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
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
//! Publishes `api::Status` after each tick and command, and records the profiles, slots, budget and repositories
//! that changed as events. It reads engine state and must not change it, besides the status channel.
use std::collections::{BTreeMap, BTreeSet};
use super::{Engine, Shutdown, unix_secs};
use crate::api;
use crate::docker::DockerApi;
use crate::github::GithubApi;
use crate::scheduler::{Plan, SlotRef};
impl<G: GithubApi + 'static, D: DockerApi + 'static> Engine<G, D> {
pub(super) fn publish(&mut self) {
let mode = self.control.mode();
// Why jobs wait and which runners are warm, from a plan of the engine as it is now, not as the tick's plan
// saw it: a start that failed, a runner stopped over budget or a pause since then would make that one wrong.
// A shutdown starts nothing, so no job says why it waits and no runner is warm.
let plan = match mode {
Shutdown::No => self.plan().0,
Shutdown::Graceful | Shutdown::Force => Plan::default(),
};
let profiles = self
.config
.profiles
.iter()
.chain(&self.retired)
.map(|(name, profile)| api::ProfileStatus {
name: name.clone(),
image: profile.image.reference(),
cpus: profile.cpus,
memory: profile.memory.0,
reseeding: self.is_reseeding(name),
slots: self
.slots
.get(name)
.into_iter()
.flatten()
.enumerate()
// Slots a reload removed show while their runner runs.
.filter(|(index, slot)| {
slot.is_some()
|| self
.config
.profiles
.get(name)
.is_some_and(|profile| *index < profile.slots as usize)
})
.map(|(index, slot)| api::SlotStatus {
slot: index + 1,
runner: slot.as_ref().map(|occupant| api::RunnerStatus {
repository: occupant.repository.clone(),
name: occupant.runner_name.clone(),
runner_id: occupant.runner_id,
container: occupant.container.clone(),
busy: occupant.busy || occupant.job.is_some(),
started_at: unix_secs(occupant.started_at),
reserved: self.takes(occupant),
warm: plan.warm.contains(&SlotRef {
profile: name.clone(),
index,
}),
}),
})
.collect(),
})
.collect();
// The queue as clients list it: a job a runner took, which GitHub still lists for a few seconds, is not counted.
let queued = self.queued(&plan.waiting);
let repositories = self.repository_statuses(&queued);
let previous = self.status.borrow().clone();
let announced = self.announced_profiles.take();
let status = api::Status {
version: env!("CARGO_PKG_VERSION").to_owned(),
shutting_down: mode != Shutdown::No,
budget: self.budget(),
profiles,
repositories,
queued,
};
self.announce_changes(&previous, announced.as_deref(), &status);
self.status.send_replace(status);
self.publish_records();
}
/// Each configured repository once, with its pools and its jobs in `queued`.
fn repository_statuses(&self, queued: &[api::QueuedJob]) -> Vec<api::RepositoryStatus> {
self.config
.repository_names()
.into_iter()
.map(|name| {
let pools: Vec<api::PoolStatus> = self
.config
.repositories
.iter()
.filter(|pool| pool.name == name)
.map(|pool| api::PoolStatus {
profile: pool.profile.clone(),
labels: pool.labels.clone(),
queued_jobs: queued
.iter()
.filter(|job| {
job.repository == name
&& job.profile.as_ref() == Some(&pool.profile)
})
.count(),
})
.collect();
api::RepositoryStatus {
profile: pools
.first()
.map(|pool| pool.profile.clone())
.unwrap_or_default(),
labels: pools
.first()
.map(|pool| pool.labels.clone())
.unwrap_or_default(),
pools,
paused: self.control.is_paused(&name),
queued_jobs: queued.iter().filter(|job| job.repository == name).count(),
name,
}
})
.collect()
}
/// The memory caps in use, with the mirror's and each seed container's part.
fn budget(&self) -> api::Budget {
api::Budget {
slice: self.config.host.slice.clone(),
memory_max: self.memory_max,
memory_in_use: self.memory_in_use(),
admission: match self.config.host.admission {
crate::config::Admission::Caps => api::AdmissionPolicy::Caps,
crate::config::Admission::Measured => api::AdmissionPolicy::Measured,
},
held: self.measured.high_water,
host: self.host_budget(),
mirror_memory: self.mirror_memory_in_use(),
seed_memory: self
.seed_memory()
.into_iter()
.map(|(profile, memory)| api::SeedMemory { profile, memory })
.collect(),
shell_memory: self.shell_budget(),
slice_cpus: self.cpus,
cpu: self
.cpus
.zip(self.cpu_budget())
.map(|(cpus, budget)| api::CpuBudget {
cpus,
budget,
in_use: self.cpu_in_use(),
held: self
.repositories
.values()
.any(|state| !state.cpu_held.is_empty()),
}),
}
}
/// The queued jobs the repositories' pools can take and none holds, and those the plan says no pool serves,
/// oldest first, with why each `waiting`; none of a repository whose last listing failed or was skipped.
