improvement: efficient /sync, mutex for federation transactions

Conduit did not send the event in /sync because of a race condition.
There is a brief moment in time where Conduit accepted the event, but
did not store it yet. So when the client /syncs it updates the since
token without sending the event. I hope I fixed it by significantly
shortening the race-condition period.

src/appservice_server.rs

@@ -46,7 +46,10 @@ where
     *reqwest_request.timeout_mut() = Some(Duration::from_secs(30));
 
     let url = reqwest_request.url().clone();
-    let mut response = globals.reqwest_client().execute(reqwest_request).await?;
+    let mut response = globals
+        .reqwest_client()
+        .execute(reqwest_request)
+        .await?;
 
     // reqwest::Response -> http::Response conversion
     let status = response.status();

src/client_server/membership.rs

@@ -615,16 +615,9 @@ async fn join_room_by_id_helper(
         // pdu without it's state. This is okay because append_pdu can't fail.
         let statehashid = db.rooms.append_to_state(&pdu, &db.globals)?;
 
-        let count = db.globals.next_count()?;
-        let mut pdu_id = room_id.as_bytes().to_vec();
-        pdu_id.push(0xff);
-        pdu_id.extend_from_slice(&count.to_be_bytes());
-
         db.rooms.append_pdu(
             &pdu,
             utils::to_canonical_object(&pdu).expect("Pdu is valid canonical object"),
-            count,
-            &pdu_id,
             &[pdu.event_id.clone()],
             db,
         )?;
@@ -905,19 +898,37 @@ pub async fn invite_helper(
         )
         .map_err(|_| Error::BadRequest(ErrorKind::InvalidParam, "Origin field is invalid."))?;
 
-        let pdu_id =
+        let mutex = Arc::clone(
-            server_server::handle_incoming_pdu(&origin, &event_id, value, true, &db, &pub_key_map)
+            db.globals
-                .await
+                .roomid_mutex
-                .map_err(|_| {
+                .write()
-                    Error::BadRequest(
+                .unwrap()
-                        ErrorKind::InvalidParam,
+                .entry(room_id.clone())
-                        "Error while handling incoming PDU.",
+                .or_default(),
-                    )
+        );
-                })?
+        let mutex_lock = mutex.lock().await;
-                .ok_or(Error::BadRequest(
+
-                    ErrorKind::InvalidParam,
+        let pdu_id = server_server::handle_incoming_pdu(
-                    "Could not accept incoming PDU as timeline event.",
+            &origin,
-                ))?;
+            &event_id,
+            &room_id,
+            value,
+            true,
+            &db,
+            &pub_key_map,
+        )
+        .await
+        .map_err(|_| {
+            Error::BadRequest(
+                ErrorKind::InvalidParam,
+                "Error while handling incoming PDU.",
+            )
+        })?
+        .ok_or(Error::BadRequest(
+            ErrorKind::InvalidParam,
+            "Could not accept incoming PDU as timeline event.",
+        ))?;
+        drop(mutex_lock);
 
         for server in db
             .rooms

src/client_server/sync.rs

@@ -88,7 +88,9 @@ pub async fn sync_events_route(
 
     let we_have_to_wait = rx.borrow().is_none();
     if we_have_to_wait {
-        let _ = rx.changed().await;
+        if let Err(e) = rx.changed().await {
+            error!("Error waiting for sync: {}", e);
+        }
     }
 
     let result = match rx
@@ -222,13 +224,16 @@ async fn sync_helper(
 
         // Database queries:
 
-        let current_shortstatehash = db.rooms.current_shortstatehash(&room_id)?;
+        let current_shortstatehash = db
+            .rooms
+            .current_shortstatehash(&room_id)?
+            .expect("All rooms have state");
 
-        // These type is Option<Option<_>>. The outer Option is None when there is no event between
+        let first_pdu_before_since = db
-        // since and the current room state, meaning there should be no updates.
+            .rooms
-        // The inner Option is None when there is an event, but there is no state hash associated
+            .pdus_until(&sender_user, &room_id, since)
-        // with it. This can happen for the RoomCreate event, so all updates should arrive.
+            .next()
-        let first_pdu_before_since = db.rooms.pdus_until(&sender_user, &room_id, since).next();
+            .transpose()?;
 
         let pdus_after_since = db
             .rooms
@@ -236,11 +241,78 @@ async fn sync_helper(
             .next()
             .is_some();
 
