|
| 1 | +//! `RMW_EVENT_MESSAGE_LOST` must actually be raised by a live subscriber. |
| 2 | +//! |
| 3 | +//! `event.rs`'s unit tests pin `MessageLossTracker`'s arithmetic. They say |
| 4 | +//! nothing about whether anything *calls* it: delete the `observe_loss(..)` line |
| 5 | +//! from the subscriber's receive path and every one of them still passes. This |
| 6 | +//! file is the detector for that wiring. |
| 7 | +//! |
| 8 | +//! Loss is induced deterministically rather than by trying to drop a packet. |
| 9 | +//! The subscriber's own key expression is published to directly, through the |
| 10 | +//! node's zenoh session, with a hand-built [`Attachment`] carrying a chosen |
| 11 | +//! sequence number — so the gap is exact and there is no timing to lose. |
| 12 | +
|
| 13 | +mod common; |
| 14 | + |
| 15 | +use std::{ |
| 16 | + sync::{Arc, Mutex}, |
| 17 | + thread, |
| 18 | + time::{Duration, Instant}, |
| 19 | +}; |
| 20 | + |
| 21 | +use common::{TestRouter, create_hiroz_context_with_endpoint}; |
| 22 | +use hiroz::{ |
| 23 | + Builder, GidArray, TypeHash, |
| 24 | + attachment::Attachment, |
| 25 | + event::ZenohEventType, |
| 26 | + ros_msg::MessageTypeInfo, |
| 27 | +}; |
| 28 | +use serde::{Deserialize, Serialize}; |
| 29 | +use serial_test::serial; |
| 30 | +use zenoh::Wait; |
| 31 | + |
| 32 | +#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)] |
| 33 | +struct Tick { |
| 34 | + counter: u64, |
| 35 | +} |
| 36 | + |
| 37 | +impl MessageTypeInfo for Tick { |
| 38 | + fn type_name() -> &'static str { |
| 39 | + "test_msgs::msg::dds_::Tick_" |
| 40 | + } |
| 41 | + fn type_hash() -> TypeHash { |
| 42 | + TypeHash::zero() |
| 43 | + } |
| 44 | +} |
| 45 | + |
| 46 | +impl hiroz::ros_msg::WithTypeInfo for Tick {} |
| 47 | + |
| 48 | +impl hiroz::msg::ZMessage for Tick { |
| 49 | + type Serdes = hiroz::msg::SerdeCdrSerdes<Tick>; |
| 50 | +} |
| 51 | + |
| 52 | +fn gid(n: u8) -> GidArray { |
| 53 | + let mut g = [0u8; 16]; |
| 54 | + g[0] = n; |
| 55 | + g |
| 56 | +} |
| 57 | + |
| 58 | +/// Wait until `total_count` for `MessageLost` stops changing, then return it. |
| 59 | +fn settled_loss_count(sub_events: &Arc<Mutex<hiroz::event::EventsManager>>) -> i32 { |
| 60 | + let mut last = -1; |
| 61 | + let mut stable_since = Instant::now(); |
| 62 | + let deadline = Instant::now() + Duration::from_secs(5); |
| 63 | + loop { |
| 64 | + let now = sub_events |
| 65 | + .lock() |
| 66 | + .unwrap() |
| 67 | + .take_event_status(ZenohEventType::MessageLost) |
| 68 | + .total_count; |
| 69 | + if now != last { |
| 70 | + last = now; |
| 71 | + stable_since = Instant::now(); |
| 72 | + } else if stable_since.elapsed() >= Duration::from_millis(400) { |
| 73 | + return now; |
| 74 | + } |
| 75 | + assert!(Instant::now() < deadline, "loss count never settled"); |
| 76 | + thread::sleep(Duration::from_millis(25)); |
| 77 | + } |
| 78 | +} |
| 79 | + |
| 80 | +/// A gap in a publisher's sequence numbers raises `MessageLost` on the |
| 81 | +/// subscriber that saw it, with the count of samples that never arrived. |
| 82 | +#[test] |
| 83 | +#[serial] |
| 84 | +fn a_sequence_gap_raises_message_lost() { |
| 85 | + const TOPIC: &str = "/message_lost_gap"; |
| 86 | + |
| 87 | + let router = TestRouter::new(); |
| 88 | + let ctx = create_hiroz_context_with_endpoint(router.endpoint()).expect("context"); |
| 89 | + let node = ctx.create_node("message_lost_node").build().expect("node"); |
| 90 | + |
| 91 | + let received = Arc::new(Mutex::new(Vec::<u64>::new())); |
| 92 | + let cb_received = received.clone(); |
| 93 | + let sub = node |
| 94 | + .create_sub::<Tick>(TOPIC) |