fn queued(&self, waiting: &BTreeMap<u64, api::Waiting>) -> Vec<api::QueuedJob> {
let taken: BTreeSet<u64> = self
.occupied()
.filter_map(|(_, occupant)| occupant.job)
.collect();
let mut queued: Vec<api::QueuedJob> = self
.repositories
.iter()
.flat_map(|(repository, state)| {
// Unlisted at the last tick: a failed listing or a shutdown, which the history does not follow.
let queued = if state.listed { &state.queued[..] } else { &[] };
queued
.iter()
.filter(|job| !taken.contains(&job.id))
.filter_map(move |job| {
let pool = self.config.pool(repository, &job.labels);
let unserved =
matches!(waiting.get(&job.id), Some(api::Waiting::NoPool { .. }));
(pool.is_some() || unserved).then_some((job, pool))
})
.map(|(job, pool)| api::QueuedJob {
id: job.id,
repository: repository.clone(),
profile: pool.map(|pool| pool.profile.clone()),
run_id: job.run_id,
run_attempt: job.run_attempt,
workflow: job.workflow.clone(),
name: job.name.clone(),
branch: job.branch.clone(),
url: job.url.clone(),
queued_at: job.created_at / 1000,
waiting: waiting.get(&job.id).cloned(),
})
})
.collect();
// IDs grow with time.
queued.sort_by_key(|job| job.id);
queued
}
/// Announces the profiles, slots, budget and repositories that changed between two statuses. The events carry
/// state, not transitions: a pause undone before the next publish announces nothing, and a client that applies
/// the events still ends where the status is. `announced` replaces the previous profiles on the first publish;
/// their slots are compared with nothing then, so every slot is announced.
fn announce_changes(
&self,
previous: &api::Status,
announced: Option<&[api::ProfileStatus]>,
status: &api::Status,
) {
let profiles_before = announced.unwrap_or(&previous.profiles);
for profile in &status.profiles {
let outline_unchanged = profiles_before
.iter()
.find(|candidate| candidate.name == profile.name)
.is_some_and(|before| same_outline(before, profile));
if !outline_unchanged {
self.store.event(api::Event::ProfileChanged {
profile: profile.clone(),
});
}
let before = previous
.profiles
.iter()
.find(|candidate| candidate.name == profile.name);
for slot in &profile.slots {
let unchanged = before
.and_then(|before| before.slots.iter().find(|old| old.slot == slot.slot))
.is_some_and(|old| old == slot);
if !unchanged {
self.store.event(api::Event::SlotChanged {
profile: profile.name.clone(),
slot: slot.clone(),
});
}
}
}
for before in profiles_before {
if !status
.profiles
.iter()
.any(|profile| profile.name == before.name)
{
self.store.event(api::Event::ProfileRemoved {
profile: before.name.clone(),
});
}
}
if previous.budget != status.budget {
self.store.event(api::Event::BudgetChanged {
budget: status.budget.clone(),
});
}
for repository in &status.repositories {
if !previous.repositories.contains(repository) {
self.store.event(api::Event::RepositoryChanged {
repository: repository.clone(),
});
}
}
for before in &previous.repositories {
if !status
.repositories
.iter()
.any(|repository| repository.name == before.name)
{
self.store.event(api::Event::RepositoryRemoved {
repository: before.name.clone(),
});
}
}
}
}
/// Whether two statuses of a profile agree on everything but the runners in its slots, which `slot.changed` announces.
fn same_outline(before: &api::ProfileStatus, now: &api::ProfileStatus) -> bool {
let api::ProfileStatus {
name,
image,
cpus,
memory,
reseeding,
slots,
} = now;
*name == before.name
&& *image == before.image
&& *cpus == before.cpus
&& *memory == before.memory
&& *reseeding == before.reseeding
&& slots
.iter()
.map(|slot| slot.slot)
.eq(before.slots.iter().map(|slot| slot.slot))
}