perf(quartz): index-driven fanout for LiveEventStore via FilterIndex<S>

Replace the SharedFlow-based broadcast in LiveEventStore with an
inverted index over filter-bearing subscribers. Each REQ registers
its filters into a per-store FilterIndex<LiveSubscription>; insert()
calls index.candidatesFor(event) and only delivers to candidates
whose Filter.match still passes. Cuts the per-event walk from
O(N_subs * N_filters) to a few hash lookups plus match() over a
small candidate set.

FilterIndex itself lives next to Filter.kt (commonMain, KMP-friendly,
AtomicReference + COW) so other call sites with the same shape
(LocalCache.observables, ObservableEventStore.changes) can reuse it.
Each filter contributes entries on its single most-selective dimension
(ids > authors > tags > tagsAll > kinds > unindexed) to keep buckets
narrow and avoid Set-dedupe work in candidatesFor.

The historical-replay race the previous SharedFlow + onSubscription
handoff closed is preserved by registering BEFORE replay starts and
deduping seen ids until EOSE.
This commit is contained in:
Claude
2026-05-07 22:37:31 +00:00
parent ede4bc5eab
commit 3946117084
4 changed files with 738 additions and 96 deletions
@@ -0,0 +1,285 @@
/*
* Copyright (c) 2025 Vitor Pamplona
*
* Permission is hereby granted, free of charge, to any person obtaining a copy of
* this software and associated documentation files (the "Software"), to deal in
* the Software without restriction, including without limitation the rights to use,
* copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the
* Software, and to permit persons to whom the Software is furnished to do so,
* subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
* FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
* COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
* AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
* WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
package com.vitorpamplona.quartz.nip01Core.relay.filters
import com.vitorpamplona.quartz.nip01Core.core.Event
import com.vitorpamplona.quartz.nip01Core.core.HexKey
import kotlin.concurrent.atomics.AtomicReference
import kotlin.concurrent.atomics.ExperimentalAtomicApi
/**
* Inverted index over a population of [Filter]-bearing subscribers.
* Given an [Event], returns the (much smaller) set of subscribers
* whose filters could match — callers still run [Filter.match] on
* each candidate to enforce negative constraints (`since` / `until`
* / `tagsAll` / etc.) but skip the per-event walk over subscribers
* that share no narrowing field with the event.
*
* Used by the relay server's `LiveEventStore` for fanout from one
* inserted event to many `REQ` subscriptions, and by the client-side
* `LocalCache.observables` registry for the same shape inside the
* app (one accepted event, many feed observers).
*
* ## Indexing strategy
*
* Each registered filter contributes entries to **one** dimension —
* the most selective indexable field. Picking one dimension instead
* of all of them avoids over-counting subscribers in [candidatesFor]
* (no `Set` dedup work) and minimises bucket churn on register /
* unregister:
*
* 1. `ids` (most selective; an id matches one event).
* 2. `authors`.
* 3. The first single-letter tag in `tags` (then `tagsAll`).
* 4. `kinds`.
* 5. None of the above → registered into [unindexedKey].
*
* Multi-filter registrations OR the per-filter selections together
* (a subscriber matches if *any* of its filters matches the event,
* which mirrors `filters.any { it.match(...) }`).
*
* ## Concurrency
*
* State is held in a single [AtomicReference] and mutated via
* copy-on-write CAS loops, mirroring the
* `nip86RelayManagement.server.BanStore` pattern. Reads in
* [candidatesFor] and [forEach] are wait-free single-load atomic.
* Writes (subscription register / unregister) copy the inner maps
* — fine for this workload because writes are subscription-rate
* (rare) while reads are event-rate (frequent).
*
* ## What the index does NOT cover
*
* - Negative filter constraints (`since`, `until`, `tagsAll`,
* `limit`-already-saturated). The candidate set is a
* super-set; callers must still run [Filter.match] on each
* candidate.
* - Subscribers driven by an arbitrary `(Event) -> Boolean`
* predicate without an underlying [Filter]. Use
* [registerUnindexed] for those — they're returned for every
* event.
