broadcast_queue_shim_for_ndk

Offline-first wrapper around the ndk package's broadcast use case.

NDK's broadcast sends a Nostr event to a set of relays and reports per-relay results. If every relay is unreachable (flaky network, app backgrounded, process killed), the event is gone from the caller's perspective. This shim sits in front of ndk.broadcast and adds:

  • Local persistence first. The event is committed to a sembast store before any network attempt. broadcast() returns once persistence is durable; delivery happens in the background and survives restarts.
  • 100 % delivery guarantee. An entry is only marked delivered once every targeted relay has returned broadcastSuccessful: true. Partial success keeps the entry pending and retries the missing retryable relays.
  • Monotonic ack history. A relay that has acked never un-acks. A delivered entry never silently flips back to pending due to a transient relay outage.
  • Terminal relay rejections. NIP-01 OK false replies with pow, blocked, invalid, restricted, or error stop retries for that relay/event pair and are exposed through BroadcastStatus.failed.
  • Dead relay cutoff. A relay that stays inaccessible for 12 consecutive online attempts is also stopped for that event.
  • No auto-deletion. Delivered entries stay in the store and can be re-broadcast later, for instance to a freshly discovered relay.
  • Relay sets resolved offline. Target an account's NIP-65 outbox, its recipients' inboxes, NIP-17 DM relays or NIP-37 private relays without knowing the URLs yet. The set is stored as is and resolved by the worker once online, then frozen.
  • Account-scoped clearing. Attribute an entry to an account with the optional pubkey argument, then wipe just that account's queue on logout via clearLocalAccountData.

Quick start

import 'package:broadcast_queue_shim_for_ndk/broadcast_queue_shim_for_ndk.dart';
import 'package:ndk/ndk.dart' hide RelaySet;
import 'package:sembast/sembast_io.dart';

Future<void> main() async {
  final db = await databaseFactoryIo.openDatabase('broadcasts.db');

  final ndk = Ndk(
    NdkConfig(
      eventVerifier: Bip340EventVerifier(),
      cache: MemCacheManager(),
    ),
  );

  final outbox = OfflineBroadcast.withNdk(ndk, db: db);
  outbox.start();

  final event = Nip01Event(
    pubKey: myPubKey,
    kind: 1,
    tags: const [],
    content: 'hello from a flaky network',
  );

  // Returns as soon as the event is persisted. Delivery is now the shim's
  // responsibility.
  await outbox.broadcast(
    event,
    relaySet: const RelaySet.explicit([
      'wss://relay.damus.io',
      'wss://nos.lol',
    ]),
  );

  // Or let the shim find the author's write relays later, offline included.
  await outbox.broadcast(event, relaySet: RelaySet.outbox(myPubKey));
}

package:ndk/ndk.dart also exports a RelaySet. In a file that imports both packages, add hide RelaySet to the ndk import.

Semantics

broadcast(event, {required RelaySet relaySet, String? pubkey})

Persists event and schedules an immediate attempt to the relays of relaySet (see Relay sets). URLs are normalized (lowercased, trailing / stripped) before storage. broadcast never waits on the network. A set that needs no lookup and holds no relay throws an ArgumentError.

Records are keyed by the pair (event.id, pubkey). If a record with the same pair already exists, the relay lists are merged: deliveredAt is preserved if every relay in the merged list is already in the entry's ack set, otherwise the entry is demoted to pending so the missing retryable relays get pushed. A relay set that needs a lookup is merged the same way once it resolves. The same event queued under a different pubkey is a separate record.

pubkey is optional and defaults to null (unattributed). See Account-scoped clearing for what it buys you.

Relay sets

A RelaySet describes where an event goes instead of listing URLs:

RelaySet.explicit(['wss://a', 'wss://b'])  // fixed URLs
RelaySet.outbox(pubkey)                    // NIP-65 write relays
RelaySet.nip65(pubkey)                     // every NIP-65 relay, read and write
RelaySet.inbox([pubkeys])                  // NIP-65 read relays of each pubkey
RelaySet.dm([pubkeys])                     // NIP-17 DM relays (kind 10050)
RelaySet.private(pubkey)                   // NIP-37 private relays (kind 10013)
RelaySet.union([a, b])                     // every relay of every set
RelaySet.fallback([a, b])                  // first set with at least one relay

Use dm, not inbox, for gift wraps. A user publishes their kind 10050 to their NIP-65 write relays, so a dm lookup first resolves the pubkey's NIP-65, then searches the kind 10050 on those write relays. If the NIP-65 is unavailable, so is the DM list. If it is not found, the kind 10050 is still searched on the discovery relays alone.

private follows the same chaining for the NIP-37 kind 10013, which must be published on the NIP-65 write relays. Its relays are encrypted to their owner, so pubkey must be an account the shim can decrypt for. With NDK, that is an account in ndk.accounts whose signer supports NIP-44. While the account is missing (for instance before the app has logged it in) or decryption fails, the lookup is unavailable and retried.

