1use std::io::Write;
2use std::sync::atomic::AtomicBool;
3use std::sync::atomic::Ordering;
4use std::sync::Arc;
5use std::time::Duration;
6
7use anyhow::Result;
8use scx_stats::prelude::*;
9use scx_stats_derive::stat_doc;
10use scx_stats_derive::Stats;
11use serde::Deserialize;
12use serde::Serialize;
13
14#[stat_doc]
15#[derive(Clone, Debug, Default, Serialize, Deserialize, Stats)]
16#[stat(top)]
17pub struct Metrics {
18 #[stat(desc = "Number of running tasks")]
19 pub nr_running: u64,
20 #[stat(desc = "Number of online CPUs")]
21 pub nr_cpus: u64,
22 #[stat(desc = "Number of kthread direct dispatches")]
23 pub nr_kthread_dispatches: u64,
24 #[stat(desc = "Number of task direct dispatches")]
25 pub nr_direct_dispatches: u64,
26 #[stat(desc = "Number of regular task dispatches")]
27 pub nr_shared_dispatches: u64,
28 #[stat(desc = "Number of delay recovery dispatches")]
30 pub nr_delay_recovery_dispatches: u64,
31 #[stat(desc = "Number of delay middle add dispatches")]
32 pub nr_delay_middle_add_dispatches: u64,
33 #[stat(desc = "Number of delay fast recovery dispatches")]
34 pub nr_delay_fast_recovery_dispatches: u64,
35 #[stat(desc = "Number of delay rate-limited dispatches")]
36 pub nr_delay_rate_limited_dispatches: u64,
37 #[stat(desc = "Number of gain floor dispatches")]
38 pub nr_gain_floor_dispatches: u64,
39 #[stat(desc = "Number of gain ceiling dispatches")]
40 pub nr_gain_ceiling_dispatches: u64,
41 #[stat(desc = "Number of delay low region samples")]
42 pub nr_delay_low_region_samples: u64,
43 #[stat(desc = "Number of delay mid region samples")]
44 pub nr_delay_mid_region_samples: u64,
45 #[stat(desc = "Number of delay high region samples")]
46 pub nr_delay_high_region_samples: u64,
47 #[stat(desc = "Number of gain floor resident samples")]
48 pub nr_gain_floor_resident_samples: u64,
49 #[stat(desc = "Number of gain mid resident samples")]
50 pub nr_gain_mid_resident_samples: u64,
51 #[stat(desc = "Number of gain ceiling resident samples")]
52 pub nr_gain_ceiling_resident_samples: u64,
53 #[stat(desc = "Number of idle select path picks")]
54 pub nr_idle_select_path_picks: u64,
55 #[stat(desc = "Number of idle enqueue path picks")]
56 pub nr_idle_enqueue_path_picks: u64,
57 #[stat(desc = "Number of idle prev CPU picks")]
58 pub nr_idle_prev_cpu_picks: u64,
59 #[stat(desc = "Number of idle primary picks")]
60 pub nr_idle_primary_picks: u64,
61 #[stat(desc = "Number of idle spill picks")]
62 pub nr_idle_spill_picks: u64,
63 #[stat(desc = "Number of idle pick failures")]
64 pub nr_idle_pick_failures: u64,
65 #[stat(desc = "Number of idle primary domain misses")]
66 pub nr_idle_primary_domain_misses: u64,
67 #[stat(desc = "Number of idle global misses")]
68 pub nr_idle_global_misses: u64,
69 #[stat(desc = "Number of waker CPU biases")]
70 pub nr_waker_cpu_biases: u64,
71 #[stat(desc = "Number of keep running reuses")]
72 pub nr_keep_running_reuses: u64,
73 #[stat(desc = "Number of keep running queue empty")]
74 pub nr_keep_running_queue_empty: u64,
75 #[stat(desc = "Number of keep running SMT blocked")]
76 pub nr_keep_running_smt_blocked: u64,
77 #[stat(desc = "Number of keep running queued work")]
78 pub nr_keep_running_queued_work: u64,
79 #[stat(desc = "Number of dispatch CPU DSQ consumes")]
80 pub nr_dispatch_cpu_dsq_consumes: u64,
81 #[stat(desc = "Number of dispatch node DSQ consumes")]
82 pub nr_dispatch_node_dsq_consumes: u64,
83}
84
85impl Metrics {
86 fn format<W: Write>(&self, w: &mut W) -> Result<()> {
87 writeln!(
88 w,
89 "[{}] tasks -> r: {:>2}/{:<2} | dispatch -> k: {:<5} d: {:<5} s: {:<5} | timely -> rec: {:<5} mid: {:<5} rl: {:<5} min: {:<5} max: {:<5}",
90 crate::SCHEDULER_NAME,
91 self.nr_running,
92 self.nr_cpus,
93 self.nr_kthread_dispatches,
94 self.nr_direct_dispatches,
95 self.nr_shared_dispatches,
96 self.nr_delay_recovery_dispatches,
97 self.nr_delay_middle_add_dispatches,
98 self.nr_delay_rate_limited_dispatches,
99 self.nr_gain_floor_dispatches,