* - Membership-driven re-evaluation paths (e.g. an addressable
* `v2` that no longer matches a filter but the observer
* already holds `v1`). Those callers must consult their own
* membership state in addition to [candidatesFor].
*/
@OptIn(ExperimentalAtomicApi::class)
class FilterIndex<S : Any> {
/**
* Bucket key. Five concrete shapes plus a sentinel for filters
* with no indexable narrowing field.
*/
private sealed interface BucketKey
private data class IdKey(
val id: HexKey,
) : BucketKey
private data class AuthorKey(
val author: HexKey,
) : BucketKey
private data class TagKey(
val letter: String,
val value: String,
) : BucketKey
private data class KindKey(
val kind: Int,
) : BucketKey
private object Unindexed : BucketKey
/**
* Single immutable snapshot. [buckets] maps a key to the set of
* subscribers registered under it; [assignments] is the reverse
* map used by [unregister] to find a subscriber's keys without
* scanning every bucket.
*/
private data class State<S>(
val buckets: Map<BucketKey, Set<S>> = emptyMap(),
val assignments: Map<S, Set<BucketKey>> = emptyMap(),
)
private val state: AtomicReference<State<S>> = AtomicReference(State())
/** Number of distinct subscribers currently registered. */
fun size(): Int = state.load().assignments.size
fun isEmpty(): Boolean = state.load().assignments.isEmpty()
/**
* Register [subscriber] under the bucket(s) selected for [filter].
* If [filter] has no indexable field the subscriber is added to
* the unindexed pool and is returned for every event.
*/
fun register(
filter: Filter,
subscriber: S,
) {
val keys = selectKeys(filter).ifEmpty { listOf(Unindexed) }
addAssignments(subscriber, keys)
}
/**
* Register [subscriber] for a list of filters (OR semantics).
* Each filter's most-selective dimension contributes its keys;
* any filter with no indexable field adds the subscriber to the
* unindexed pool, which dominates dispatch (the subscriber
* matches every event).
*/
fun register(
filters: List<Filter>,
subscriber: S,
) {
if (filters.isEmpty()) {
addAssignments(subscriber, listOf(Unindexed))
return
}
val keys = mutableListOf<BucketKey>()
for (f in filters) {
val perFilter = selectKeys(f)
if (perFilter.isEmpty()) {
keys.add(Unindexed)
} else {
keys.addAll(perFilter)
}
}
addAssignments(subscriber, keys)
}
/**
* Register [subscriber] in the unindexed pool. Use this for
* subscribers driven by an opaque predicate where the index
* can't infer a narrowing field.
*/
fun registerUnindexed(subscriber: S) = addAssignments(subscriber, listOf(Unindexed))
/**
* Remove [subscriber] from every bucket it was registered in.
* No-op if the subscriber isn't currently registered.
*/
fun unregister(subscriber: S) {
while (true) {
val current = state.load()
val keys = current.assignments[subscriber] ?: return
val newBuckets = current.buckets.toMutableMap()
for (key in keys) {
val cur = newBuckets[key] ?: continue
val next = cur - subscriber
if (next.isEmpty()) {
newBuckets.remove(key)
} else {
newBuckets[key] = next
}
}
val newAssignments = current.assignments - subscriber
if (state.compareAndSet(current, State(newBuckets, newAssignments))) return
}
}
/**
* Subscribers whose filters might match [event]. The result is a
* super-set: callers must still run `filter.match(event)` on each
* candidate to handle negative constraints.
*
* Iteration order is insertion-stable per call but otherwise
* unspecified.
*/
fun candidatesFor(event: Event): Set<S> {
val s = state.load()
if (s.buckets.isEmpty()) return emptySet()
val result = LinkedHashSet<S>()
s.buckets[Unindexed]?.let { result.addAll(it) }
s.buckets[IdKey(event.id)]?.let { result.addAll(it) }
s.buckets[AuthorKey(event.pubKey)]?.let { result.addAll(it) }
s.buckets[KindKey(event.kind)]?.let { result.addAll(it) }
for (tag in event.tags) {
if (tag.size >= 2 && tag[0].length == 1) {
s.buckets[TagKey(tag[0], tag[1])]?.let { result.addAll(it) }
}
}
return result
}
/**
* Visit every registered subscriber. Used by callers that need
* to broadcast something the index can't help with (e.g.
