exponential backoff is now just bees. did you want bees? no? well you have them now. congrats

scale rocksdb background jobs and subcompactions

change rocksdb default error level to info from error

delete unused num_threads function

fix warns from cargo

conduwuit-example.toml

@@ -1066,7 +1066,7 @@
 # 3 to 5 = Statistics with possible performance impact.
 # 6 = All statistics.
 #
-#rocksdb_stats_level = 1
+#rocksdb_stats_level = 3
 
 # This is a password that can be configured that will let you login to the
 # server bot account (currently `@conduit`) for emergency troubleshooting
@@ -1680,11 +1680,9 @@
 #stream_amplification = 1024
 
 # Number of sender task workers; determines sender parallelism. Default is
-# '0' which means the value is determined internally, likely matching the
+# number of CPU cores. Override by setting a different value.
-# number of tokio worker-threads or number of cores, etc. Override by
-# setting a non-zero value.
 #
-#sender_workers = 0
+#sender_workers = 4
 
 # Enables listener sockets; can be set to false to disable listening. This
 # option is intended for developer/diagnostic purposes only.

src/core/config/mod.rs

@@ -1235,7 +1235,7 @@ pub struct Config {
 	/// 3 to 5 = Statistics with possible performance impact.
 	/// 6 = All statistics.
 	///
-	/// default: 1
+	/// default: 3
 	#[serde(default = "default_rocksdb_stats_level")]
 	pub rocksdb_stats_level: u8,
 
@@ -1917,12 +1917,10 @@ pub struct Config {
 	pub stream_amplification: usize,
 
 	/// Number of sender task workers; determines sender parallelism. Default is
-	/// '0' which means the value is determined internally, likely matching the
+	/// core count. Override by setting a different value.
-	/// number of tokio worker-threads or number of cores, etc. Override by
-	/// setting a non-zero value.
 	///
-	/// default: 0
+	/// default: core count
-	#[serde(default)]
+	#[serde(default = "default_sender_workers")]
 	pub sender_workers: usize,
 
 	/// Enables listener sockets; can be set to false to disable listening. This
@@ -2153,45 +2151,48 @@ fn default_database_backups_to_keep() -> i16 { 1 }
 
 fn default_db_write_buffer_capacity_mb() -> f64 { 48.0 + parallelism_scaled_f64(4.0) }
 
-fn default_db_cache_capacity_mb() -> f64 { 128.0 + parallelism_scaled_f64(64.0) }
+fn default_db_cache_capacity_mb() -> f64 { 512.0 + parallelism_scaled_f64(512.0) }
 
-fn default_pdu_cache_capacity() -> u32 { parallelism_scaled_u32(10_000).saturating_add(100_000) }
+fn default_pdu_cache_capacity() -> u32 { parallelism_scaled_u32(50_000).saturating_add(100_000) }
 
 fn default_cache_capacity_modifier() -> f64 { 1.0 }
 
 fn default_auth_chain_cache_capacity() -> u32 {
-	parallelism_scaled_u32(10_000).saturating_add(100_000)
-}
-
-fn default_shorteventid_cache_capacity() -> u32 {
 	parallelism_scaled_u32(50_000).saturating_add(100_000)
 }
 
+fn default_shorteventid_cache_capacity() -> u32 {
+	parallelism_scaled_u32(100_000).saturating_add(100_000)
+}
+
 fn default_eventidshort_cache_capacity() -> u32 {
-	parallelism_scaled_u32(25_000).saturating_add(100_000)
+	parallelism_scaled_u32(50_000).saturating_add(100_000)
 }
 
 fn default_eventid_pdu_cache_capacity() -> u32 {
-	parallelism_scaled_u32(25_000).saturating_add(100_000)
+	parallelism_scaled_u32(50_000).saturating_add(100_000)
 }
 
 fn default_shortstatekey_cache_capacity() -> u32 {
-	parallelism_scaled_u32(10_000).saturating_add(100_000)
+	parallelism_scaled_u32(50_000).saturating_add(100_000)
 }
 
 fn default_statekeyshort_cache_capacity() -> u32 {
-	parallelism_scaled_u32(10_000).saturating_add(100_000)
+	parallelism_scaled_u32(50_000).saturating_add(100_000)
 }
 
 fn default_servernameevent_data_cache_capacity() -> u32 {
-	parallelism_scaled_u32(100_000).saturating_add(500_000)
+	parallelism_scaled_u32(100_000).saturating_add(100_000)
 }
 
