follow-feed (kinds=[1,6] × 150 authors, ORDER BY created_at DESC LIMIT 500)
was geode's 5.5× loss (97.7ms vs strfry 17.6ms). Investigated whether any
change is worth it, across read + write + size.
Read (FollowFeedReadBenchmark, in-memory, scale 5 ≈ 1.05M events):
prolific-recent sparse-old
current 5.7 ms 1.9 ms
scan (strfry) 1.0 ms 1601.9 ms
union 316.9 ms 20.0 ms
- scan (created_at index + early LIMIT) wins for active follows but is
catastrophic for sparse/inactive follows AND grows with corpus size
(234ms→1601ms from scale 1→5) — following rarely-posting accounts is
common, so it'd be a severe regression.
- union (300 per-branch subqueries) is dominated by branch overhead.
- current is the only robust option — flat across scale, bounded by the
followed set, never catastrophic. The 97.7ms is a worst case (the 150
MOST prolific authors, disk-bound reading all their matching rows).
No safe SQL-level swap exists; each alternative trades geode's worst case
for a worse one on a common workload. The only universal improvement is
strfry's app-level k-way merge (O(LIMIT+streams)) — a real new executor,
not a SQL tweak.
Write & size: neutral for every candidate — all reuse existing indexes
(query_by_kind_pubkey_created / query_by_created_at_id), none adds a
CREATE INDEX, so ingest throughput and storage are untouched regardless of
choice. A new index was considered and rejected (taxes every write, helps
one shape, reverts under ANALYZE).
Decision: keep the current composite plan. Full write-up in
quartz/plans/2026-07-04-follow-feed-read-tradeoff.md.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012EZeWww5TJnzBZKPoc6mvU
Quartz Guide for Clients
Here's how to structure a new Twitter-like client.
Architecture
Set up a Context class to wire Quartz components together. Usually there is only one instance of this class.
object AppGraph {
// application-wide scope
private val scope = CoroutineScope(Dispatchers.IO + SupervisorJob())
// the local db
val sqlite = EventStore(dbName = "demo-events.db")
// the local cache that keeps only one copy of each event in memory
val interned = InterningEventStore(sqlite)
// the observable db, that you can produce flows that auto update
val db = ObservableEventStore(interned)
// the client to access relays
val client = NostrClient(websocketBuilder = KtorWebSocket.Builder())
// sends all events, regardless of the subscription, to the local db
val collector = EventCollector(client) { event, _ ->
runCatching {
db.insert(event)
}
}
// update this variable when a user logs in, starts with a guest
var signer: NostrSigner = NostrSignerInternal(KeyPair())
init {
// Periodic NIP-40 sweep — drops expired events from SQLite and
// emits StoreChange.DeleteExpired so live projections drop them
// too. Without this the on-disk store grows monotonically.
scope.launch {
while (isActive) {
delay(15.minutes)
runCatching { db.deleteExpiredEvents() }
}
}
}
}
Then use a view model to subscribe to relays and the local db at the same time, like this:
class NotesFeed(
private val db: ObservableEventStore,
private val client: NostrClient,
) {
private val subId = newSubId()
private val filter = Filter(kinds = listOf(TextNoteEvent.KIND), limit = 100)
private val relays =
setOf(
"wss://relay.damus.io".normalizeRelayUrl(),
"wss://nos.lol".normalizeRelayUrl(),
"wss://relay.nostr.band".normalizeRelayUrl(),
)
val notes: Flow<ProjectionState<TextNoteEvent>> =
db
.project<TextNoteEvent>(filter)
.filterItems { it.value.isNewThread() }
.onStart { client.subscribe(subId, relays.associateWith { listOf(filter) }) }
.onCompletion { client.unsubscribe(subId) }
}
class FeedViewModel(
private val db: ObservableEventStore,
private val client: NostrClient,
) : ViewModel() {
val notesFeed = NotesFeed(db, client)
val feed = notesFeed
.flow
.stateIn(viewModelScope, SharingStarted.WhileSubscribed(5_000), ProjectionState.Loading)
fun send(text: String, signer: NostrSigner) {
viewModelScope.launch {
val signed = signer.sign<TextNoteEvent>(TextNoteEvent.build(text))
// Hits the bus → projection picks it up alongside any inbound relay copy.
db.insert(signed)
client.publish(signed, relays)
}
}
}
Notice that the notes flow is ready for the UI and automatically subscribes
and unsubscribes to any group of relays and filters the user wants. Similarly,
the send function updates both the local db and the relay.