* `LocalCache.refreshDeletedNoteObservers` — the deletion path
* has no event-shape to consult, every observer must see it).
*/
fun forEach(action: (S) -> Unit) {
for (sub in state.load().assignments.keys) action(sub)
}
private fun addAssignments(
subscriber: S,
keys: List<BucketKey>,
) {
if (keys.isEmpty()) return
val keySet = keys.toSet()
while (true) {
val current = state.load()
val newBuckets = current.buckets.toMutableMap()
for (key in keySet) {
val cur = newBuckets[key] ?: emptySet()
if (subscriber in cur) continue
newBuckets[key] = cur + subscriber
}
val existing = current.assignments[subscriber]
val merged = if (existing == null) keySet else existing + keySet
val newAssignments = current.assignments + (subscriber to merged)
if (state.compareAndSet(current, State(newBuckets, newAssignments))) return
}
}
/**
* Pick the most-selective indexable dimension for [filter] and
* expand it into one [BucketKey] per value. Returns an empty
* list if no field is indexable — caller maps that to [Unindexed].
*/
private fun selectKeys(filter: Filter): List<BucketKey> {
if (!filter.ids.isNullOrEmpty()) {
return filter.ids.map { IdKey(it) }
}
if (!filter.authors.isNullOrEmpty()) {
return filter.authors.map { AuthorKey(it) }
}
if (!filter.tags.isNullOrEmpty()) {
val first =
filter.tags.entries.firstOrNull {
it.key.length == 1 && it.value.isNotEmpty()
}
if (first != null) return first.value.map { TagKey(first.key, it) }
}
if (!filter.tagsAll.isNullOrEmpty()) {
val first =
filter.tagsAll.entries.firstOrNull {
it.key.length == 1 && it.value.isNotEmpty()
}
if (first != null) return first.value.map { TagKey(first.key, it) }
}
if (!filter.kinds.isNullOrEmpty()) {
return filter.kinds.map { KindKey(it) }
}
return emptyList()
}
}
@@ -22,10 +22,11 @@ package com.vitorpamplona.quartz.nip01Core.relay.server
import com.vitorpamplona.quartz.nip01Core.core.Event
import com.vitorpamplona.quartz.nip01Core.relay.filters.Filter
import com.vitorpamplona.quartz.nip01Core.relay.filters.FilterIndex
import com.vitorpamplona.quartz.nip01Core.store.IEventStore
import kotlinx.coroutines.channels.BufferOverflow
import kotlinx.coroutines.flow.MutableSharedFlow
import kotlinx.coroutines.flow.onSubscription
import kotlinx.coroutines.awaitCancellation
import kotlin.concurrent.atomics.AtomicReference
import kotlin.concurrent.atomics.ExperimentalAtomicApi
/**
* A reactive event store that combines historical data retrieval with live event streaming.
@@ -35,21 +36,41 @@ import kotlinx.coroutines.flow.onSubscription
* End of Stored Events (EOSE), and then continues to stream matching new events as they
* are inserted.
*
* Live fanout from `insert` to interested subscribers is index-driven via [FilterIndex]:
* each [query] registers its filters; [insert] looks up the candidate set and only delivers
* to those whose filters actually match. This avoids the quadratic
* O(N_subscribers × N_filters_per_sub) per-event walk that a naïve broadcast would do.
*
* @property store The underlying persistent storage for events.
*/
@OptIn(ExperimentalAtomicApi::class)
class LiveEventStore(
private val store: IEventStore,
) {
private val newEventStream =
MutableSharedFlow<Event>(
replay = 0,
extraBufferCapacity = 100, // Optional: adjust for backpressure
onBufferOverflow = BufferOverflow.DROP_LATEST, // Default behavior
)
private val index = FilterIndex<LiveSubscription>()
/**
* One live REQ subscription. Carries the filters (for the
* post-index `match` re-check needed for negative constraints
* like `since` / `until` / `tagsAll`) and the delivery callback
* the index dispatches into. Identity-keyed inside [FilterIndex].