| 95 | + .build_with_callback(move |msg: Tick| { |
| 96 | + cb_received.lock().unwrap().push(msg.counter); |
| 97 | + }) |
| 98 | + .expect("subscriber"); |
| 99 | + |
| 100 | + // Publish straight onto the subscriber's own key expression, so the |
| 101 | + // sequence numbers are ours to choose. |
| 102 | + let ke = node |
| 103 | + .keyexpr_format() |
| 104 | + .topic_key_expr(sub.entity()) |
| 105 | + .expect("topic key expr"); |
| 106 | + let session = node.session(); |
| 107 | + let publisher_gid = gid(42); |
| 108 | + |
| 109 | + let put = |sn: i64, counter: u64| { |
| 110 | + let zbuf = <hiroz::msg::SerdeCdrSerdes<Tick> as hiroz::msg::ZSerializer>::serialize_to_zbuf( |
| 111 | + &Tick { counter }, |
| 112 | + ); |
| 113 | + session |
| 114 | + .put((*ke).clone(), zenoh::bytes::ZBytes::from(zbuf)) |
| 115 | + .attachment(Attachment::new(sn, publisher_gid)) |
| 116 | + .wait() |
| 117 | + .expect("put"); |
| 118 | + }; |
| 119 | + |
| 120 | + thread::sleep(Duration::from_millis(300)); |
| 121 | + |
| 122 | + put(0, 0); // baseline — first from this publisher, never counted |
| 123 | + put(1, 1); // contiguous |
| 124 | + put(5, 5); // 2, 3 and 4 never arrived |
| 125 | + |
| 126 | + let lost = settled_loss_count(sub.events_mgr()); |
| 127 | + |
| 128 | + assert_eq!( |
| 129 | + lost, 3, |
| 130 | + "expected the three skipped sequence numbers to be reported as lost; \ |
| 131 | + got {lost}. Zero means the receive path never fed the loss tracker" |
| 132 | + ); |
| 133 | + assert_eq!( |
| 134 | + received.lock().unwrap().len(), |
| 135 | + 3, |
| 136 | + "all three published samples should still have been delivered — \ |
| 137 | + detecting loss must not drop anything" |
| 138 | + ); |
| 139 | +} |
| 140 | + |
| 141 | +/// A subscriber that joins late has not "lost" the history it was never sent. |
| 142 | +/// |
| 143 | +/// Without the first-sample exemption this reports the publisher's sequence |
| 144 | +/// number as the loss count, so every late joiner looks catastrophically lossy. |
| 145 | +#[test] |
| 146 | +#[serial] |
| 147 | +fn joining_late_reports_no_loss() { |
| 148 | + const TOPIC: &str = "/message_lost_late_join"; |
| 149 | + |
| 150 | + let router = TestRouter::new(); |
| 151 | + let ctx = create_hiroz_context_with_endpoint(router.endpoint()).expect("context"); |
| 152 | + let node = ctx |
| 153 | + .create_node("message_lost_late_node") |
| 154 | + .build() |
| 155 | + .expect("node"); |
| 156 | + |
| 157 | + let sub = node |
| 158 | + .create_sub::<Tick>(TOPIC) |
| 159 | + .build_with_callback(|_msg: Tick| {}) |
| 160 | + .expect("subscriber"); |
| 161 | + |
| 162 | + let ke = node |
| 163 | + .keyexpr_format() |
| 164 | + .topic_key_expr(sub.entity()) |
| 165 | + .expect("topic key expr"); |
| 166 | + let session = node.session(); |
| 167 | + |
| 168 | + thread::sleep(Duration::from_millis(300)); |
| 169 | + |
| 170 | + // First sample this subscriber ever sees from this publisher, and it is |
| 171 | + // already well into the publisher's stream. |
| 172 | + let zbuf = <hiroz::msg::SerdeCdrSerdes<Tick> as hiroz::msg::ZSerializer>::serialize_to_zbuf( |
| 173 | + &Tick { counter: 9000 }, |
| 174 | + ); |
| 175 | + session |
| 176 | + .put((*ke).clone(), zenoh::bytes::ZBytes::from(zbuf)) |
| 177 | + .attachment(Attachment::new(9000, gid(7))) |
| 178 | + .wait() |
| 179 | + .expect("put"); |
| 180 | + |
| 181 | + let lost = settled_loss_count(sub.events_mgr()); |
| 182 | + |
| 183 | + assert_eq!( |
| 184 | + lost, 0, |
| 185 | + "a late joiner must not be charged for history it was never sent" |
| 186 | + ); |
| 187 | +} |
0 commit comments