2019-07-08 13:31:42 +02:00
|
|
|
//! Receives data from Redis, sorts it by `ClientAgent`, and stores it until
|
|
|
|
//! polled by the correct `ClientAgent`. Also manages sububscriptions and
|
|
|
|
//! unsubscriptions to/from Redis.
|
2019-10-09 20:46:56 +02:00
|
|
|
mod message_queues;
|
|
|
|
use crate::{
|
|
|
|
config::{self, RedisInterval},
|
2020-03-19 01:37:10 +01:00
|
|
|
log_fatal,
|
2020-03-20 01:54:23 +01:00
|
|
|
parse_client_request::subscription::{self, postgres, PgPool, Timeline},
|
2019-10-09 20:46:56 +02:00
|
|
|
pubsub_cmd,
|
|
|
|
redis_to_client_stream::redis::{redis_cmd, RedisConn, RedisStream},
|
2019-10-09 02:35:26 +02:00
|
|
|
};
|
2019-04-30 15:44:51 +02:00
|
|
|
use futures::{Async, Poll};
|
2020-03-19 01:37:10 +01:00
|
|
|
use lru::LruCache;
|
2019-10-09 20:46:56 +02:00
|
|
|
pub use message_queues::{MessageQueues, MsgQueue};
|
2019-04-30 15:44:51 +02:00
|
|
|
use serde_json::Value;
|
2020-03-19 01:37:10 +01:00
|
|
|
use std::{collections::HashMap, net, time};
|
2019-09-28 05:29:11 +02:00
|
|
|
use tokio::io::Error;
|
2019-05-10 07:47:29 +02:00
|
|
|
use uuid::Uuid;
|
2019-05-09 05:02:01 +02:00
|
|
|
|
2019-07-08 21:21:02 +02:00
|
|
|
/// The item that streams from Redis and is polled by the `ClientAgent`
|
2019-04-30 15:44:51 +02:00
|
|
|
#[derive(Debug)]
|
|
|
|
pub struct Receiver {
|
2019-10-09 20:46:56 +02:00
|
|
|
pub pubsub_connection: RedisStream,
|
2019-07-06 02:08:50 +02:00
|
|
|
secondary_redis_connection: net::TcpStream,
|
2019-10-09 02:35:26 +02:00
|
|
|
redis_poll_interval: RedisInterval,
|
2019-07-08 13:31:42 +02:00
|
|
|
redis_polled_at: time::Instant,
|
2020-03-19 01:37:10 +01:00
|
|
|
timeline: Timeline,
|
2019-05-09 05:02:01 +02:00
|
|
|
manager_id: Uuid,
|
2019-10-09 20:46:56 +02:00
|
|
|
pub msg_queues: MessageQueues,
|
2020-03-19 01:37:10 +01:00
|
|
|
clients_per_timeline: HashMap<Timeline, i32>,
|
|
|
|
cache: Cache,
|
|
|
|
pool: PgPool,
|
|
|
|
}
|
|
|
|
#[derive(Debug)]
|
|
|
|
struct Cache {
|
|
|
|
id_to_hashtag: LruCache<i64, String>,
|
|
|
|
hashtag_to_id: LruCache<String, i64>,
|
|
|
|
}
|
|
|
|
impl Cache {
|
|
|
|
fn new(size: usize) -> Self {
|
|
|
|
Self {
|
|
|
|
id_to_hashtag: LruCache::new(size),
|
|
|
|
hashtag_to_id: LruCache::new(size),
|
|
|
|
}
|
|
|
|
}
|
2019-05-09 17:52:05 +02:00
|
|
|
}
|
2019-04-30 15:44:51 +02:00
|
|
|
impl Receiver {
|
2019-07-06 02:08:50 +02:00
|
|
|
/// Create a new `Receiver`, with its own Redis connections (but, as yet, no
|
|
|
|
/// active subscriptions).
|
2020-03-19 01:37:10 +01:00
|
|
|
pub fn new(redis_cfg: config::RedisConfig, pool: PgPool) -> Self {
|
2019-10-03 06:34:41 +02:00
|
|
|
let RedisConn {
|
|
|
|
primary: pubsub_connection,
|
|
|
|
secondary: secondary_redis_connection,
|
|
|
|
namespace: redis_namespace,
|
2019-10-04 00:02:23 +02:00
|
|
|
polling_interval: redis_poll_interval,
|
|
|
|
} = RedisConn::new(redis_cfg);
|
2019-10-03 06:34:41 +02:00
|
|
|
|
2019-04-30 15:44:51 +02:00
|
|
|
Self {
|
2019-10-09 20:46:56 +02:00
|
|
|
pubsub_connection: RedisStream::from_stream(pubsub_connection)
|
|
|
|
.with_namespace(redis_namespace),
|
2019-05-10 07:47:29 +02:00
|
|
|
secondary_redis_connection,
|
2019-10-04 00:02:23 +02:00
|
|
|
redis_poll_interval,
|
2019-07-08 13:31:42 +02:00
|
|
|
redis_polled_at: time::Instant::now(),
|
2020-03-19 01:37:10 +01:00
|
|
|
timeline: Timeline::empty(),
|
2019-07-06 02:08:50 +02:00
|
|
|
manager_id: Uuid::default(),
|
2020-03-19 01:37:10 +01:00
|
|
|
msg_queues: MessageQueues(HashMap::new()),
|
|
|
|
clients_per_timeline: HashMap::new(),
|
|
|
|
cache: Cache::new(1000), // should this be a run-time option?