*/
private class LiveSubscription(
val filters: List<Filter>,
val deliver: (Event) -> Unit,
)
suspend fun insert(event: Event) {
store.insert(event)
newEventStream.tryEmit(event)
// Live fanout. The index returns a super-set; `match` enforces
// negative constraints. Synchronous delivery — callers are
// expected to keep `deliver` cheap (typically a `tryEmit` to
// a per-connection outbound queue).
for (sub in index.candidatesFor(event)) {
if (sub.filters.any { it.match(event) }) {
sub.deliver(event)
}
}
}
suspend fun query(
@@ -57,42 +78,51 @@ class LiveEventStore(
onEach: (Event) -> Unit,
onEose: () -> Unit,
) {
// Order matters: register the live collector BEFORE replaying
// stored events and signalling EOSE. Otherwise an event emitted
// between EOSE and `collect` is lost because [newEventStream] has
// replay=0. The race is only occasionally visible for kinds the
// store persists (insert latency masks it) but fires reliably for
// ephemeral kinds (20000-29999) where insert is a no-op — and
// ephemeral events MUST still reach matching live subscribers per
// NIP-01.
// During the historical replay, mark ids the store has
// emitted so the live path can dedupe. The index registers
// *before* the replay starts (otherwise an event accepted
// mid-replay would slip past the live path entirely — same
// race the previous SharedFlow-based implementation closed
// with `onSubscription`). Anything the store and the live
// path both observe gets dropped here on the live side.
//
// Side effect of registering the collector first: an event
// inserted *during* `store.query` will be both replayed by the
// store AND emitted to the live stream. We dedupe by tracking
// ids seen during the historical replay and skipping them on
// the live path. The set is dropped after EOSE so live-only
// events don't accumulate memory.
var inHistoricalPhase = true
var seenIds: HashSet<String>? = HashSet()
val historicalOnEach: (Event) -> Unit = { event ->
seenIds?.add(event.id)
onEach(event)
}
newEventStream
.onSubscription {
store.query(filters, historicalOnEach)
onEose()
// Free the dedupe set once we've crossed EOSE: from
// here on the live stream is the only source of
// events, so duplicates aren't possible.
inHistoricalPhase = false
seenIds = null
}.collect { newEvent ->
if (inHistoricalPhase && seenIds?.contains(newEvent.id) == true) return@collect
if (filters.any { it.match(newEvent) }) {
onEach(newEvent)
}
// Held in an AtomicReference because the live-dispatch
// coroutine (which calls `deliver` from `insert`) needs to
// see the post-EOSE handoff promptly. Once cleared to null,
// the dedupe check short-circuits and every live event is
// forwarded. The set itself is mutated only from the
// historical-replay closure below, which runs on the same
// coroutine that owns `query` — no cross-thread mutation.
val seenIds = AtomicReference<HashSet<String>?>(HashSet())
val sub =
LiveSubscription(
filters = filters,
deliver = { event ->
val seen = seenIds.load()
if (seen != null && seen.contains(event.id)) return@LiveSubscription
onEach(event)
},
)
index.register(filters, sub)
try {
store.query<Event>(filters) { event ->
seenIds.load()?.add(event.id)
onEach(event)
}
onEose()
// Drop the dedupe set so the live path stops paying for
// it. From this point the index drives delivery and
// duplicates are no longer possible.
seenIds.store(null)
// Suspend until the caller's coroutine is cancelled
// (e.g. NIP-01 CLOSE or connection drop). The `finally`
// unregisters from the index.
