1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
use crate::error::ObixFault;
use es_entity::errlanes::Laned;
use tokio::sync::mpsc;
use tracing::Instrument;
use std::time::Duration;
use crate::handle::{OwnedTaskHandle, spawn_supervised};
/// A message from the pg LISTEN pump to a cache loop.
pub(crate) enum NotifyMessage {
/// A notification arrived on the subscribed channel.
Notification(sqlx::postgres::PgNotification),
/// The LISTEN connection was lost and has been re-established.
/// Notifications sent while it was down are gone — the cache must
/// resync against the tables instead of trusting the notify stream.
Resync,
}
const INITIAL_RECONNECT_BACKOFF: Duration = Duration::from_millis(50);
const MAX_RECONNECT_BACKOFF: Duration = Duration::from_secs(5);
pub async fn spawn_pg_listener(
pool: &sqlx::PgPool,
channel: &'static str,
notification_tx: mpsc::Sender<NotifyMessage>,
) -> Result<OwnedTaskHandle, sqlx::Error> {
let pool = pool.clone();
// The initial connection stays fail-fast: a broken setup should surface
// at init, not as an endlessly-retrying background task.
let mut listener = connect_and_listen(&pool, channel).await?;
let pump = async move {
loop {
// `try_recv` (unlike `recv`) surfaces connection loss as
// `Ok(None)` after sqlx has already reconnected and re-issued
// LISTEN. `recv` would swallow that marker — and any notifications sent
// while the connection was down would be silently lost with no
// resync, stalling every consumer until the next notification.
match listener.try_recv().await {
Ok(Some(notification)) => {
// A LISTEN connection can still receive stray channels.
if notification.channel() != channel {
continue;
}
// If send fails, receiver is dropped, so break
if let Err(e) = notification_tx
.send(NotifyMessage::Notification(notification))
.await
{
record_forward_failed(&e);
break;
}
}
Ok(None) => {
// Connection was lost; sqlx already reconnected and
// re-issued LISTEN. Tell the cache to resync for
// whatever was notified during the gap.
record_connection_lost();
if let Err(e) = notification_tx.send(NotifyMessage::Resync).await {
record_forward_failed(&e);
break;
}
}
Err(error) => {
// Reconnecting inside `try_recv` failed (e.g. the pool
// was saturated at that moment). This pump must outlive
// transient failures: exiting here drops the notification
// sender, which kills the cache loop and leaves the
// outbox permanently deaf while every consumer still
// looks healthy. Rebuild the listener with capped
// backoff, then resync.
record_recv_error(&error);
let mut backoff = INITIAL_RECONNECT_BACKOFF;
listener = loop {
tokio::time::sleep(backoff).await;
match connect_and_listen(&pool, channel).await {
Ok(listener) => break listener,
Err(error) => {
record_reconnect_failed(error);
backoff = (backoff * 2).min(MAX_RECONNECT_BACKOFF);
}
}
};
record_reconnected();
// Resync only after LISTEN is active again: anything
// published before this point is covered by the resync
// (the head re-read happens after), anything after by
// regular notifications.
if let Err(e) = notification_tx.send(NotifyMessage::Resync).await {
record_forward_failed(&e);
break;
}
}
}
}
}
.instrument(tracing::info_span!("obix.pg_listener", channel));
Ok(OwnedTaskHandle::new(spawn_supervised(
"obix::pg_listener",
pump,
)))
}
async fn connect_and_listen(
pool: &sqlx::PgPool,
channel: &'static str,
) -> Result<sqlx::postgres::PgListener, sqlx::Error> {
let mut listener = sqlx::postgres::PgListener::connect_with(pool).await?;
listener.listen(channel).await?;
Ok(listener)
}
#[tracing::instrument(
name = "obix.pg_listener.recv_error",
level = "error",
skip_all,
fields(otel.status_code = "ERROR", error = %error),
)]
fn record_recv_error(error: &sqlx::Error) {}
#[tracing::instrument(
name = "obix.pg_listener.connection_lost",
level = "error",
fields(otel.status_code = "ERROR"),
)]
fn record_connection_lost() {}
#[tracing::instrument(
name = "obix.pg_listener.reconnect_failed",
level = "error",
skip_all,
fields(
otel.status_code = "ERROR",
error = tracing::field::Empty,
error.lane = tracing::field::Empty,
error.code = tracing::field::Empty,
error.level = tracing::field::Empty,
exception.message = tracing::field::Empty,
exception.type = tracing::field::Empty,
),
)]
fn record_reconnect_failed(fault: impl Into<ObixFault>) {
fault.into().record(&tracing::Span::current());
}
#[tracing::instrument(name = "obix.pg_listener.reconnected", level = "warn")]
fn record_reconnected() {}
#[tracing::instrument(
name = "obix.pg_listener.forward_failed",
level = "error",
skip_all,
fields(otel.status_code = "ERROR", error = %error),
)]
fn record_forward_failed<T>(error: &tokio::sync::mpsc::error::SendError<T>) {}