maintain list of events in db. must convert allh handlers to use it.

src/bot/event_handlers.rs

@@ -41,13 +41,7 @@ fn check_message_age(
     }
 }
 
-/// Determine whether or not to process a received message. This check
-/// is necessary in addition to the event processing check because we
-/// may receive message events when entering a room for the first
-/// time, and we don't want to respond to things before the bot was in
-/// the channel, but we do want to respond to things that were sent if
-/// the bot left and rejoined quickly.
-async fn should_process_message<'a>(
+async fn should_process<'a>(
     bot: &DiceBot,
     event: &SyncMessageEvent<MessageEventContent>,
 ) -> Result<(String, String), BotError> {
@@ -147,12 +141,12 @@ impl EventEmitter for DiceBot {
     async fn on_stripped_state_member(
         &self,
         room: SyncRoom,
-        event: &StrippedStateEvent<MemberEventContent>,
+        room_member: &StrippedStateEvent<MemberEventContent>,
         _: Option<MemberEventContent>,
     ) {
         if let SyncRoom::Invited(room) = room {
             if let Some(user_id) = self.client.user_id().await {
-                if event.state_key != user_id {
+                if room_member.state_key != user_id {
                     return;
                 }
             }
@@ -169,22 +163,17 @@ impl EventEmitter for DiceBot {
 
     async fn on_room_message(&self, room: SyncRoom, event: &SyncMessageEvent<MessageEventContent>) {
         if let SyncRoom::Joined(room) = room {
-            //Clone to avoid holding lock.
-            let room = room.read().await.clone();
-            let room_id = room.room_id.as_str();
-            if !should_process_event(&self.db, room_id, event.event_id.as_str()) {
-                return;
-            }
+            let (msg_body, sender_username) =
+                if let Ok((msg_body, sender_username)) = should_process(self, &event).await {
+                    (msg_body, sender_username)
+                } else {
+                    return;
+                };
 
-            let (msg_body, sender_username) = if let Ok((msg_body, sender_username)) =
-                should_process_message(self, &event).await
-            {
-                (msg_body, sender_username)
-            } else {
-                return;
-            };
+            //we clone here to hold the lock for as little time as possible.
+            let real_room = room.read().await.clone();
 
-            self.execute_commands(&room, &sender_username, &msg_body)
+            self.execute_commands(&real_room, &sender_username, &msg_body)
                 .await;
         }
     }

src/db.rs

@@ -25,18 +25,12 @@ impl Database {
     fn new_db(db: sled::Db) -> Result<Database, DataError> {
         let migrations = db.open_tree("migrations")?;
 
-        let database = Database {
+        Ok(Database {
             db: db.clone(),
             variables: Variables::new(&db)?,
             migrations: Migrations(migrations),
             rooms: Rooms::new(&db)?,
-        };
-
-        //Start any event handlers.
-        database.rooms.start_handler();
-
-        info!("Opened database.");
-        Ok(database)
+        })
     }
 
     pub fn new<P: AsRef<Path>>(path: P) -> Result<Database, DataError> {

src/db/errors.rs

@@ -26,9 +26,6 @@ pub enum DataError {
     #[error("expected i32, but i32 schema was violated")]
     I32SchemaViolation,
 
-    #[error("unexpected or corruptd data bytes")]
-    InvalidValue,
-
     #[error("expected string, but utf8 schema was violated: {0}")]
     Utf8chemaViolation(#[from] std::str::Utf8Error),
 

src/db/rooms.rs

@@ -1,14 +1,11 @@
 use crate::db::errors::DataError;
-use crate::db::schema::convert_u64;
-use byteorder::BigEndian;
-use log::{error, log_enabled, trace};
+use byteorder::LittleEndian;
 use sled::transaction::TransactionalTree;
 use sled::Transactional;
 use sled::{CompareAndSwapError, Tree};
 use std::collections::HashSet;
 use std::str;
 use std::time::{SystemTime, UNIX_EPOCH};
-use tokio::task::JoinHandle;
 use zerocopy::byteorder::U64;
 use zerocopy::AsBytes;
 