100 self.nr_gain_ceiling_dispatches
101 )?;
102 Ok(())
103 }
104
105 fn delta(&self, rhs: &Self) -> Self {
106 Self {
107 nr_kthread_dispatches: self.nr_kthread_dispatches - rhs.nr_kthread_dispatches,
108 nr_direct_dispatches: self.nr_direct_dispatches - rhs.nr_direct_dispatches,
109 nr_shared_dispatches: self.nr_shared_dispatches - rhs.nr_shared_dispatches,
110 nr_delay_recovery_dispatches: self.nr_delay_recovery_dispatches
111 - rhs.nr_delay_recovery_dispatches,
112 nr_delay_middle_add_dispatches: self.nr_delay_middle_add_dispatches
113 - rhs.nr_delay_middle_add_dispatches,
114 nr_delay_fast_recovery_dispatches: self.nr_delay_fast_recovery_dispatches
115 - rhs.nr_delay_fast_recovery_dispatches,
116 nr_delay_rate_limited_dispatches: self.nr_delay_rate_limited_dispatches
117 - rhs.nr_delay_rate_limited_dispatches,
118 nr_gain_floor_dispatches: self.nr_gain_floor_dispatches - rhs.nr_gain_floor_dispatches,
119 nr_gain_ceiling_dispatches: self.nr_gain_ceiling_dispatches
120 - rhs.nr_gain_ceiling_dispatches,
121 nr_delay_low_region_samples: self.nr_delay_low_region_samples
122 - rhs.nr_delay_low_region_samples,
123 nr_delay_mid_region_samples: self.nr_delay_mid_region_samples
124 - rhs.nr_delay_mid_region_samples,
125 nr_delay_high_region_samples: self.nr_delay_high_region_samples
126 - rhs.nr_delay_high_region_samples,
127 nr_gain_floor_resident_samples: self.nr_gain_floor_resident_samples
128 - rhs.nr_gain_floor_resident_samples,
129 nr_gain_mid_resident_samples: self.nr_gain_mid_resident_samples
130 - rhs.nr_gain_mid_resident_samples,
131 nr_gain_ceiling_resident_samples: self.nr_gain_ceiling_resident_samples
132 - rhs.nr_gain_ceiling_resident_samples,
133 nr_idle_select_path_picks: self.nr_idle_select_path_picks
134 - rhs.nr_idle_select_path_picks,
135 nr_idle_enqueue_path_picks: self.nr_idle_enqueue_path_picks
136 - rhs.nr_idle_enqueue_path_picks,
137 nr_idle_prev_cpu_picks: self.nr_idle_prev_cpu_picks - rhs.nr_idle_prev_cpu_picks,
138 nr_idle_primary_picks: self.nr_idle_primary_picks - rhs.nr_idle_primary_picks,
139 nr_idle_spill_picks: self.nr_idle_spill_picks - rhs.nr_idle_spill_picks,
140 nr_idle_pick_failures: self.nr_idle_pick_failures - rhs.nr_idle_pick_failures,
141 nr_idle_primary_domain_misses: self.nr_idle_primary_domain_misses
142 - rhs.nr_idle_primary_domain_misses,
143 nr_idle_global_misses: self.nr_idle_global_misses - rhs.nr_idle_global_misses,
144 nr_waker_cpu_biases: self.nr_waker_cpu_biases - rhs.nr_waker_cpu_biases,
145 nr_keep_running_reuses: self.nr_keep_running_reuses - rhs.nr_keep_running_reuses,
146 nr_keep_running_queue_empty: self.nr_keep_running_queue_empty
147 - rhs.nr_keep_running_queue_empty,
148 nr_keep_running_smt_blocked: self.nr_keep_running_smt_blocked
149 - rhs.nr_keep_running_smt_blocked,
150 nr_keep_running_queued_work: self.nr_keep_running_queued_work
151 - rhs.nr_keep_running_queued_work,
152 nr_dispatch_cpu_dsq_consumes: self.nr_dispatch_cpu_dsq_consumes
153 - rhs.nr_dispatch_cpu_dsq_consumes,
154 nr_dispatch_node_dsq_consumes: self.nr_dispatch_node_dsq_consumes
155 - rhs.nr_dispatch_node_dsq_consumes,
156 ..self.clone()
157 }
158 }
159}
160
161pub fn server_data() -> StatsServerData<(), Metrics> {
162 let open: Box<dyn StatsOpener<(), Metrics>> = Box::new(move |(req_ch, res_ch)| {
163 req_ch.send(())?;
164 let mut prev = res_ch.recv()?;
165
166 let read: Box<dyn StatsReader<(), Metrics>> = Box::new(move |_args, (req_ch, res_ch)| {
167 req_ch.send(())?;
168 let cur = res_ch.recv()?;
169 let delta = cur.delta(&prev);
170 prev = cur;
171 delta.to_json()
172 });
173
174 Ok(read)
175 });
176
177 StatsServerData::new()
178 .add_meta(Metrics::meta())
179 .add_ops("top", StatsOps { open, close: None })
180}
181
182pub fn monitor(intv: Duration, shutdown: Arc<AtomicBool>) -> Result<()> {
183 scx_utils::monitor_stats::<Metrics>(
184 &[],
185 intv,
186 || shutdown.load(Ordering::Relaxed),
187 |metrics| metrics.format(&mut std::io::stdout()),
188 )
189}