|
|
|
|
pool,
|
2019-04-30 15:44:51 +02:00
|
|
|
}
|
|
|
|
}
|
2019-05-10 07:47:29 +02:00
|
|
|
|
2019-07-06 02:08:50 +02:00
|
|
|
/// Assigns the `Receiver` a new timeline to monitor and runs other
|
|
|
|
/// first-time setup.
|
|
|
|
///
|
2019-07-08 13:31:42 +02:00
|
|
|
/// Note: this method calls `subscribe_or_unsubscribe_as_needed`,
|
2019-07-06 02:08:50 +02:00
|
|
|
/// so Redis PubSub subscriptions are only updated when a new timeline
|
|
|
|
/// comes under management for the first time.
|
2020-03-19 01:37:10 +01:00
|
|
|
pub fn manage_new_timeline(&mut self, manager_id: Uuid, timeline: Timeline) {
|
2019-07-06 02:08:50 +02:00
|
|
|
self.manager_id = manager_id;
|
2020-03-19 01:37:10 +01:00
|
|
|
self.timeline = timeline;
|
2019-07-08 13:31:42 +02:00
|
|
|
self.msg_queues
|
2019-07-06 02:08:50 +02:00
|
|
|
.insert(self.manager_id, MsgQueue::new(timeline));
|
|
|
|
self.subscribe_or_unsubscribe_as_needed(timeline);
|
2019-04-30 15:44:51 +02:00
|
|
|
}
|
2019-05-10 07:47:29 +02:00
|
|
|
|
2019-07-08 21:21:02 +02:00
|
|
|
/// Set the `Receiver`'s manager_id and target_timeline fields to the appropriate
|
2019-07-06 02:08:50 +02:00
|
|
|
/// value to be polled by the current `StreamManager`.
|
2020-03-19 01:37:10 +01:00
|
|
|
pub fn configure_for_polling(&mut self, manager_id: Uuid, timeline: Timeline) {
|
2019-07-06 02:08:50 +02:00
|
|
|
self.manager_id = manager_id;
|
2020-03-19 01:37:10 +01:00
|
|
|
self.timeline = timeline;
|
|
|
|
}
|
|
|
|
|
|
|
|
fn if_hashtag_timeline_get_hashtag_name(&mut self, timeline: Timeline) -> Option<String> {
|
2020-03-20 01:54:23 +01:00
|
|
|
use subscription::Stream::*;
|
2020-03-19 01:37:10 +01:00
|
|
|
if let Timeline(Hashtag(id), _, _) = timeline {
|
|
|
|
let cached_tag = self.cache.id_to_hashtag.get(&id).map(String::from);
|
|
|
|
let tag = match cached_tag {
|
|
|
|
Some(tag) => tag,
|
|
|
|
None => {
|
|
|
|
let new_tag = postgres::select_hashtag_name(&id, self.pool.clone())
|
|
|
|
.unwrap_or_else(|_| log_fatal!("No hashtag associated with tag #{}", &id));
|
|
|
|
self.cache.hashtag_to_id.put(new_tag.clone(), id);
|
|
|
|
self.cache.id_to_hashtag.put(id, new_tag.clone());
|
|
|
|
new_tag.to_string()
|
|
|
|
}
|
|
|
|
};
|
|
|
|
Some(tag)
|
|
|
|
} else {
|
|
|
|
None
|
|
|
|
}
|
2019-05-10 07:47:29 +02:00
|
|
|
}
|
|
|
|
|
2019-07-06 02:08:50 +02:00
|
|
|
/// Drop any PubSub subscriptions that don't have active clients and check
|
|
|
|
/// that there's a subscription to the current one. If there isn't, then
|
|
|
|
/// subscribe to it.