awaitCancellation()
} finally {
index.unregister(sub)
}
}
suspend fun count(filters: List<Filter>) = store.count(filters)
@@ -0,0 +1,252 @@
/*
* Copyright (c) 2025 Vitor Pamplona
*
* Permission is hereby granted, free of charge, to any person obtaining a copy of
* this software and associated documentation files (the "Software"), to deal in
* the Software without restriction, including without limitation the rights to use,
* copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the
* Software, and to permit persons to whom the Software is furnished to do so,
* subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
* FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
* COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
* AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
* WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
package com.vitorpamplona.quartz.nip01Core.relay.filters
import com.vitorpamplona.quartz.nip01Core.core.Event
import kotlin.test.Test
import kotlin.test.assertEquals
import kotlin.test.assertFalse
import kotlin.test.assertTrue
class FilterIndexTest {
private val authorA = "a".repeat(64)
private val authorB = "b".repeat(64)
private val authorC = "c".repeat(64)
private val pTag1 = "1".repeat(64)
private val pTag2 = "2".repeat(64)
private fun event(
id: String = "e".repeat(64),
pubkey: String = authorA,
kind: Int = 1,
tags: Array<Array<String>> = emptyArray(),
createdAt: Long = 1_700_000_000,
) = Event(
id = id,
pubKey = pubkey,
createdAt = createdAt,
kind = kind,
tags = tags,
content = "",
sig = "",
)
/** Distinct identity wrapper so tests can hold a stable handle. */
private data class Sub(
val name: String,
)
@Test
fun emptyIndexReturnsNoCandidates() {
val index = FilterIndex<Sub>()
assertTrue(index.isEmpty())
assertTrue(index.candidatesFor(event()).isEmpty())
}
@Test
fun authorFilterMatchesByAuthor() {
val index = FilterIndex<Sub>()
val s = Sub("s")
index.register(Filter(authors = listOf(authorA)), s)
assertTrue(s in index.candidatesFor(event(pubkey = authorA)))
assertFalse(s in index.candidatesFor(event(pubkey = authorB)))
}
@Test
fun multipleAuthorsAllRoutedToSameSubscriber() {
val index = FilterIndex<Sub>()
val s = Sub("s")
index.register(Filter(authors = listOf(authorA, authorB)), s)
assertTrue(s in index.candidatesFor(event(pubkey = authorA)))
assertTrue(s in index.candidatesFor(event(pubkey = authorB)))
assertFalse(s in index.candidatesFor(event(pubkey = authorC)))
}
@Test
fun kindFilterMatchesByKind() {
val index = FilterIndex<Sub>()
val s = Sub("s")
index.register(Filter(kinds = listOf(1, 7)), s)
assertTrue(s in index.candidatesFor(event(kind = 1)))
assertTrue(s in index.candidatesFor(event(kind = 7)))
assertFalse(s in index.candidatesFor(event(kind = 30023)))
}
@Test
fun authorWinsOverKindWhenBothPresent() {
// Filter has authors AND kinds — the more-selective dimension
// (authors) is used. An event with the right kind but a
// different author must NOT appear in candidates, otherwise
// the index didn't actually narrow.
val index = FilterIndex<Sub>()
val s = Sub("s")
index.register(Filter(authors = listOf(authorA), kinds = listOf(1)), s)
assertTrue(s in index.candidatesFor(event(pubkey = authorA, kind = 1)))
// Wrong author, right kind — index excludes correctly.
assertFalse(s in index.candidatesFor(event(pubkey = authorB, kind = 1)))
// Right author, wrong kind — index includes; Filter.match
// would post-reject. Exposed candidate is acceptable.
assertTrue(s in index.candidatesFor(event(pubkey = authorA, kind = 7)))
}
@Test
fun idFilterMostSelective() {
val index = FilterIndex<Sub>()
val s = Sub("s")
val targetId = "9".repeat(64)
index.register(Filter(ids = listOf(targetId), kinds = listOf(1)), s)
assertTrue(s in index.candidatesFor(event(id = targetId)))
assertFalse(s in index.candidatesFor(event(id = "8".repeat(64))))
}
@Test
fun tagFilterMatchesEventsCarryingTheTag() {
val index = FilterIndex<Sub>()
val s = Sub("s")
index.register(Filter(tags = mapOf("p" to listOf(pTag1))), s)
val matching = event(tags = arrayOf(arrayOf("p", pTag1)))
val nonMatching = event(tags = arrayOf(arrayOf("p", pTag2)))
assertTrue(s in index.candidatesFor(matching))
assertFalse(s in index.candidatesFor(nonMatching))
}
@Test
fun unindexedFilterMatchesEverything() {
// A filter with no narrowing field (e.g. just `since`) lives
// in the unindexed pool. Every event must include it.