-        let since_shortstatehash = first_pdu_before_since.as_ref().map(|pdu| {
+        let since_shortstatehash = first_pdu_before_since
-            db.rooms
+            .as_ref()
-                .pdu_shortstatehash(&pdu.as_ref().ok()?.1.event_id)
+            .map(|pdu| {
-                .ok()?
+                db.rooms
-        });
+                    .pdu_shortstatehash(&pdu.1.event_id)
+                    .transpose()
+                    .expect("all pdus have state")
+            })
+            .transpose()?;
+
+        // Calculates joined_member_count, invited_member_count and heroes
+        let calculate_counts = || {
+            let joined_member_count = db.rooms.room_members(&room_id).count();
+            let invited_member_count = db.rooms.room_members_invited(&room_id).count();
+
+            // Recalculate heroes (first 5 members)
+            let mut heroes = Vec::new();
+
+            if joined_member_count + invited_member_count <= 5 {
+                // Go through all PDUs and for each member event, check if the user is still joined or
+                // invited until we have 5 or we reach the end
+
+                for hero in db
+                    .rooms
+                    .all_pdus(&sender_user, &room_id)
+                    .filter_map(|pdu| pdu.ok()) // Ignore all broken pdus
+                    .filter(|(_, pdu)| pdu.kind == EventType::RoomMember)
+                    .map(|(_, pdu)| {
+                        let content = serde_json::from_value::<
+                            ruma::events::room::member::MemberEventContent,
+                        >(pdu.content.clone())
+                        .map_err(|_| Error::bad_database("Invalid member event in database."))?;
+
+                        if let Some(state_key) = &pdu.state_key {
+                            let user_id = UserId::try_from(state_key.clone()).map_err(|_| {
+                                Error::bad_database("Invalid UserId in member PDU.")
+                            })?;
+
+                            // The membership was and still is invite or join
+                            if matches!(
+                                content.membership,
+                                MembershipState::Join | MembershipState::Invite
+                            ) && (db.rooms.is_joined(&user_id, &room_id)?
+                                || db.rooms.is_invited(&user_id, &room_id)?)
+                            {
+                                Ok::<_, Error>(Some(state_key.clone()))
+                            } else {
+                                Ok(None)
+                            }
+                        } else {
+                            Ok(None)
+                        }
+                    })
+                    // Filter out buggy users
+                    .filter_map(|u| u.ok())
+                    // Filter for possible heroes
+                    .flatten()
+                {
+                    if heroes.contains(&hero) || hero == sender_user.as_str() {
+                        continue;
+                    }
+
+                    heroes.push(hero);
+                }
+            }
+
+            (
+                Some(joined_member_count),
+                Some(invited_member_count),
+                heroes,
+            )
+        };
 
         let (
             heroes,
@@ -248,63 +320,107 @@ async fn sync_helper(
             invited_member_count,
             joined_since_last_sync,
             state_events,
-        ) = if pdus_after_since && Some(current_shortstatehash) != since_shortstatehash {
+        ) = if since_shortstatehash.is_none() {
-            let current_state = db.rooms.room_state_full(&room_id)?;
+            // Probably since = 0, we will do an initial sync
-            let current_members = current_state
+            let (joined_member_count, invited_member_count, heroes) = calculate_counts();
-                .iter()
-                .filter(|(key, _)| key.0 == EventType::RoomMember)
-                .map(|(key, value)| (&key.1, value)) // Only keep state key
-                .collect::<Vec<_>>();
-            let encrypted_room = current_state
-                .get(&(EventType::RoomEncryption, "".to_owned()))
-                .is_some();
-            let since_state = since_shortstatehash
-                .as_ref()
-                .map(|since_shortstatehash| {
-                    since_shortstatehash
-                        .map(|since_shortstatehash| db.rooms.state_full(since_shortstatehash))
-                        .transpose()
-                })
-                .transpose()?;
 
-            let since_encryption = since_state.as_ref().map(|state| {
+            let current_state_ids = db.rooms.state_full_ids(current_shortstatehash)?;
-                state
+            let state_events = current_state_ids
-                    .as_ref()
+                .iter()
-                    .map(|state| state.get(&(EventType::RoomEncryption, "".to_owned())))
+                .map(|id| db.rooms.get_pdu(id))
-            });
+                .filter_map(|r| r.ok().flatten())
+                .collect::<Vec<_>>();
+
+            (
+                heroes,
+                joined_member_count,
+                invited_member_count,
+                true,
+                state_events
+            )
+        } else if !pdus_after_since || since_shortstatehash == Some(current_shortstatehash) {
+            // No state changes
+            (Vec::new(), None, None, false, Vec::new())
+        } else {
+            // Incremental /sync
+            let since_shortstatehash = since_shortstatehash.unwrap();
+
+            let since_sender_member = db
+                .rooms
+                .state_get(
+                    since_shortstatehash,
+                    &EventType::RoomMember,
+                    sender_user.as_str(),
+                )?
+                .and_then(|pdu| {
+                    serde_json::from_value::<Raw<ruma::events::room::member::MemberEventContent>>(
+                        pdu.content.clone(),
+                    )
+                    .expect("Raw::from_value always works")
+                    .deserialize()
+                    .map_err(|_| Error::bad_database("Invalid PDU in database."))
+                    .ok()
+                });
+
+            let joined_since_last_sync = since_sender_member
+                .map_or(true, |member| member.membership != MembershipState::Join);
+
+            let current_state_ids = db.rooms.state_full_ids(current_shortstatehash)?;
+
+            let since_state_ids = db.rooms.state_full_ids(since_shortstatehash)?;
+
+            let state_events = if joined_since_last_sync {
+                current_state_ids
+                    .iter()
+                    .map(|id| db.rooms.get_pdu(id))
+                    .filter_map(|r| r.ok().flatten())
+                    .collect::<Vec<_>>()
+            } else {
+                current_state_ids
+                    .difference(&since_state_ids)
+                    .filter(|id| {
+                        !timeline_pdus
+                            .iter()
+                            .any(|(_, timeline_pdu)| timeline_pdu.event_id == **id)
+                    })
+                    .map(|id| db.rooms.get_pdu(id))
+                    .filter_map(|r| r.ok().flatten())
+                    .collect()
+            };
+
+            let encrypted_room = db
+                .rooms
+                .state_get(current_shortstatehash, &EventType::RoomEncryption, "")?
+                .is_some();
+
+            let since_encryption =
+                db.rooms
+                    .state_get(since_shortstatehash, &EventType::RoomEncryption, "")?;
 