NostrClient connects on-demand: the first subscribe(...) or publish(...) to a relay triggers the socket. There's no need to call client.connect() at startup — it's only useful for resuming after a prior disconnect().
Building a reactive feed UI
A feed screen reads from the view model's feed flow, which only updates when new events arrive or are deleted due to kind 5 deletions, vanish requests or expirations.
fun main() {
application {
val state = rememberWindowState(size = DpSize(560.dp, 720.dp))
Window(onCloseRequest = ::exitApplication, state = state, title = "Nostr Kind 1 Demo") {
MaterialTheme {
val viewModel = remember {
FeedViewModel(AppGraph.db, AppGraph.client, AppGraph.signer)
}
val noteState by viewModel.feed.collectAsStateWithLifecycle()
when (noteState) {
is ProjectionState.Loading -> LoadingFeed()
is ProjectionState.Loaded -> Feed(noteState.items)
}
}
}
}
}
@Composable
private fun LoadingFeed() {
Box(modifier = Modifier.fillMaxSize(), contentAlignment = Alignment.Center) {
CircularProgressIndicator()
}
}
@Composable
private fun Feed(items: List<MutableStateFlow<TextNoteEvent>>) {
LazyColumn(modifier = Modifier.fillMaxSize()) {
items(items = items, key = { it.value.id }) { handle ->
NoteRow(handle)
HorizontalDivider()
}
}
}
@Composable
private fun NoteRow(handle: MutableStateFlow<TextNoteEvent>) {
val event by handle.collectAsStateWithLifecycle()
Text(
text = event.content,
style = MaterialTheme.typography.bodyMedium,
modifier = Modifier.padding(top = 4.dp),
)
}
Notice how each how also subscribe for changes. This is important to receive updates from replaceable and addressable events.
Appendix A
Quartz doesn't offer a Ktor websocket, but you can use this one as reference.
/**
* Ktor-based [WebSocket] for talking to a Nostr relay.
*
* Quartz exposes [WebsocketBuilder] as the only seam between its relay-pool
* and the underlying transport, so all this class has to do is open a Ktor
* websocket session, forward incoming text frames to [out], and let Quartz
* drive sends.
*/
class KtorWebSocket(
private val url: NormalizedRelayUrl,
private val httpClient: HttpClient,
private val out: WebSocketListener,
) : WebSocket {
private val scope = CoroutineScope(Dispatchers.IO + SupervisorJob())
private var session: DefaultWebSocketSession? = null
private var readerJob: Job? = null
override fun needsReconnect(): Boolean = session == null
override fun connect() {
readerJob =
scope.launch {
try {
val s = httpClient.webSocketSession(urlString = url.url)
session = s
out.onOpen(0, false)
for (frame in s.incoming) {
if (frame is Frame.Text) {
out.onMessage(frame.readText())
}
}
val reason = s.closeReason.await()
out.onClosed(
code =
reason?.code?.toInt() ?: CloseReason.Codes.NORMAL.code
.toInt(),
reason = reason?.message ?: "",
)
} catch (t: Throwable) {
out.onFailure(t, null, null)
} finally {
session = null
}
}
}
override fun disconnect() {
val s = session
session = null
readerJob?.cancel()
readerJob = null
if (s != null) {
runBlocking { s.close(CloseReason(CloseReason.Codes.NORMAL, "client disconnect")) }
}
scope.cancel()
}
override fun send(msg: String): Boolean {
val s = session ?: return false
scope.launch { s.send(msg) }
return true
}
/**
* The factory Quartz hands to [com.vitorpamplona.quartz.nip01Core.relay.client.NostrClient].
* One [HttpClient] is shared by every relay in the pool.
*/
class Builder(
private val httpClient: HttpClient = defaultClient(),
) : WebsocketBuilder {
override fun build(
url: NormalizedRelayUrl,
out: WebSocketListener,
): WebSocket = KtorWebSocket(url, httpClient, out)
companion object {
fun defaultClient() =
HttpClient(CIO) {
install(WebSockets)
}
}
}
}