val index = FilterIndex<Sub>()
val s = Sub("s")
index.register(Filter(since = 1L), s)
assertTrue(s in index.candidatesFor(event(pubkey = authorA, kind = 1)))
assertTrue(s in index.candidatesFor(event(pubkey = authorB, kind = 30023)))
}
@Test
fun registerUnindexedExplicit() {
val index = FilterIndex<Sub>()
val s = Sub("s")
index.registerUnindexed(s)
assertTrue(s in index.candidatesFor(event(pubkey = authorA)))
assertTrue(s in index.candidatesFor(event(pubkey = authorB)))
}
@Test
fun unregisterRemovesFromAllBuckets() {
val index = FilterIndex<Sub>()
val s = Sub("s")
index.register(Filter(authors = listOf(authorA, authorB)), s)
assertEquals(1, index.size())
index.unregister(s)
assertEquals(0, index.size())
assertFalse(s in index.candidatesFor(event(pubkey = authorA)))
assertFalse(s in index.candidatesFor(event(pubkey = authorB)))
}
@Test
fun unregisterOfUnknownIsNoOp() {
val index = FilterIndex<Sub>()
val s = Sub("s")
index.unregister(s) // should not throw
assertEquals(0, index.size())
}
@Test
fun multiFilterRegistrationOrsAllSelections() {
// Subscriber wants events from authorA OR kind 30023.
val index = FilterIndex<Sub>()
val s = Sub("s")
index.register(
filters =
listOf(
Filter(authors = listOf(authorA)),
Filter(kinds = listOf(30023)),
),
subscriber = s,
)
assertTrue(s in index.candidatesFor(event(pubkey = authorA, kind = 1)))
assertTrue(s in index.candidatesFor(event(pubkey = authorB, kind = 30023)))
assertFalse(s in index.candidatesFor(event(pubkey = authorB, kind = 1)))
}
@Test
fun multipleSubscribersUnionInCandidates() {
val index = FilterIndex<Sub>()
val s1 = Sub("s1")
val s2 = Sub("s2")
val s3 = Sub("s3")
index.register(Filter(authors = listOf(authorA)), s1)
index.register(Filter(authors = listOf(authorB)), s2)
index.register(Filter(kinds = listOf(1)), s3)
val cands = index.candidatesFor(event(pubkey = authorA, kind = 1))
// s1 hits via author, s3 via kind, s2 must be excluded.
assertTrue(s1 in cands)
assertTrue(s3 in cands)
assertFalse(s2 in cands)
}
@Test
fun forEachVisitsEverySubscriberOnce() {
val index = FilterIndex<Sub>()
val s1 = Sub("s1")
val s2 = Sub("s2")
val s3 = Sub("s3")
index.register(Filter(authors = listOf(authorA, authorB)), s1)
index.register(Filter(kinds = listOf(1)), s2)
index.registerUnindexed(s3)
val visited = mutableListOf<Sub>()
index.forEach { visited.add(it) }
assertEquals(3, visited.size)
assertEquals(setOf(s1, s2, s3), visited.toSet())
}
@Test
fun registerThenUnregisterLeavesNoStaleBuckets() {
// Churn test — repeatedly add and remove a subscriber and
// verify the index ends up empty (no leaked bucket entries).
val index = FilterIndex<Sub>()
val s = Sub("s")
repeat(100) {
index.register(
Filter(authors = listOf(authorA), kinds = listOf(1, 7), tags = mapOf("p" to listOf(pTag1))),
s,
)
index.unregister(s)
}
assertTrue(index.isEmpty())
assertTrue(index.candidatesFor(event(pubkey = authorA)).isEmpty())
}
}