             // Calculations:
-            let new_encrypted_room =
+            let new_encrypted_room = encrypted_room && since_encryption.is_none();
-                encrypted_room && since_encryption.map_or(true, |encryption| encryption.is_none());
 
-            let send_member_count = since_state.as_ref().map_or(true, |since_state| {
+            let send_member_count = state_events
-                since_state.as_ref().map_or(true, |since_state| {
+                .iter()
-                    current_members.len()
+                .any(|event| event.kind == EventType::RoomMember);
-                        != since_state
-                            .iter()
-                            .filter(|(key, _)| key.0 == EventType::RoomMember)
-                            .count()
-                })
-            });
-
-            let since_sender_member = since_state.as_ref().map(|since_state| {
-                since_state.as_ref().and_then(|state| {
-                    state
-                        .get(&(EventType::RoomMember, sender_user.as_str().to_owned()))
-                        .and_then(|pdu| {
-                            serde_json::from_value::<
-                                Raw<ruma::events::room::member::MemberEventContent>,
-                            >(pdu.content.clone())
-                            .expect("Raw::from_value always works")
-                            .deserialize()
-                            .map_err(|_| Error::bad_database("Invalid PDU in database."))
-                            .ok()
-                        })
-                })
-            });
 
             if encrypted_room {
-                for (user_id, current_member) in current_members {
+                for (user_id, current_member) in db
+                    .rooms
+                    .room_members(&room_id)
+                    .filter_map(|r| r.ok())
+                    .filter_map(|user_id| {
+                        db.rooms
+                            .state_get(
+                                current_shortstatehash,
+                                &EventType::RoomMember,
+                                user_id.as_str(),
+                            )
+                            .ok()
+                            .flatten()
+                            .map(|current_member| (user_id, current_member))
+                    })
+                {
                     let current_membership = serde_json::from_value::<
                         Raw<ruma::events::room::member::MemberEventContent>,
                     >(current_member.content.clone())
@@ -313,31 +429,23 @@ async fn sync_helper(
                     .map_err(|_| Error::bad_database("Invalid PDU in database."))?
                     .membership;
 
-                    let since_membership =
+                    let since_membership = db
-                        since_state
+                        .rooms
-                            .as_ref()
+                        .state_get(
-                            .map_or(MembershipState::Leave, |since_state| {
+                            since_shortstatehash,
-                                since_state
+                            &EventType::RoomMember,
-                                    .as_ref()
+                            user_id.as_str(),
-                                    .and_then(|since_state| {
+                        )?
-                                        since_state
+                        .and_then(|since_member| {
-                                            .get(&(EventType::RoomMember, user_id.clone()))
+                            serde_json::from_value::<
-                                            .and_then(|since_member| {
+                                Raw<ruma::events::room::member::MemberEventContent>,
-                                                serde_json::from_value::<
+                            >(since_member.content.clone())
-                                                Raw<ruma::events::room::member::MemberEventContent>,
+                            .expect("Raw::from_value always works")
-                                            >(
+                            .deserialize()
-                                                since_member.content.clone()
+                            .map_err(|_| Error::bad_database("Invalid PDU in database."))
-                                            )
+                            .ok()
-                                            .expect("Raw::from_value always works")
+                        })
-                                            .deserialize()
+                        .map_or(MembershipState::Leave, |member| member.membership);
-                                            .map_err(|_| {
-                                                Error::bad_database("Invalid PDU in database.")
-                                            })
-                                            .ok()
-                                            })
-                                    })
-                                    .map_or(MembershipState::Leave, |member| member.membership)
-                            });
 
                     let user_id = UserId::try_from(user_id.clone())
                         .map_err(|_| Error::bad_database("Invalid UserId in member PDU."))?;
@@ -359,10 +467,6 @@ async fn sync_helper(
                 }
             }
 
