts. Multiple interceptors are supported and execute in registration order.

Named HttpClient

The engine resolves its transport from IHttpClientFactory under the name RestSyncTransport.HttpClientName ("Shiny.Data.Sync"). Use the named-client pattern to attach base addresses, Polly handlers, and so on:

builder.Services
    .AddHttpClient(RestSyncTransport.HttpClientName, c =>
    {
        c.BaseAddress = new Uri("https://api.example.com");
        c.Timeout = TimeSpan.FromMinutes(2);
    })
    .AddPolicyHandler(GetRetryPolicy());

Inbox response shape

The default RestSyncTransport expects:

{
  "cursor": "<opaque next cursor>",
  "hasMore": false,
  "items": [
    { "id": "<entity id>", "verb": "Create|Update|Delete", "payload": { ... } }
  ]
}

If hasMore: true, the engine immediately re-pulls with the new cursor — drains the full delta set in one PullNow/PullAll call. HTTP 304 Not Modified is treated as "no changes" and just bumps LastPulledAt without touching the cursor. Apps that need a different shape can implement ISyncTransport directly.

Batching

Set endpoint.Batch = true to coalesce multiple queued ops for one endpoint into a single POST {url}/batch request. Coalescing rules:

• Trailing Delete wins — any preceding Create/Update for the same entity drop
• Create + Update(s) → single Create with the latest payload
• Update + Update(s) → single Update with the latest payload

Batch response shape expected from the server:

{
  "results": [
    { "id": "<op-id>", "status": 200, "body": { ... }, "error": null }
  ]
}

Batching applies to the HttpClient process path (Android Foreground Service, Windows, Linux, Blazor). The iOS path sends one NSURLSession upload task per op by design.

Custom Transports

Override ISyncTransport to use a non-REST protocol (gRPC, GraphQL, custom envelopes):

public class GrpcSyncTransport : ISyncTransport { /* ... */ }

builder.Services.AddDataSync<MyDelegate>(opts => { /* ... */ });
builder.Services.AddSingleton<ISyncTransport, GrpcSyncTransport>();   // overrides the default REST transport

Caveats

• Apple OnBeforeSend doesn't see the body. Uploads stream from disk on iOS / Mac Catalyst, so the stub HttpRequestMessage passed to OnBeforeSend has headers only. Signers that hash the body (AWS SigV4, etc.) won't work on Apple — use server-side validation or the Android/desktop path.
• Apple PullAll is fire-and-forget. The returned Task completes once tasks are kicked off, not when the downloads finish — observe PullCompleted for completion.
• Apple session shares the connection limit. Uploads and downloads share HttpMaximumConnectionsPerHost = 4; a saturated outbox queues inbox pulls behind it.
• Inbox in-flight requests are lost on process exit for HttpClient platforms. Only iOS / Mac Catalyst gets the survives-app-kill guarantee for pulls.