-fn default_stateinfo_cache_capacity() -> u32 { parallelism_scaled_u32(100) }
+fn default_stateinfo_cache_capacity() -> u32 {
+	parallelism_scaled_u32(500).clamp(100, 12000)
+}
 
-fn default_roomid_spacehierarchy_cache_capacity() -> u32 { parallelism_scaled_u32(1000) }
+fn default_roomid_spacehierarchy_cache_capacity() -> u32 {
+	parallelism_scaled_u32(500).clamp(100, 12000) }
 
-fn default_dns_cache_entries() -> u32 { 32768 }
+fn default_dns_cache_entries() -> u32 { 327680 }
 
 fn default_dns_min_ttl() -> u64 { 60 * 180 }
 
@@ -2297,7 +2298,7 @@ fn default_typing_client_timeout_max_s() -> u64 { 45 }
 
 fn default_rocksdb_recovery_mode() -> u8 { 1 }
 
-fn default_rocksdb_log_level() -> String { "error".to_owned() }
+fn default_rocksdb_log_level() -> String { "info".to_owned() }
 
 fn default_rocksdb_log_time_to_roll() -> usize { 0 }
 
@@ -2329,7 +2330,7 @@ fn default_rocksdb_compression_level() -> i32 { 32767 }
 #[allow(clippy::doc_markdown)]
 fn default_rocksdb_bottommost_compression_level() -> i32 { 32767 }
 
-fn default_rocksdb_stats_level() -> u8 { 1 }
+fn default_rocksdb_stats_level() -> u8 { 3 }
 
 // I know, it's a great name
 #[must_use]
@@ -2384,14 +2385,13 @@ fn default_admin_log_capture() -> String {
 fn default_admin_room_tag() -> String { "m.server_notice".to_owned() }
 
 #[allow(clippy::as_conversions, clippy::cast_precision_loss)]
-fn parallelism_scaled_f64(val: f64) -> f64 { val * (sys::available_parallelism() as f64) }
+pub fn parallelism_scaled_f64(val: f64) -> f64 { val * (sys::available_parallelism() as f64) }
 
-fn parallelism_scaled_u32(val: u32) -> u32 {
+pub fn parallelism_scaled_u32(val: u32) -> u32 { val.saturating_mul(sys::available_parallelism() as u32) }
-	let val = val.try_into().expect("failed to cast u32 to usize");
-	parallelism_scaled(val).try_into().unwrap_or(u32::MAX)
-}
 
-fn parallelism_scaled(val: usize) -> usize { val.saturating_mul(sys::available_parallelism()) }
+pub fn parallelism_scaled_i32(val: i32) -> i32 { val.saturating_mul(sys::available_parallelism() as i32) }
+
+pub fn parallelism_scaled(val: usize) -> usize { val.saturating_mul(sys::available_parallelism()) }
 
 fn default_trusted_server_batch_size() -> usize { 256 }
 
@@ -2411,6 +2411,8 @@ fn default_stream_width_scale() -> f32 { 1.0 }
 
 fn default_stream_amplification() -> usize { 1024 }
 
+fn default_sender_workers() -> usize { parallelism_scaled(1) }
+
 fn default_client_receive_timeout() -> u64 { 75 }
 
 fn default_client_request_timeout() -> u64 { 180 }

src/core/matrix/state_res/event_auth.rs

@@ -13,6 +13,7 @@ use ruma::{
 		power_levels::RoomPowerLevelsEventContent,
 		third_party_invite::RoomThirdPartyInviteEventContent,
 	},
+	EventId,
 	int,
 	serde::{Base64, Raw},
 };
@@ -21,7 +22,6 @@ use serde::{
 	de::{Error as _, IgnoredAny},
 };
 use serde_json::{from_str as from_json_str, value::RawValue as RawJsonValue};
-
 use super::{
 	Error, Event, Result, StateEventType, StateKey, TimelineEventType,
 	power_levels::{
@@ -149,8 +149,8 @@ where
 	for<'a> &'a E: Event + Send,
 {
 	debug!(
-		event_id = %incoming_event.event_id(),
+		event_id = format!("{}", incoming_event.event_id()),
-		event_type = ?incoming_event.event_type(),
+		event_type = format!("{}", incoming_event.event_type()),
 		"auth_check beginning"
 	);
 