-            let joined_since_last_sync = since_sender_member.map_or(true, |member| {
-                member.map_or(true, |member| member.membership != MembershipState::Join)
-            });
-
             if joined_since_last_sync && encrypted_room || new_encrypted_room {
                 // If the user is in a new encrypted room, give them all joined users
                 device_list_updates.extend(
@@ -382,100 +486,11 @@ async fn sync_helper(
             }
 
             let (joined_member_count, invited_member_count, heroes) = if send_member_count {
-                let joined_member_count = db.rooms.room_members(&room_id).count();
+                calculate_counts()
-                let invited_member_count = db.rooms.room_members_invited(&room_id).count();
-
-                // Recalculate heroes (first 5 members)
-                let mut heroes = Vec::new();
-
-                if joined_member_count + invited_member_count <= 5 {
-                    // Go through all PDUs and for each member event, check if the user is still joined or
-                    // invited until we have 5 or we reach the end
-
-                    for hero in db
-                        .rooms
-                        .all_pdus(&sender_user, &room_id)
-                        .filter_map(|pdu| pdu.ok()) // Ignore all broken pdus
-                        .filter(|(_, pdu)| pdu.kind == EventType::RoomMember)
-                        .map(|(_, pdu)| {
-                            let content = serde_json::from_value::<
-                                ruma::events::room::member::MemberEventContent,
-                            >(pdu.content.clone())
-                            .map_err(|_| {
-                                Error::bad_database("Invalid member event in database.")
-                            })?;
-
-                            if let Some(state_key) = &pdu.state_key {
-                                let user_id =
-                                    UserId::try_from(state_key.clone()).map_err(|_| {
-                                        Error::bad_database("Invalid UserId in member PDU.")
-                                    })?;
-
-                                // The membership was and still is invite or join
-                                if matches!(
-                                    content.membership,
-                                    MembershipState::Join | MembershipState::Invite
-                                ) && (db.rooms.is_joined(&user_id, &room_id)?
-                                    || db.rooms.is_invited(&user_id, &room_id)?)
-                                {
-                                    Ok::<_, Error>(Some(state_key.clone()))
-                                } else {
-                                    Ok(None)
-                                }
-                            } else {
-                                Ok(None)
-                            }
-                        })
-                        // Filter out buggy users
-                        .filter_map(|u| u.ok())
-                        // Filter for possible heroes
-                        .flatten()
-                    {
-                        if heroes.contains(&hero) || hero == sender_user.as_str() {
-                            continue;
-                        }
-
-                        heroes.push(hero);
-                    }
-                }
-
-                (
-                    Some(joined_member_count),
-                    Some(invited_member_count),
-                    heroes,
-                )
             } else {
                 (None, None, Vec::new())
             };
 
-            let state_events = if joined_since_last_sync {
-                current_state
-                    .iter()
-                    .map(|(_, pdu)| pdu.to_sync_state_event())
-                    .collect()
-            } else {
-                match since_state {
-                    None => Vec::new(),
-                    Some(Some(since_state)) => current_state
-                        .iter()
-                        .filter(|(key, value)| {
-                            since_state.get(key).map(|e| &e.event_id) != Some(&value.event_id)
-                        })
-                        .filter(|(_, value)| {
-                            !timeline_pdus.iter().any(|(_, timeline_pdu)| {
-                                timeline_pdu.kind == value.kind
-                                    && timeline_pdu.state_key == value.state_key
-                            })
-                        })
-                        .map(|(_, pdu)| pdu.to_sync_state_event())
-                        .collect(),
-                    Some(None) => current_state
-                        .iter()
-                        .map(|(_, pdu)| pdu.to_sync_state_event())
-                        .collect(),
-                }
-            };
-
             (
                 heroes,
                 joined_member_count,
@@ -483,8 +498,6 @@ async fn sync_helper(
                 joined_since_last_sync,
                 state_events,
             )
-        } else {
-            (Vec::new(), None, None, false, Vec::new())
         };
 
         // Look for device list updates in this room
@@ -575,7 +588,10 @@ async fn sync_helper(
                 events: room_events,
             },
             state: sync_events::State {
-                events: state_events,
+                events: state_events
+                    .iter()
+                    .map(|pdu| pdu.to_sync_state_event())
+                    .collect(),
             },
             ephemeral: sync_events::Ephemeral { events: edus },
         };

src/client_server/to_device.rs

@@ -22,7 +22,9 @@ pub async fn send_event_to_device_route(
     let sender_user = body.sender_user.as_ref().expect("user is authenticated");
     let sender_device = body.sender_device.as_deref();
 
+    // TODO: uncomment when https://github.com/vector-im/element-android/issues/3589 is solved
     // Check if this is a new transaction id
+    /*
     if db
         .transaction_ids
         .existing_txnid(sender_user, sender_device, &body.txn_id)?
@@ -30,6 +32,7 @@ pub async fn send_event_to_device_route(
     {
         return Ok(send_event_to_device::Response.into());
     }
+    */
 
     for (target_user_id, map) in &body.messages {
         for (target_device_id_maybe, event) in map {

src/database/abstraction.rs

@@ -111,7 +111,7 @@ impl Tree for SledEngineTree {
         backwards: bool,
     ) -> Box<dyn Iterator<Item = (Box<[u8]>, Box<[u8]>)>> {
         let iter = if backwards {
-            self.0.range(..from)
+            self.0.range(..=from)
         } else {
             self.0.range(from..)
         };