@@ -23,22 +20,10 @@ pub struct Rooms {
     /// Username -> list of room IDs user is in.
     pub(in crate::db) username_roomids: Tree,
 
-    /// Room ID(str) 0xff event ID(str) -> timestamp. Records event
-    /// IDs that we have received, so we do not process twice.
+    /// Room ID 0xff event ID -> received timestamp.
     pub(in crate::db) roomeventid_timestamp: Tree,
-
-    /// Room ID(str) 0xff timestamp(u64) -> event ID. Records event
-    /// IDs with timestamp as the primary key instead. Exists to allow
-    /// easy scanning of old roomeventid_timestamp records for
-    /// removal. Be careful with u64, it can have 0xff and 0xfe bytes.
-    /// A simple split on 0xff will not work with this key. Instead,
-    /// it is meant to be split on the first 0xff byte only, and
-    /// separated into room ID and timestamp.
-    pub(in crate::db) roomtimestamp_eventid: Tree,
 }
 
-/// An enum that can hold either a regular sled Tree, or a
-/// Transactional tree.
 #[derive(Clone, Copy)]
 enum TxableTree<'a> {
     Tree(&'a Tree),
@@ -57,107 +42,39 @@ impl<'a> Into<TxableTree<'a>> for &'a TransactionalTree {
     }
 }
 
-/// A set of functions that can be used with a sled::Tree that stores
-/// HashSets as its values. Atomicity is partially handled. If the
-/// Tree is a transactional tree, operations will be atomic.
-/// Otherwise, there is a potential non-atomic step.
-mod hashset_tree {
-    use super::*;
-
-    fn insert_set<'a, T: Into<TxableTree<'a>>>(
-        tree: T,
-        key: &[u8],
-        set: HashSet<String>,
-    ) -> Result<(), DataError> {
-        let serialized = bincode::serialize(&set)?;
-        match tree.into() {
-            TxableTree::Tree(tree) => tree.insert(key, serialized)?,
-            TxableTree::Tx(tx) => tx.insert(key, serialized)?,
-        };
-        Ok(())
+fn get_set<'a, T: Into<TxableTree<'a>>>(tree: T, key: &[u8]) -> Result<HashSet<String>, DataError> {
+    let set: HashSet<String> = match tree.into() {
+        TxableTree::Tree(tree) => tree.get(key)?,
+        TxableTree::Tx(tx) => tx.get(key)?,
     }
+    .map(|bytes| bincode::deserialize::<HashSet<String>>(&bytes))
+    .unwrap_or(Ok(HashSet::new()))?;
 
-    pub(super) fn get_set<'a, T: Into<TxableTree<'a>>>(
-        tree: T,
-        key: &[u8],
-    ) -> Result<HashSet<String>, DataError> {
-        let set: HashSet<String> = match tree.into() {
-            TxableTree::Tree(tree) => tree.get(key)?,
-            TxableTree::Tx(tx) => tx.get(key)?,
-        }
-        .map(|bytes| bincode::deserialize::<HashSet<String>>(&bytes))
-        .unwrap_or(Ok(HashSet::new()))?;
-
-        Ok(set)
-    }
-
-    pub(super) fn remove_from_set<'a, T: Into<TxableTree<'a>> + Copy>(
-        tree: T,
-        key: &[u8],
-        value_to_remove: &str,
-    ) -> Result<(), DataError> {
-        let mut set = get_set(tree, key)?;
-        set.remove(value_to_remove);
-        insert_set(tree, key, set)?;
-        Ok(())
-    }
-
-    pub(super) fn add_to_set<'a, T: Into<TxableTree<'a>> + Copy>(
-        tree: T,
-        key: &[u8],
-        value_to_add: String,
-    ) -> Result<(), DataError> {
-        let mut set = get_set(tree, key)?;
-        set.insert(value_to_add);
-        insert_set(tree, key, set)?;
-        Ok(())
-    }
+    Ok(set)
 }
 