|
2020-03-19 01:37:10 +01:00
|
|
|
fn subscribe_or_unsubscribe_as_needed(&mut self, timeline: Timeline) {
|
2019-09-28 23:57:37 +02:00
|
|
|
let start_time = std::time::Instant::now();
|
2020-03-19 01:37:10 +01:00
|
|
|
let timelines_to_modify = self.msg_queues.calculate_timelines_to_add_or_drop(timeline);
|
2019-05-09 05:02:01 +02:00
|
|
|
|
2019-05-10 07:47:29 +02:00
|
|
|
// Record the lower number of clients subscribed to that channel
|
2019-07-08 13:31:42 +02:00
|
|
|
for change in timelines_to_modify {
|
2020-03-19 01:37:10 +01:00
|
|
|
let timeline = change.timeline;
|
|
|
|
let opt_hashtag = self.if_hashtag_timeline_get_hashtag_name(timeline);
|
|
|
|
let opt_hashtag = opt_hashtag.as_ref();
|
|
|
|
|
2019-05-09 05:02:01 +02:00
|
|
|
let count_of_subscribed_clients = self
|
2019-05-10 07:47:29 +02:00
|
|
|
.clients_per_timeline
|
2020-03-19 01:37:10 +01:00
|
|
|
.entry(timeline)
|
2019-10-02 06:03:18 +02:00
|
|
|
.and_modify(|n| *n += change.in_subscriber_number)
|
|
|
|
.or_insert_with(|| 1);
|
2020-03-19 01:37:10 +01:00
|
|
|
|
2019-05-10 07:47:29 +02:00
|
|
|
// If no clients, unsubscribe from the channel
|
2019-05-09 05:02:01 +02:00
|
|
|
if *count_of_subscribed_clients <= 0 {
|
2020-03-19 01:37:10 +01:00
|
|
|
pubsub_cmd!("unsubscribe", self, timeline.to_redis_str(opt_hashtag));
|
2019-10-02 06:03:18 +02:00
|
|
|
} else if *count_of_subscribed_clients == 1 && change.in_subscriber_number == 1 {
|
2020-03-19 01:37:10 +01:00
|
|
|
pubsub_cmd!("subscribe", self, timeline.to_redis_str(opt_hashtag));
|
2019-07-08 13:31:42 +02:00
|
|
|
}
|
|
|
|
}
|
2019-09-28 23:57:37 +02:00
|
|
|
if start_time.elapsed().as_millis() > 1 {
|
|
|
|
log::warn!("Sending cmd to Redis took: {:?}", start_time.elapsed());
|
|
|
|
};
|
2019-07-06 02:08:50 +02:00
|
|
|
}
|
|
|
|
}
|
2019-07-08 13:31:42 +02:00
|
|
|
|
|
|
|
/// The stream that the ClientAgent polls to learn about new messages.
|
2019-07-06 02:08:50 +02:00
|
|
|
impl futures::stream::Stream for Receiver {
|
2019-04-30 15:44:51 +02:00
|
|
|
type Item = Value;
|
|
|
|
type Error = Error;
|
|
|
|
|
2019-07-08 13:31:42 +02:00
|
|
|
/// Returns the oldest message in the `ClientAgent`'s queue (if any).
|
|
|
|
///
|
|
|
|
/// Note: This method does **not** poll Redis every time, because polling
|
|
|
|
/// Redis is signifiantly more time consuming that simply returning the
|
|
|
|
/// message already in a queue. Thus, we only poll Redis if it has not
|
|
|
|
/// been polled lately.
|
2019-04-30 15:44:51 +02:00
|
|
|
fn poll(&mut self) -> Poll<Option<Value>, Self::Error> {
|
2019-10-09 20:46:56 +02:00
|
|
|
let (timeline, id) = (self.timeline.clone(), self.manager_id);
|
2019-10-09 02:35:26 +02:00
|
|
|
if self.redis_polled_at.elapsed() > *self.redis_poll_interval {
|
2020-03-19 01:37:10 +01:00
|
|
|
for (raw_timeline, msg_value) in self.pubsub_connection.poll_redis() {
|
|
|
|
let hashtag = if raw_timeline.starts_with("hashtag") {
|
|
|
|
let tag_name = raw_timeline
|
|
|
|
.split(':')
|
|
|
|
.nth(1)
|
|
|
|
.unwrap_or_else(|| log_fatal!("No hashtag found in `{}`", raw_timeline))
|
|
|
|
.to_string();
|
|
|
|
let tag_id = *self
|
|
|
|
.cache
|
|
|
|
.hashtag_to_id
|
|
|
|
.get(&tag_name)
|
|
|
|
.unwrap_or_else(|| log_fatal!("No cached id for `{}`", tag_name));
|
|
|
|
Some(tag_id)
|
|
|
|
} else {
|
|
|
|
None
|
|
|
|
};
|
|
|
|
let timeline = Timeline::from_redis_str(&raw_timeline, hashtag);
|
|
|
|
for msg_queue in self.msg_queues.values_mut() {
|
|
|
|
if msg_queue.timeline == timeline {
|
|
|
|
msg_queue.messages.push_back(msg_value.clone());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2019-07-08 13:31:42 +02:00
|
|
|
self.redis_polled_at = time::Instant::now();
|
|
|
|
}
|
2019-04-30 15:44:51 +02:00
|
|
|
|
2019-05-10 07:47:29 +02:00
|
|
|
// Record current time as last polled time
|
2019-10-09 20:46:56 +02:00
|
|
|
self.msg_queues.update_time_for_target_queue(id);
|
2019-04-30 15:44:51 +02:00
|
|
|
|
2019-05-10 07:47:29 +02:00
|
|
|
// If the `msg_queue` being polled has any new messages, return the first (oldest) one
|
2019-10-09 20:46:56 +02:00
|
|
|
match self.msg_queues.oldest_msg_in_target_queue(id, timeline) {
|
2019-09-28 23:57:37 +02:00
|
|
|
Some(value) => Ok(Async::Ready(Some(value))),
|
2019-05-10 07:47:29 +02:00
|
|
|
_ => Ok(Async::NotReady),
|
2019-04-30 15:44:51 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|