src/database/globals.rs

@@ -1,12 +1,9 @@
 use crate::{database::Config, utils, ConduitResult, Error, Result};
 use log::{error, info};
-use ruma::{
+use ruma::{DeviceId, EventId, MilliSecondsSinceUnixEpoch, RoomId, ServerName, ServerSigningKeyId, UserId, api::{
-    api::{
         client::r0::sync::sync_events,
         federation::discovery::{ServerSigningKeys, VerifyKey},
-    },
+    }};
-    DeviceId, EventId, MilliSecondsSinceUnixEpoch, ServerName, ServerSigningKeyId, UserId,
-};
 use rustls::{ServerCertVerifier, WebPKIVerifier};
 use std::{
     collections::{BTreeMap, HashMap},
@@ -15,7 +12,7 @@ use std::{
     sync::{Arc, RwLock},
     time::{Duration, Instant},
 };
-use tokio::sync::Semaphore;
+use tokio::sync::{Mutex, Semaphore};
 use trust_dns_resolver::TokioAsyncResolver;
 
 use super::abstraction::Tree;
@@ -38,6 +35,7 @@ pub struct Globals {
     pub bad_event_ratelimiter: Arc<RwLock<BTreeMap<EventId, RateLimitState>>>,
     pub bad_signature_ratelimiter: Arc<RwLock<BTreeMap<Vec<String>, RateLimitState>>>,
     pub servername_ratelimiter: Arc<RwLock<BTreeMap<Box<ServerName>, Arc<Semaphore>>>>,
+    pub roomid_mutex: RwLock<BTreeMap<RoomId, Arc<Mutex<()>>>>,
     pub sync_receivers: RwLock<
         BTreeMap<
             (UserId, Box<DeviceId>),
@@ -165,6 +163,7 @@ impl Globals {
             bad_event_ratelimiter: Arc::new(RwLock::new(BTreeMap::new())),
             bad_signature_ratelimiter: Arc::new(RwLock::new(BTreeMap::new())),
             servername_ratelimiter: Arc::new(RwLock::new(BTreeMap::new())),
+            roomid_mutex: RwLock::new(BTreeMap::new()),
             sync_receivers: RwLock::new(BTreeMap::new()),
         };
 

src/database/rooms.rs

@@ -20,12 +20,7 @@ use ruma::{
     state_res::{self, Event, RoomVersion, StateMap},
     uint, EventId, RoomAliasId, RoomId, RoomVersionId, ServerName, UserId,
 };
-use std::{
+use std::{collections::{BTreeMap, BTreeSet, HashMap, HashSet}, convert::{TryFrom, TryInto}, mem, sync::{Arc, RwLock}};
-    collections::{BTreeMap, HashMap, HashSet},
-    convert::{TryFrom, TryInto},
-    mem,
-    sync::{Arc, RwLock},
-};
 
 use super::{abstraction::Tree, admin::AdminCommand, pusher};
 
@@ -89,7 +84,7 @@ pub struct Rooms {
 impl Rooms {
     /// Builds a StateMap by iterating over all keys that start
     /// with state_hash, this gives the full state for the given state_hash.
-    pub fn state_full_ids(&self, shortstatehash: u64) -> Result<Vec<EventId>> {
+    pub fn state_full_ids(&self, shortstatehash: u64) -> Result<BTreeSet<EventId>> {
         Ok(self
             .stateid_shorteventid
             .scan_prefix(shortstatehash.to_be_bytes().to_vec())
@@ -668,11 +663,10 @@ impl Rooms {
         &self,
         pdu: &PduEvent,
         mut pdu_json: CanonicalJsonObject,
-        count: u64,
-        pdu_id: &[u8],
         leaves: &[EventId],
         db: &Database,
-    ) -> Result<()> {
+    ) -> Result<Vec<u8>> {
+        // returns pdu id
         // Make unsigned fields correct. This is not properly documented in the spec, but state
         // events need to have previous content in the unsigned field, so clients can easily
         // interpret things like membership changes
@@ -710,20 +704,30 @@ impl Rooms {
 
         self.replace_pdu_leaves(&pdu.room_id, leaves)?;
 
+        let count1 = db.globals.next_count()?;
         // Mark as read first so the sending client doesn't get a notification even if appending
         // fails
         self.edus
-            .private_read_set(&pdu.room_id, &pdu.sender, count, &db.globals)?;
+            .private_read_set(&pdu.room_id, &pdu.sender, count1, &db.globals)?;
         self.reset_notification_counts(&pdu.sender, &pdu.room_id)?;
 
+        let count2 = db.globals.next_count()?;
+        let mut pdu_id = pdu.room_id.as_bytes().to_vec();
+        pdu_id.push(0xff);
+        pdu_id.extend_from_slice(&count2.to_be_bytes());
+
+        // There's a brief moment of time here where the count is updated but the pdu does not
+        // exist. This could theoretically lead to dropped pdus, but it's extremely rare
+
         self.pduid_pdu.insert(
-            pdu_id,
+            &pdu_id,
             &serde_json::to_vec(&pdu_json).expect("CanonicalJsonObject is always a valid"),
         )?;
 
         // This also replaces the eventid of any outliers with the correct
         // pduid, removing the place holder.
-        self.eventid_pduid.insert(pdu.event_id.as_bytes(), pdu_id)?;
+        self.eventid_pduid
+            .insert(pdu.event_id.as_bytes(), &pdu_id)?;
 