Decryption goes through ndk.decryptedEventPayloads: a given kind 10013 event is decrypted once, concurrent lookups share that decryption, and later lookups read the plaintext from NDK's cache without the account or its signer. The signer is asked again only when the user publishes a new list. That cache is as persistent as the CacheManager given to NDK, and clearing it is up to the app through NDK's cache API: clearLocalAccountData does not touch it. Once frozen, the private relays are also stored in plain text in the queue's sembast database.

The set is persisted with the entry, which stays pending with an empty relays list and a non-null pendingRelaySet. The worker resolves it on its next attempt and, the first time every lookup concludes, freezes the result into relays and clears pendingRelaySet. Later retries push to those frozen relays and never look the lists up again. A set made only of explicit sets is resolved at enqueue time.

Each relay list lookup ends in one of three states:

  • found: the list exists (it may hold no relay of the wanted kind).
  • not found: at least one relay sent EOSE and none holds the list. It counts as an empty list.
  • unavailable: no relay answered (offline, timeouts, disconnections, CLOSED). Whether the list exists is unknown.

If any lookup the result depends on is unavailable, nothing is frozen: the entry records resolutionError, increments resolutionAttempts and retries with backoff. This matters for fallback: an offline lookup never falls through to the next set, so going offline cannot lock an entry onto its fallback relays. fallback only moves on when a set resolves to no relay.

If the whole set resolves to no relay, the entry becomes failed with resolutionError: 'relay set resolved to no relay'. Calling broadcast again with a set queues a new resolution on the same entry.

OfflineBroadcast.withNdk reads kind 10002, 10050 and 10013 through ndk.requests.query, NDK's cache first, so a list NDK has already seen resolves offline. The lists are queried on defaultIndexerRelays (coracle, yabu.me, purplepag.es, nos.social indexers), not on NDK's bootstrap relays, plus the write relays for a kind 10050 or 10013. Pass relayListDiscoveryRelays to query other indexers. With the default constructor, pass your own relayListFn; without one, only explicit sets are accepted.

Only the relay query is timed out (relayListQueryTimeout, NDK's query timeout by default). The shim never times out a lookup as a whole, so a remote signer (NIP-46, NIP-55) can wait for user approval as long as it takes. dispose does not wait for such a lookup: the resolution is abandoned without writing and starts over on the next run.

A "not found" answer from a fast relay can arrive while a slower relay holding the list times out.

Account-scoped clearing

pubkey labels the local account an entry was queued under, so its queue can be dropped on logout:

await outbox.broadcast(event, relaySet: ..., pubkey: myPubkey);
...
await outbox.clearLocalAccountData(pubkey: myPubkey); // wipe this account
await outbox.clearAllLocalData();                     // wipe everything
  • It is a plain label, not event.pubKey. An event may be queued under an account that did not sign it (rebroadcasting someone else's note), and for a gift wrap (kind 1059) event.pubKey is an ephemeral throwaway key. Pass the real sending account so NIP-17 DMs and other wraps can still be cleared.
  • Leaving pubkey null leaves the entry unattributed: retried like any other, keyed by the bare event.id, but never removed by clearLocalAccountData. Entries written before 0.4.0 are unattributed.
  • clearLocalAccountData removes matching entries for good, delivered or not; pending ones will never be retried again. It is safe to call while attempts are in flight (a concurrent attempt no-ops once its record is gone), but do not broadcast for an account you are clearing, as an overlapping enqueue can re-create its record.
  • get, watch, and rebroadcast take the same optional pubkey to select which record they act on. watchPending and listAll span all accounts; filter on QueuedBroadcast.pubkey for one.
  • clearAllLocalData empties the shim's store only; other sembast stores in the same database are untouched.

retryNow()

Forces an immediate scan of due entries, bypassing the online check. Use it as an explicit override (e.g. when the user pulls to refresh).

Connectivity awareness

OfflineBroadcast.withNdk() subscribes to ndk.connectivity.relayConnectivityChanges and pauses the periodic retry timer while no public relay is connected. As soon as a public relay comes online, the shim replays everything that's due. Loopback addresses, RFC1918 IPv4, ULA/link-local IPv6, and mDNS .local names are excluded from the "is online" computation so a local dev relay cannot mask a real outage.

