Skip to main content

runnerd_core/
config.rs

1//! `runnerd.toml`: the repositories to serve, the runner profiles they use, the shared caches and the host
2//! budget they must fit in. Builds without the `host` feature, so clients can read a config; its only I/O is
3//! reading the file and asking `systemctl` for a slice's `MemoryMax`. It never talks to Docker or GitHub.
4
5use std::collections::BTreeMap;
6use std::fmt;
7use std::net::{Ipv4Addr, Ipv6Addr};
8use std::path::{Path, PathBuf};
9use std::process::Command;
10use std::str::FromStr;
11use std::time::Duration;
12
13use serde::{Deserialize, Deserializer, de};
14
15use crate::api;
16
17mod budget;
18
19pub use budget::{BudgetBlock, CheckBudget, Remainder, Verdict};
20
21/// A parsed `runnerd.toml`. `validate` checks it against the host's budget.
22#[derive(Debug, Deserialize)]
23#[serde(deny_unknown_fields)]
24pub struct Config {
25    /// The host's budget and isolation.
26    pub host: Host,
27    /// The caches runners share.
28    pub caches: Caches,
29    /// Runner profiles by name.
30    pub profiles: BTreeMap<String, Profile>,
31    /// The repositories to serve, one pool per entry: a repository named by several entries is served by each of
32    /// their profiles, a job by the one whose labels match it (ADR 0058).
33    pub repositories: Vec<Repository>,
34    /// What is kept of past jobs.
35    #[serde(default)]
36    pub history: History,
37    /// How much the records of failed jobs may keep.
38    #[serde(default)]
39    pub records: Records,
40    /// How sizing advice reads the job history.
41    #[serde(default)]
42    pub sizing: Sizing,
43    /// Who may use the dashboard through another address.
44    #[serde(default)]
45    pub web: Web,
46    /// Where the daemon sends notifications; none by default.
47    #[serde(default)]
48    pub notify: Notify,
49}
50
51/// `[host]`: where runner containers run and how they are isolated.
52#[derive(Debug, Deserialize)]
53#[serde(deny_unknown_fields)]
54pub struct Host {
55    /// The systemd slice every runner container is placed in; its `MemoryMax` is the budget.
56    pub slice: String,
57    /// How runner containers are kept from the host.
58    pub isolation: Isolation,
59    /// Must be set for `isolation = "unsafe-dind"`, so a copied config cannot turn it on silently.
60    #[serde(default)]
61    pub allow_unsafe: bool,
62    /// How often GitHub is asked for queued jobs.
63    #[serde(default = "default_poll_interval", deserialize_with = "duration")]
64    pub poll_interval: Duration,
65    /// What a runner must fit in before it starts. Default `caps`.
66    #[serde(default)]
67    pub admission: Admission,
68    /// Set, a runner also starts only when its predicted CPU fits the slice's CPUs times this; absent, CPU is not
69    /// admitted.
70    #[serde(default)]
71    pub cpu_overcommit: Option<f64>,
72}
73
74impl Host {
75    /// Refuses a `cpu_overcommit` that is not above 0 and at most `MAX_CPU_OVERCOMMIT`.
76    fn check_cpu_overcommit(&self) -> Result<(), ConfigError> {
77        match self.cpu_overcommit {
78            Some(value) if !(value > 0.0 && value <= MAX_CPU_OVERCOMMIT) => {
79                Err(ConfigError::CpuOvercommit { value })
80            }
81            _ => Ok(()),
82        }
83    }
84}
85
86/// The largest `cpu_overcommit` the config check takes.
87pub const MAX_CPU_OVERCOMMIT: f64 = 16.0;
88
89/// `admission` under `[host]`: what a new runner must fit in the slice's `MemoryMax` beside the running ones.
90#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Deserialize)]
91#[serde(rename_all = "kebab-case")]
92pub enum Admission {
93    /// Every runner's memory cap: `slots × memory` over all profiles, with the mirror, must fit `MemoryMax`.
94    #[default]
95    Caps,
96    /// Every runner's predicted peak, from the job history; `slots` only bounds how many run at once, and one
97    /// runner of each profile with the mirror must fit `MemoryMax`. Nothing starts while the slice is at 90% of it.
98    Measured,
99}
100
101impl Admission {
102    /// The policy as the API names it.
103    pub fn policy(self) -> api::AdmissionPolicy {
104        match self {
105            Admission::Caps => api::AdmissionPolicy::Caps,
106            Admission::Measured => api::AdmissionPolicy::Measured,
107        }
108    }
109}
110
111fn default_poll_interval() -> Duration {
112    Duration::from_secs(15)
113}
114
115/// How runner containers are kept from the host: `isolation` under `[host]`.