         // See if the event matches any known pushers
         for user in db
@@ -911,7 +915,7 @@ impl Rooms {
             _ => {}
         }
 
-        Ok(())
+        Ok(pdu_id)
     }
 
     pub fn reset_notification_counts(&self, user_id: &UserId, room_id: &RoomId) -> Result<()> {
@@ -1208,6 +1212,7 @@ impl Rooms {
             state_key,
             redacts,
         } = pdu_builder;
+
         // TODO: Make sure this isn't called twice in parallel
         let prev_events = self
             .get_pdu_leaves(&room_id)?
@@ -1356,11 +1361,9 @@ impl Rooms {
         // pdu without it's state. This is okay because append_pdu can't fail.
         let statehashid = self.append_to_state(&pdu, &db.globals)?;
 
-        self.append_pdu(
+        let pdu_id = self.append_pdu(
             &pdu,
             pdu_json,
-            count,
-            &pdu_id,
             // Since this PDU references all pdu_leaves we can update the leaves
             // of the room
             &[pdu.event_id.clone()],
@@ -1497,7 +1500,7 @@ impl Rooms {
         prefix.push(0xff);
 
         let mut current = prefix.clone();
-        current.extend_from_slice(&until.to_be_bytes());
+        current.extend_from_slice(&(until.saturating_sub(1)).to_be_bytes()); // -1 because we don't want event at `until`
 
         let current: &[u8] = &current;
 

src/database/users.rs

@@ -726,10 +726,9 @@ impl Users {
         json.insert("sender".to_owned(), sender.to_string().into());
         json.insert("content".to_owned(), content);
 
-        self.todeviceid_events.insert(
+        let value = serde_json::to_vec(&json).expect("Map::to_vec always works");
-            &key,
+
-            &serde_json::to_vec(&json).expect("Map::to_vec always works"),
+        self.todeviceid_events.insert(&key, &value)?;
-        )?;
 
         Ok(())
     }
@@ -774,7 +773,7 @@ impl Users {
 
         for (key, _) in self
             .todeviceid_events
-            .iter_from(&last, true)
+            .iter_from(&last, true) // this includes last
             .take_while(move |(k, _)| k.starts_with(&prefix))
             .map(|(key, _)| {
                 Ok::<_, Error>((

src/ruma_wrapper.rs

@@ -318,6 +318,7 @@ where
             }),
             Err(e) => {
                 warn!("{:?}", e);
+                // Bad Json
                 Failure((Status::raw(583), ()))
             }
         }

src/server_server.rs

@@ -624,13 +624,44 @@ pub async fn send_transaction_message_route(
             }
         };
 
+        // 0. Check the server is in the room
+        let room_id = match value
+            .get("room_id")
+            .and_then(|id| RoomId::try_from(id.as_str()?).ok())
+        {
+            Some(id) => id,
+            None => {
+                // Event is invalid
+                resolved_map.insert(event_id, Err("Event needs a valid RoomId.".to_string()));
+                continue;
+            }
+        };
+
+        let mutex = Arc::clone(
+            db.globals
+                .roomid_mutex
+                .write()
+                .unwrap()
+                .entry(room_id.clone())
+                .or_default(),
+        );
+        let mutex_lock = mutex.lock().await;
         let start_time = Instant::now();
         resolved_map.insert(
             event_id.clone(),
-            handle_incoming_pdu(&body.origin, &event_id, value, true, &db, &pub_key_map)
+            handle_incoming_pdu(
-                .await
+                &body.origin,
-                .map(|_| ()),
+                &event_id,
+                &room_id,
+                value,
+                true,
+                &db,
+                &pub_key_map,
+            )
+            .await
+            .map(|_| ()),
         );
+        drop(mutex_lock);
 
         let elapsed = start_time.elapsed();
         if elapsed > Duration::from_secs(1) {
@@ -782,8 +813,8 @@ pub async fn send_transaction_message_route(
 type AsyncRecursiveResult<'a, T, E> = Pin<Box<dyn Future<Output = StdResult<T, E>> + 'a + Send>>;
 
 /// When receiving an event one needs to:
-/// 0. Skip the PDU if we already know about it
+/// 0. Check the server is in the room
-/// 1. Check the server is in the room
+/// 1. Skip the PDU if we already know about it
 /// 2. Check signatures, otherwise drop
 /// 3. Check content hash, redact if doesn't match
 /// 4. Fetch any missing auth events doing all checks listed here starting at 1. These are not
@@ -808,6 +839,7 @@ type AsyncRecursiveResult<'a, T, E> = Pin<Box<dyn Future<Output = StdResult<T, E
 pub fn handle_incoming_pdu<'a>(
     origin: &'a ServerName,
     event_id: &'a EventId,
+    room_id: &'a RoomId,
     value: BTreeMap<String, CanonicalJsonValue>,
     is_timeline_event: bool,
     db: &'a Database,
@@ -815,24 +847,6 @@ pub fn handle_incoming_pdu<'a>(
 ) -> AsyncRecursiveResult<'a, Option<Vec<u8>>, String> {
     Box::pin(async move {
         // TODO: For RoomVersion6 we must check that Raw<..> is canonical do we anywhere?: https://matrix.org/docs/spec/rooms/v6#canonical-json
-
-        // 0. Skip the PDU if we already have it as a timeline event
-        if let Ok(Some(pdu_id)) = db.rooms.get_pdu_id(&event_id) {
-            return Ok(Some(pdu_id.to_vec()));
-        }
-
-        // 1. Check the server is in the room
-        let room_id = match value
-            .get("room_id")
-            .and_then(|id| RoomId::try_from(id.as_str()?).ok())
-        {
-            Some(id) => id,
-            None => {
-                // Event is invalid
-                return Err("Event needs a valid RoomId.".to_string());
-            }
-        };
-
         match db.rooms.exists(&room_id) {
             Ok(true) => {}
             _ => {
@@ -840,6 +854,11 @@ pub fn handle_incoming_pdu<'a>(
             }
         }
 