For non-NDK setups, pass any Stream<bool> onlineSignal to the default constructor:

OfflineBroadcast(
  broadcastFn: ...,
  db: db,
  onlineSignal: yourConnectivityStream, // true while online, false otherwise
);

If you don't pass anything, the shim assumes it is always online and the periodic timer runs unconditionally (pre-0.2 behavior).

rebroadcast(id, {String? pubkey, String? relay})

ackedRelays and deliveredAt are monotonic. rebroadcast never rewrites the past; it queues a one-shot push via a transient forcedRelays override that the next attempt consumes. pubkey selects which record to act on, the same value passed to broadcast; it returns null if no such record exists.

  • rebroadcast(id): schedules an immediate push to every relay in the entry's relays list, including those that already acked. Useful when you suspect a relay dropped your event. Acks and deliveredAt are preserved regardless of the new attempt's outcome.
  • rebroadcast(id, relay: r): pushes to that single relay. If r is new to the entry, it joins the target list and the entry is demoted to pending until r acks. If r was already there, the historical state is preserved.

What "success" means

The full target set must ack. NDK's own considerDonePercent knob is not used as a delivery threshold; it only governs when the underlying future completes, which is a different question.

Terminal failures

NIP-01 requires failed OK messages to start with a machine-readable prefix followed by :. When a relay returns OK false with one of these prefixes, the shim records the message in QueuedBroadcast.terminalErrors and stops retrying that relay for the event:

  • blocked
  • error
  • invalid
  • pow
  • restricted

If every target relay is either acked or terminally rejected, and at least one relay is terminally rejected, the entry status becomes BroadcastStatus.failed and it leaves watchPending(). Non-terminal failures, including rate-limited, unknown prefixes, missing colons, and global broadcaster exceptions remain retryable. Timeouts, no response, and transport-level relay failures remain retryable until the relay has been inaccessible for 12 consecutive online attempts.

Manual rebroadcast(...) can still force another push. If a previously terminally rejected relay later succeeds, its terminal error is cleared and the entry can become delivered.

What the shim does NOT do

  • It never signs. Whatever event you pass is forwarded as-is to ndk.broadcast.broadcast. If the event is unsigned, NDK signs it using its configured EventSigner. The shim has no opinion on signing.
  • It never auto-deletes records. Even after full delivery, an entry stays in the database until you remove it, via clearLocalAccountData / clearAllLocalData or by clearing sembast records directly.
  • It does not apply a max-attempts limit. Retryable failures continue with exponential backoff until they ack, are manually rebroadcast, become a terminal NIP-01 rejection, or hit the inaccessible-relay cutoff.

Tuning

OfflineBroadcast.withNdk(
  ndk,
  db: db,
  storeName: 'broadcasts',                       // sembast store name
  tickInterval: const Duration(seconds: 30),     // periodic retry scan
  initialBackoff: const Duration(seconds: 5),    // backoff floor
  maxBackoff: const Duration(minutes: 30),       // backoff ceiling
  perAttemptTimeout: const Duration(seconds: 10),// gives up on a single NDK call after this
  maxInaccessibleAttemptsPerRelay: 12,           // stops dead relays while online
  relayListDiscoveryRelays: defaultIndexerRelays, // relays queried for relay lists
  relayListQueryTimeout: null,                   // relay list query timeout, NDK's by default
);

Architecture in one diagram

caller.broadcast(event, relaySet)
        │
        ▼
  sembast write ──── durable, returns to caller here
        │
        ▼
  resolve pendingRelaySet (once)         ────►  retry with backoff while unavailable
        │
        ▼
  ndk.broadcast.broadcast(event, specificRelays: remaining)
        │
        ▼
  await broadcastDoneFuture
        │
        ▼
  per-relay union into ackedRelays       ────►  delivered when ⊇ relays
  terminal OK false into terminalErrors  ────►  failed when all remaining are terminal
  inaccessible relay attempts            ────►  terminal after 12 online failures
  retryable error into lastErrors              (otherwise schedule backoff)

A Timer.periodic scans findDue every tickInterval and replays whatever is overdue. retryNow() runs the same scan immediately.

Testing your integration

OfflineBroadcast is fully unit-testable without NDK. Pass a custom BroadcastFn to the default constructor:

final outbox = OfflineBroadcast(
  broadcastFn: (event, relays) => NdkBroadcastResponse(
    publishEvent: event,
    broadcastDoneStream: Stream.value([
      for (final r in relays)
        RelayBroadcastResponse(
          relayUrl: r,
          okReceived: true,
          broadcastSuccessful: true,
        ),
    ]),
  ),
  db: await newDatabaseFactoryMemory().openDatabase('test.db'),
);

The package's own test suite uses exactly this approach; see test/offline_broadcast_test.dart.

License

MIT

Libraries

broadcast_queue_shim_for_ndk
Offline-first shim around the ndk package's broadcast use case.