@@ -242,20 +242,44 @@ where
 	}
 	*/
 
-	let (room_create_event, power_levels_event, sender_member_event) = join3(
+	// let (room_create_event, power_levels_event, sender_member_event) = join3(
-		fetch_state(&StateEventType::RoomCreate, ""),
+	// 	fetch_state(&StateEventType::RoomCreate, ""),
-		fetch_state(&StateEventType::RoomPowerLevels, ""),
+	// 	fetch_state(&StateEventType::RoomPowerLevels, ""),
-		fetch_state(&StateEventType::RoomMember, sender.as_str()),
+	// 	fetch_state(&StateEventType::RoomMember, sender.as_str()),
-	)
+	// )
-	.await;
+	// .await;
+
+	let room_create_event = fetch_state(&StateEventType::RoomCreate, "").await;
+	let power_levels_event = fetch_state(&StateEventType::RoomPowerLevels, "").await;
+	let sender_member_event = fetch_state(&StateEventType::RoomMember, sender.as_str()).await;
 
 	let room_create_event = match room_create_event {
 		| None => {
-			warn!("no m.room.create event in auth chain");
+			error!(
+				create_event = room_create_event.as_ref().map(Event::event_id).unwrap_or(<&EventId>::try_from("$unknown").unwrap()).as_str(),
+				power_levels = power_levels_event.as_ref().map(Event::event_id).unwrap_or(<&EventId>::try_from("$unknown").unwrap()).as_str(),
+				member_event = sender_member_event.as_ref().map(Event::event_id).unwrap_or(<&EventId>::try_from("$unknown").unwrap()).as_str(),
+				"no m.room.create event found for {} ({})!",
+				incoming_event.event_id().as_str(),
+				incoming_event.room_id().as_str()
+			);
 			return Ok(false);
 		},
 		| Some(e) => e,
 	};
+	// just re-check 1.2 to work around a bug
+	let Some(room_id_server_name) = incoming_event.room_id().server_name() else {
+		warn!("room ID has no servername");
+		return Ok(false);
+	};
+
+	if room_id_server_name != room_create_event.sender().server_name() {
+		warn!(
+			"servername of room ID origin ({}) does not match servername of m.room.create sender ({})",
+			room_id_server_name,
+			room_create_event.sender().server_name());
+		return Ok(false);
+	}
 
 	if incoming_event.room_id() != room_create_event.room_id() {
 		warn!("room_id of incoming event does not match room_id of m.room.create event");

src/core/matrix/state_res/mod.rs

@@ -733,8 +733,12 @@ where
 	Fut: Future<Output = Option<E>> + Send,
 	E: Event + Send + Sync,
 {
+	let mut room_id = None;
 	while let Some(sort_ev) = event {
 		debug!(event_id = sort_ev.event_id().as_str(), "mainline");
+		if room_id.is_none() {
+			room_id = Some(sort_ev.room_id().to_owned());
+		}
 
 		let id = sort_ev.event_id();
 		if let Some(depth) = mainline_map.get(id) {
@@ -753,7 +757,7 @@ where
 			}
 		}
 	}
-	// Did not find a power level event so we default to zero
+	warn!("could not find a power event in the mainline map for {room_id:?}, defaulting to zero depth");
 	Ok(0)
 }
 

src/database/engine/cf_opts.rs

@@ -29,7 +29,7 @@ fn descriptor_cf_options(
 	set_table_options(&mut opts, &desc, cache)?;
 
 	opts.set_min_write_buffer_number(1);
-	opts.set_max_write_buffer_number(2);
+	opts.set_max_write_buffer_number(3);
 	opts.set_write_buffer_size(desc.write_size);
 
 	opts.set_target_file_size_base(desc.file_size);

src/database/engine/db_opts.rs

@@ -1,8 +1,6 @@
-use std::{cmp, convert::TryFrom};
+use conduwuit::{Config, Result};
-
-use conduwuit::{Config, Result, utils};
 use rocksdb::{Cache, DBRecoveryMode, Env, LogLevel, Options, statistics::StatsLevel};
-
+use conduwuit::config::{parallelism_scaled_i32, parallelism_scaled_u32};
 use super::{cf_opts::cache_size_f64, logger::handle as handle_log};
 