-/// Functions that specifically relate to the "timestamp index" tree,
-/// which is stored on the Room sinstance as a tree called
-/// roomtimestamp_eventid. Tightly coupled to the event watcher in the
-/// Rooms impl, and only factored out for unit testing.
-mod timestamp_index {
-    use super::*;
+fn remove_from_set<'a, T: Into<TxableTree<'a>> + Copy>(
+    tree: T,
+    key: &[u8],
+    value_to_remove: &str,
+) -> Result<(), DataError> {
+    let mut set = get_set(tree, key)?;
+    set.remove(value_to_remove);
+    insert_set(tree, key, set)?;
+    Ok(())
+}
 
-    /// Insert an entry from the main roomeventid_timestamp Tree into
-    /// the timestamp index. Keys in this Tree are stored as room ID
-    /// 0xff timestamp, with the value being a hashset of event IDs
-    /// received at the time. The parameters come from an insert to
-    /// that Tree, where the key is room ID 0xff event ID, and the
-    /// value is the timestamp.
-    pub(super) fn insert(
-        roomtimestamp_eventid: &Tree,
-        key: &[u8],
-        timestamp_bytes: &[u8],
-    ) -> Result<(), DataError> {
-        let parts: Vec<&[u8]> = key.split(|&b| b == 0xff).collect();
-        if let [room_id, event_id] = parts[..] {
-            let mut ts_key = room_id.to_vec();
-            ts_key.push(0xff);
-            ts_key.extend_from_slice(&timestamp_bytes);
-            trace_index_record(room_id, event_id, &timestamp_bytes);
-
-            let event_id = str::from_utf8(event_id)?;
-            hashset_tree::add_to_set(roomtimestamp_eventid, &ts_key, event_id.to_owned())?;
-            Ok(())
-        } else {
-            Err(DataError::InvalidValue)
-        }
-    }
-
-    /// Log a trace message.
-    fn trace_index_record(room_id: &[u8], event_id: &[u8], timestamp: &[u8]) {
-        if log_enabled!(log::Level::Trace) {
-            trace!(
-                "Recording event {} | {} received at {} in timestamp index.",
-                str::from_utf8(room_id).unwrap_or("[invalid room id]"),
-                str::from_utf8(event_id).unwrap_or("[invalid event id]"),
-                convert_u64(timestamp).unwrap_or(0)
-            );
-        }
-    }
+fn insert_set<'a, T: Into<TxableTree<'a>>>(
+    tree: T,
+    key: &[u8],
+    set: HashSet<String>,
+) -> Result<(), DataError> {
+    let serialized = bincode::serialize(&set)?;
+    match tree.into() {
+        TxableTree::Tree(tree) => tree.insert(key, serialized)?,
+        TxableTree::Tx(tx) => tx.insert(key, serialized)?,
+    };
+    Ok(())
 }
 
 impl Rooms {
@@ -167,33 +84,6 @@ impl Rooms {
             roomid_usernames: db.open_tree("roomid_usernames")?,
             username_roomids: db.open_tree("username_roomids")?,
             roomeventid_timestamp: db.open_tree("roomeventid_timestamp")?,
-            roomtimestamp_eventid: db.open_tree("roomtimestamp_eventid")?,
-        })
-    }
-
-    /// Start an event subscriber that listens for inserts made by the
-    /// `should_process` function. This event handler duplicates the
-    /// entry by timestamp instead of event ID.
-    pub(in crate::db) fn start_handler(&self) -> JoinHandle<()> {
-        //Clone due to lifetime requirements.
-        let roomeventid_timestamp = self.roomeventid_timestamp.clone();
-        let roomtimestamp_eventid = self.roomtimestamp_eventid.clone();
-
-        tokio::spawn(async move {
-            let mut subscriber = roomeventid_timestamp.watch_prefix(b"");
-
-            // TODO make this handler receive kill messages somehow so
-            // we can unit test it and gracefully shut it down.
-            while let Some(event) = (&mut subscriber).await {
-                if let sled::Event::Insert { key, value } = event {
-                    match timestamp_index::insert(&roomtimestamp_eventid, &key, &value) {
-                        Err(e) => {
-                            error!("Unable to update the timestamp index: {}", e);
-                        }
-                        _ => (),
-                    }
-                }
-            }
         })
     }
 