+        // 1. Skip the PDU if we already have it as a timeline event
+        if let Ok(Some(pdu_id)) = db.rooms.get_pdu_id(&event_id) {
+            return Ok(Some(pdu_id.to_vec()));
+        }
+
         // We go through all the signatures we see on the value and fetch the corresponding signing
         // keys
         fetch_required_signing_keys(&value, &pub_key_map, db)
@@ -899,7 +918,7 @@ pub fn handle_incoming_pdu<'a>(
         // 5. Reject "due to auth events" if can't get all the auth events or some of the auth events are also rejected "due to auth events"
         // EDIT: Step 5 is not applied anymore because it failed too often
         debug!("Fetching auth events for {}", incoming_pdu.event_id);
-        fetch_and_handle_events(db, origin, &incoming_pdu.auth_events, pub_key_map)
+        fetch_and_handle_events(db, origin, &incoming_pdu.auth_events, &room_id, pub_key_map)
             .await
             .map_err(|e| e.to_string())?;
 
@@ -1000,13 +1019,13 @@ pub fn handle_incoming_pdu<'a>(
 
         if incoming_pdu.prev_events.len() == 1 {
             let prev_event = &incoming_pdu.prev_events[0];
-            let state_vec = db
+            let state = db
                 .rooms
                 .pdu_shortstatehash(prev_event)
                 .map_err(|_| "Failed talking to db".to_owned())?
                 .map(|shortstatehash| db.rooms.state_full_ids(shortstatehash).ok())
                 .flatten();
-            if let Some(mut state_vec) = state_vec {
+            if let Some(mut state) = state {
                 if db
                     .rooms
                     .get_pdu(prev_event)
@@ -1016,25 +1035,31 @@ pub fn handle_incoming_pdu<'a>(
                     .state_key
                     .is_some()
                 {
-                    state_vec.push(prev_event.clone());
+                    state.insert(prev_event.clone());
                 }
                 state_at_incoming_event = Some(
-                    fetch_and_handle_events(db, origin, &state_vec, pub_key_map)
+                    fetch_and_handle_events(
-                        .await
+                        db,
-                        .map_err(|_| "Failed to fetch state events locally".to_owned())?
+                        origin,
-                        .into_iter()
+                        &state.into_iter().collect::<Vec<_>>(),
-                        .map(|pdu| {
+                        &room_id,
+                        pub_key_map,
+                    )
+                    .await
+                    .map_err(|_| "Failed to fetch state events locally".to_owned())?
+                    .into_iter()
+                    .map(|pdu| {
+                        (
                             (
-                                (
+                                pdu.kind.clone(),
-                                    pdu.kind.clone(),
+                                pdu.state_key
-                                    pdu.state_key
+                                    .clone()
-                                        .clone()
+                                    .expect("events from state_full_ids are state events"),
-                                        .expect("events from state_full_ids are state events"),
+                            ),
-                                ),
+                            pdu,
-                                pdu,
+                        )
-                            )
+                    })
-                        })
+                    .collect(),
-                        .collect(),
                 );
             }
             // TODO: set incoming_auth_events?
@@ -1057,12 +1082,18 @@ pub fn handle_incoming_pdu<'a>(
             {
                 Ok(res) => {
                     debug!("Fetching state events at event.");
-                    let state_vec =
+                    let state_vec = match fetch_and_handle_events(
-                        match fetch_and_handle_events(&db, origin, &res.pdu_ids, pub_key_map).await
+                        &db,
-                        {
+                        origin,
-                            Ok(state) => state,
+                        &res.pdu_ids,
-                            Err(_) => return Err("Failed to fetch state events.".to_owned()),
+                        &room_id,
-                        };
+                        pub_key_map,
+                    )
+                    .await
+                    {
+                        Ok(state) => state,
+                        Err(_) => return Err("Failed to fetch state events.".to_owned()),
+                    };
 
                     let mut state = BTreeMap::new();
                     for pdu in state_vec {
@@ -1088,8 +1119,14 @@ pub fn handle_incoming_pdu<'a>(
                     }
 