 /// Create database-wide options suitable for opening the database. This also
@@ -23,8 +21,8 @@ pub(crate) fn db_options(config: &Config, env: &Env, row_cache: &Cache) -> Resul
 	set_logging_defaults(&mut opts, config);
 
 	// Processing
-	opts.set_max_background_jobs(num_threads::<i32>(config)?);
+	opts.set_max_background_jobs(parallelism_scaled_i32(1));
-	opts.set_max_subcompactions(num_threads::<u32>(config)?);
+	opts.set_max_subcompactions(parallelism_scaled_u32(1));
 	opts.set_avoid_unnecessary_blocking_io(true);
 	opts.set_max_file_opening_threads(0);
 
@@ -126,15 +124,3 @@ fn set_logging_defaults(opts: &mut Options, config: &Config) {
 		opts.set_callback_logger(rocksdb_log_level, &handle_log);
 	}
 }
-
-fn num_threads<T: TryFrom<usize>>(config: &Config) -> Result<T> {
-	const MIN_PARALLELISM: usize = 2;
-
-	let requested = if config.rocksdb_parallelism_threads != 0 {
-		config.rocksdb_parallelism_threads
-	} else {
-		utils::available_parallelism()
-	};
-
-	utils::math::try_into::<T, usize>(cmp::max(MIN_PARALLELISM, requested))
-}

src/service/presence/mod.rs

@@ -100,7 +100,7 @@ impl Service {
 	/// Pings the presence of the given user in the given room, setting the
 	/// specified state.
 	pub async fn ping_presence(&self, user_id: &UserId, new_state: &PresenceState) -> Result<()> {
-		const REFRESH_TIMEOUT: u64 = 60 * 1000;
+		const REFRESH_TIMEOUT: u64 = 60 * 1000 * 4;
 
 		let last_presence = self.db.get_presence(user_id).await;
 		let state_changed = match last_presence {

src/service/resolver/actual.rs

@@ -306,14 +306,12 @@ impl super::Service {
 
 	#[tracing::instrument(name = "srv", level = "debug", skip(self))]
 	async fn query_srv_record(&self, hostname: &'_ str) -> Result<Option<FedDest>> {
-		let hostnames =
-			[format!("_matrix-fed._tcp.{hostname}."), format!("_matrix._tcp.{hostname}.")];
-
-		for hostname in hostnames {
 		self.services.server.check_running()?;
 
 		debug!("querying SRV for {hostname:?}");
-			let hostname = hostname.trim_end_matches('.');
+
+		let hostname_suffix = format!("_matrix-fed._tcp.{hostname}.");
+		let hostname = hostname_suffix.trim_end_matches('.');
 		match self.resolver.resolver.srv_lookup(hostname).await {
 			| Err(e) => Self::handle_resolve_error(&e, hostname)?,
 			| Ok(result) => {
@@ -328,7 +326,6 @@ impl super::Service {
 				}));
 			},
 		}
-		}
 
 		Ok(None)
 	}

src/service/resolver/dns.rs

@@ -53,9 +53,9 @@ impl Resolver {
 		opts.cache_size = config.dns_cache_entries as usize;
 		opts.preserve_intermediates = true;
 		opts.negative_min_ttl = Some(Duration::from_secs(config.dns_min_ttl_nxdomain));
-		opts.negative_max_ttl = Some(Duration::from_secs(60 * 60 * 24 * 30));
+		opts.negative_max_ttl = Some(Duration::from_secs(60 * 60 * 24));
 		opts.positive_min_ttl = Some(Duration::from_secs(config.dns_min_ttl));
-		opts.positive_max_ttl = Some(Duration::from_secs(60 * 60 * 24 * 7));
+		opts.positive_max_ttl = Some(Duration::from_secs(60 * 60 * 24));
 		opts.timeout = Duration::from_secs(config.dns_timeout);
 		opts.attempts = config.dns_attempts as usize;
 		opts.try_tcp_on_error = config.dns_tcp_fallback;