@@ -208,7 +98,7 @@ impl Rooms {
         key.push(0xff);
         key.extend_from_slice(event_id.as_bytes());
 
-        let timestamp: U64<BigEndian> = U64::new(
+        let timestamp: U64<LittleEndian> = U64::new(
             SystemTime::now()
                 .duration_since(UNIX_EPOCH)
                 .expect("Clock has gone backwards")
@@ -226,21 +116,25 @@ impl Rooms {
     }
 
     pub fn get_rooms_for_user(&self, username: &str) -> Result<HashSet<String>, DataError> {
-        hashset_tree::get_set(&self.username_roomids, username.as_bytes())
+        get_set(&self.username_roomids, username.as_bytes())
     }
 
     pub fn get_users_in_room(&self, room_id: &str) -> Result<HashSet<String>, DataError> {
-        hashset_tree::get_set(&self.roomid_usernames, room_id.as_bytes())
+        get_set(&self.roomid_usernames, room_id.as_bytes())
     }
 
     pub fn add_user_to_room(&self, username: &str, room_id: &str) -> Result<(), DataError> {
         (&self.username_roomids, &self.roomid_usernames).transaction(
             |(tx_username_rooms, tx_room_usernames)| {
                 let username_key = &username.as_bytes();
-                hashset_tree::add_to_set(tx_username_rooms, username_key, room_id.to_owned())?;
+                let mut user_to_rooms = get_set(tx_username_rooms, username_key)?;
+                user_to_rooms.insert(room_id.to_string());
+                insert_set(tx_username_rooms, username_key, user_to_rooms)?;
 
                 let roomid_key = &room_id.as_bytes();
-                hashset_tree::add_to_set(tx_room_usernames, roomid_key, username.to_owned())?;
+                let mut room_to_users = get_set(tx_room_usernames, roomid_key)?;
+                room_to_users.insert(username.to_string());
+                insert_set(tx_room_usernames, roomid_key, room_to_users)?;
 
                 Ok(())
             },
@@ -253,10 +147,14 @@ impl Rooms {
         (&self.username_roomids, &self.roomid_usernames).transaction(
             |(tx_username_rooms, tx_room_usernames)| {
                 let username_key = &username.as_bytes();
-                hashset_tree::remove_from_set(tx_username_rooms, username_key, room_id)?;
+                let mut user_to_rooms = get_set(tx_username_rooms, username_key)?;
+                user_to_rooms.remove(room_id);
+                insert_set(tx_username_rooms, username_key, user_to_rooms)?;
 
                 let roomid_key = &room_id.as_bytes();
-                hashset_tree::remove_from_set(tx_room_usernames, roomid_key, username)?;
+                let mut room_to_users = get_set(tx_room_usernames, roomid_key)?;
+                room_to_users.remove(username);
+                insert_set(tx_room_usernames, roomid_key, room_to_users)?;
 
                 Ok(())
             },
@@ -269,15 +167,11 @@ impl Rooms {
         (&self.username_roomids, &self.roomid_usernames).transaction(
             |(tx_username_roomids, tx_roomid_usernames)| {
                 let roomid_key = room_id.as_bytes();
-                let users_in_room = hashset_tree::get_set(tx_roomid_usernames, roomid_key)?;
+                let users_in_room = get_set(tx_roomid_usernames, roomid_key)?;
 