                     debug!("Fetching auth chain events at event.");
-                    match fetch_and_handle_events(&db, origin, &res.auth_chain_ids, pub_key_map)
+                    match fetch_and_handle_events(
-                        .await
+                        &db,
+                        origin,
+                        &res.auth_chain_ids,
+                        &room_id,
+                        pub_key_map,
+                    )
+                    .await
                     {
                         Ok(state) => state,
                         Err(_) => return Err("Failed to fetch auth chain.".to_owned()),
@@ -1219,8 +1256,14 @@ pub fn handle_incoming_pdu<'a>(
             for map in &fork_states {
                 let mut state_auth = vec![];
                 for auth_id in map.values().flat_map(|pdu| &pdu.auth_events) {
-                    match fetch_and_handle_events(&db, origin, &[auth_id.clone()], pub_key_map)
+                    match fetch_and_handle_events(
-                        .await
+                        &db,
+                        origin,
+                        &[auth_id.clone()],
+                        &room_id,
+                        pub_key_map,
+                    )
+                    .await
                     {
                         // This should always contain exactly one element when Ok
                         Ok(events) => state_auth.extend_from_slice(&events),
@@ -1326,6 +1369,7 @@ pub(crate) fn fetch_and_handle_events<'a>(
     db: &'a Database,
     origin: &'a ServerName,
     events: &'a [EventId],
+    room_id: &'a RoomId,
     pub_key_map: &'a RwLock<BTreeMap<String, BTreeMap<String, String>>>,
 ) -> AsyncRecursiveResult<'a, Vec<Arc<PduEvent>>, Error> {
     Box::pin(async move {
@@ -1379,6 +1423,7 @@ pub(crate) fn fetch_and_handle_events<'a>(
                             match handle_incoming_pdu(
                                 origin,
                                 &event_id,
+                                &room_id,
                                 value.clone(),
                                 false,
                                 db,
@@ -1592,21 +1637,14 @@ pub(crate) fn append_incoming_pdu(
     new_room_leaves: HashSet<EventId>,
     state: &StateMap<Arc<PduEvent>>,
 ) -> Result<Vec<u8>> {
-    let count = db.globals.next_count()?;
-    let mut pdu_id = pdu.room_id.as_bytes().to_vec();
-    pdu_id.push(0xff);
-    pdu_id.extend_from_slice(&count.to_be_bytes());
-
     // We append to state before appending the pdu, so we don't have a moment in time with the
     // pdu without it's state. This is okay because append_pdu can't fail.
     db.rooms
         .set_event_state(&pdu.event_id, state, &db.globals)?;
 
-    db.rooms.append_pdu(
+    let pdu_id = db.rooms.append_pdu(
         pdu,
         pdu_json,
-        count,
-        &pdu_id,
         &new_room_leaves.into_iter().collect::<Vec<_>>(),
         &db,
     )?;
@@ -1664,7 +1702,8 @@ pub(crate) fn append_incoming_pdu(
                     .filter_map(|r| r.ok())
                     .any(|member| users.iter().any(|regex| regex.is_match(member.as_str())))
             {
-                db.sending.send_pdu_appservice(&appservice.0, &pdu_id)?;
+                db.sending
+                    .send_pdu_appservice(&appservice.0, &pdu_id)?;
             }
         }
     }
@@ -1874,7 +1913,11 @@ pub fn get_room_state_ids_route(
             "Pdu state not found.",
         ))?;
 
-    let pdu_ids = db.rooms.state_full_ids(shortstatehash)?;
+    let pdu_ids = db
+        .rooms
+        .state_full_ids(shortstatehash)?
+        .into_iter()
+        .collect();
 
     let mut auth_chain_ids = BTreeSet::<EventId>::new();
     let mut todo = BTreeSet::new();
@@ -2120,18 +2163,36 @@ pub async fn create_join_event_route(
     )
     .map_err(|_| Error::BadRequest(ErrorKind::InvalidParam, "Origin field is invalid."))?;
 
-    let pdu_id = handle_incoming_pdu(&origin, &event_id, value, true, &db, &pub_key_map)
+    let mutex = Arc::clone(
-        .await
+        db.globals
-        .map_err(|_| {
+            .roomid_mutex
-            Error::BadRequest(
+            .write()
-                ErrorKind::InvalidParam,
+            .unwrap()
-                "Error while handling incoming PDU.",
+            .entry(body.room_id.clone())
-            )
+            .or_default(),
-        })?
+    );
-        .ok_or(Error::BadRequest(
+    let mutex_lock = mutex.lock().await;
+    let pdu_id = handle_incoming_pdu(
+        &origin,
+        &event_id,
+        &body.room_id,
+        value,
+        true,
+        &db,
+        &pub_key_map,
+    )
+    .await
+    .map_err(|_| {
+        Error::BadRequest(
             ErrorKind::InvalidParam,
-            "Could not accept incoming PDU as timeline event.",
+            "Error while handling incoming PDU.",
-        ))?;
+        )
+    })?
+    .ok_or(Error::BadRequest(
+        ErrorKind::InvalidParam,
+        "Could not accept incoming PDU as timeline event.",
+    ))?;
+    drop(mutex_lock);
 
     let state_ids = db.rooms.state_full_ids(shortstatehash)?;