src/service/rooms/event_handler/fetch_and_handle_outliers.rs

@@ -79,7 +79,7 @@ where
 			{
 				// Exponential backoff
 				const MIN_DURATION: u64 = 60 * 2;
-				const MAX_DURATION: u64 = 60 * 60 * 8;
+				const MAX_DURATION: u64 = 60 * 60;
 				if continue_exponential_backoff_secs(
 					MIN_DURATION,
 					MAX_DURATION,

src/service/rooms/event_handler/handle_outlier_pdu.rs

@@ -122,10 +122,7 @@ where
 	}
 
 	// The original create event must be in the auth events
-	if !matches!(
+	if !auth_events.contains_key(&(StateEventType::RoomCreate, String::new().into())) {
-		auth_events.get(&(StateEventType::RoomCreate, String::new().into())),
-		Some(_) | None
-	) {
 		return Err!(Request(InvalidParam("Incoming event refers to wrong create event.")));
 	}
 

src/service/rooms/event_handler/handle_prev_pdu.rs

@@ -46,7 +46,7 @@ where
 	{
 		// Exponential backoff
 		const MIN_DURATION: u64 = 5 * 60;
-		const MAX_DURATION: u64 = 60 * 60 * 24;
+		const MAX_DURATION: u64 = 60 * 60;
 		if continue_exponential_backoff_secs(MIN_DURATION, MAX_DURATION, time.elapsed(), *tries) {
 			debug!(
 				?tries,

src/service/rooms/event_handler/upgrade_outlier_pdu.rs

@@ -5,7 +5,7 @@ use conduwuit::{
 	matrix::{Event, EventTypeExt, PduEvent, StateKey, state_res},
 	trace,
 	utils::stream::{BroadbandExt, ReadyExt},
-	warn,
+	warn
 };
 use futures::{FutureExt, StreamExt, future::ready};
 use ruma::{CanonicalJsonValue, RoomId, ServerName, events::StateEventType};
@@ -175,7 +175,7 @@ where
 	let extremities: Vec<_> = self
 		.services
 		.state
-		.get_forward_extremities(room_id)
+		.get_forward_extremities(room_id, &state_lock)
 		.map(ToOwned::to_owned)
 		.ready_filter(|event_id| {
 			// Remove any that are referenced by this incoming event's prev_events
@@ -193,6 +193,8 @@ where
 		.collect()
 		.await;
 
+	if extremities.len() == 0 { info!("Retained zero extremities when upgrading outlier PDU to timeline PDU with {} previous events, event id: {}", incoming_pdu.prev_events.len(), incoming_pdu.event_id) }
+	
 	debug!(
 		"Retained {} extremities checked against {} prev_events",
 		extremities.len(),

src/service/rooms/state/mod.rs

@@ -388,6 +388,7 @@ impl Service {
 	pub fn get_forward_extremities<'a>(
 		&'a self,
 		room_id: &'a RoomId,
+		_state_lock: &'a RoomMutexGuard,
 	) -> impl Stream<Item = &EventId> + Send + '_ {
 		let prefix = (room_id, Interfix);
 

src/service/rooms/timeline/create.rs

@@ -42,7 +42,7 @@ pub async fn create_hash_and_sign_event(
 	let prev_events: Vec<OwnedEventId> = self
 		.services
 		.state
-		.get_forward_extremities(room_id)
+		.get_forward_extremities(room_id, _mutex_lock)
 		.take(20)
 		.map(Into::into)
 		.collect()

src/service/sending/mod.rs

@@ -401,16 +401,10 @@ impl Service {
 
 fn num_senders(args: &crate::Args<'_>) -> usize {
 	const MIN_SENDERS: usize = 1;
-	// Limit the number of senders to the number of workers threads or number of
+	// Limit the maximum number of senders to the number of cores.
-	// cores, conservatively.
+	let max_senders = available_parallelism();
-	let max_senders = args
-		.server
-		.metrics
-		.num_workers()
-		.min(available_parallelism());
 
-	// If the user doesn't override the default 0, this is intended to then default
+	// default is 4 senders. clamp between 1 and core count.
-	// to 1 for now as multiple senders is experimental.
 	args.server
 		.config
 		.sender_workers