                 //Remove the room ID from every user's room ID list.
                 for username in users_in_room {
-                    hashset_tree::remove_from_set(
-                        tx_username_roomids,
-                        username.as_bytes(),
-                        room_id,
-                    )?;
+                    remove_from_set(tx_username_roomids, username.as_bytes(), room_id)?;
                 }
 
                 //Remove this room entry for the room ID -> username tree.
@@ -306,7 +200,6 @@ mod tests {
     #[test]
     fn add_user_to_room() {
         let rooms = create_test_instance();
-
         rooms
             .add_user_to_room("testuser", "myroom")
             .expect("Could not add user to room");
@@ -332,7 +225,6 @@ mod tests {
     #[test]
     fn remove_user_from_room() {
         let rooms = create_test_instance();
-
         rooms
             .add_user_to_room("testuser", "myroom")
             .expect("Could not add user to room");
@@ -356,7 +248,6 @@ mod tests {
     #[test]
     fn clear_info() {
         let rooms = create_test_instance();
-
         rooms
             .add_user_to_room("testuser", "myroom1")
             .expect("Could not add user to room1");
@@ -391,37 +282,4 @@ mod tests {
         assert_eq!(expected_users_in_room2, users_in_room2);
         assert_eq!(expected_rooms_for_user, rooms_for_user);
     }
-
-    #[test]
-    fn insert_to_timestamp_index() {
-        let rooms = create_test_instance();
-
-        // Insertion into timestamp index based on data that would go
-        // into main room x eventID -> timestamp tree.
-        let mut key = b"myroom".to_vec();
-        key.push(0xff);
-        key.extend_from_slice(b"myeventid");
-
-        let timestamp: U64<BigEndian> = U64::new(1000);
-
-        let result = timestamp_index::insert(
-            &rooms.roomtimestamp_eventid,
-            key.as_bytes(),
-            timestamp.as_bytes(),
-        );
-
-        assert!(result.is_ok());
-
-        // Retrieval of data from the timestamp index tree.
-        let mut ts_key = b"myroom".to_vec();
-        ts_key.push(0xff);
-        ts_key.extend_from_slice(timestamp.as_bytes());
-
-        let expected_events: HashSet<String> =
-            vec![String::from("myeventid")].into_iter().collect();
-
-        let event_ids = hashset_tree::get_set(&rooms.roomtimestamp_eventid, &ts_key)
-            .expect("Could not get set out of Tree");
-        assert_eq!(expected_events, event_ids);
-    }
 }

src/db/schema.rs

@@ -1,5 +1,5 @@
 use crate::db::errors::DataError;
-use byteorder::{BigEndian, LittleEndian};
+use byteorder::LittleEndian;
 use zerocopy::byteorder::{I32, U32, U64};
 use zerocopy::LayoutVerified;
 
@@ -13,8 +13,6 @@ type LittleEndianU32Layout<'a> = LayoutVerified<&'a [u8], U32<LittleEndian>>;
 #[allow(dead_code)]
 type LittleEndianU64Layout<'a> = LayoutVerified<&'a [u8], U64<LittleEndian>>;
 
-type BigEndianU64Layout<'a> = LayoutVerified<&'a [u8], U64<BigEndian>>;
-
 /// Convert bytes to an i32 with zero-copy deserialization. An error
 /// is returned if the bytes do not represent an i32.
 pub(super) fn convert_i32(raw_value: &[u8]) -> Result<i32, DataError> {
@@ -41,10 +39,10 @@ pub(super) fn convert_u32(raw_value: &[u8]) -> Result<u32, DataError> {
 
 #[allow(dead_code)]
 pub(super) fn convert_u64(raw_value: &[u8]) -> Result<u64, DataError> {
-    let layout = BigEndianU64Layout::new_unaligned(raw_value.as_ref());
+    let layout = LittleEndianU64Layout::new_unaligned(raw_value.as_ref());
 
     if let Some(layout) = layout {
-        let value: U64<BigEndian> = *layout;
+        let value: U64<LittleEndian> = *layout;
         Ok(value.get())
     } else {
         Err(DataError::I32SchemaViolation)