116#[derive(Debug, Clone, Copy, PartialEq, Eq, Deserialize)]
117#[serde(rename_all = "kebab-case")]
118pub enum Isolation {
119    /// Sysbox: Docker inside an unprivileged container.
120    Sysbox,
121    /// Privileged Docker-in-Docker: a job can take over the host.
122    UnsafeDind,
123}
124
125/// `[caches]`: what runners share and keep between jobs.
126#[derive(Debug, Clone, Deserialize)]
127#[serde(deny_unknown_fields)]
128pub struct Caches {
129    /// The Docker Hub mirror.
130    pub mirror: Mirror,
131    /// The shared pnpm store.
132    pub pnpm: CacheLimit,
133    /// The shared Maven repository.
134    pub maven: CacheLimit,
135    /// When a profile has no job running and its slot 1 changed its layers by at least this much, slot 1's
136    /// layers become the profile's new snapshot.
137    #[serde(default = "default_reseed_at")]
138    pub reseed_at: ByteSize,
139    /// A reseed that would make a profile's snapshot larger than this starts it over from slot 1's changes alone,
140    /// or from the image when those are larger too.
141    #[serde(default = "default_snapshot_max")]
142    pub snapshot_max: ByteSize,
143    /// The build cache every runner shares.
144    #[serde(default)]
145    pub build: BuildCache,
146}
147
148/// `[caches.build]`: the build cache every runner shares (ADR 0052): Gradle's, and the other tools' the runner image
149/// points at it. Every field has a default.
150#[derive(Debug, Clone, PartialEq, Eq, Deserialize)]
151#[serde(deny_unknown_fields, default)]
152pub struct BuildCache {
153    /// Whether runners get it. Default true.
154    pub enabled: bool,
155    /// Above this size, the entries used least recently go when no job runs. Default 10G.
156    pub max_size: ByteSize,
157    /// The branches whose pushes, scheduled and manually started runs may write to it besides each repository's
158    /// default branch, which always may. Every other job only reads. Default empty.
159    pub branches: Vec<String>,
160    /// Where its entries are kept; default `$XDG_CACHE_HOME/runnerd/build-cache`.
161    pub directory: Option<PathBuf>,
162}
163
164impl BuildCache {
165    fn validate(&self) -> Result<(), ConfigError> {
166        if self.enabled && self.max_size.0 == 0 {
167            return Err(ConfigError::ZeroLimit {
168                field: "caches.build.max_size".to_owned(),
169            });
170        }
171        Ok(())
172    }
173}
174
175impl Default for BuildCache {
176    fn default() -> BuildCache {
177        BuildCache {
178            enabled: true,
179            max_size: ByteSize(10 << 30),
180            branches: Vec::new(),
181            directory: None,
182        }
183    }
184}
185
186fn default_reseed_at() -> ByteSize {
187    ByteSize(2 << 30)
188}
189
190/// Above the 8.7G a profile running Flutter and Android builds needs: its Flutter SDK, NDK and Gradle caches.
191fn default_snapshot_max() -> ByteSize {
192    ByteSize(10 << 30)
193}
194
195fn default_mirror_max_size() -> ByteSize {
196    ByteSize(10 << 30)
197}
198
199/// The Docker Hub pull-through mirror shared by all runners.
200#[derive(Debug, Clone, Deserialize)]
201#[serde(deny_unknown_fields)]
202pub struct Mirror {
203    /// Whether runners pull through the mirror.
204    pub enabled: bool,
205    /// The mirror container's memory limit.
206    pub memory: ByteSize,
207    /// The mirror container's CPUs.
208    pub cpus: u32,
209    /// How long content not pulled again is kept.
210    #[serde(deserialize_with = "duration")]
211    pub ttl: Duration,
212    /// Its storage is emptied above this size when no job runs.
213    #[serde(default = "default_mirror_max_size")]
214    pub max_size: ByteSize,
215}
216
217/// A size limit on a shared cache.
218#[derive(Debug, Clone, Deserialize)]
219#[serde(deny_unknown_fields)]
220pub struct CacheLimit {
221    /// The cache is emptied above this size when no job runs.
222    pub max_size: ByteSize,
223}
224
225/// Where runnerd's release workflow publishes the runner images, one tag per stage and version, such as
226/// `ghcr.io/ayagmar/runnerd-runner:jvm-0.0.16`.
227pub const PUBLISHED_RUNNER: &str = "ghcr.io/ayagmar/runnerd-runner";
228
229/// A runner image with its limits; every job of the profile gets one container with these limits.
230#[derive(Debug, Clone, Deserialize)]
231#[serde(try_from = "ProfileToml")]
232pub struct Profile {
233    /// The image its runners run: `image_stage` or `image`, exactly one of them.
234    pub image: Image,
235    /// CPUs per slot.
236    pub cpus: u32,
237    /// Memory per slot; the container's hard limit.
238    pub memory: ByteSize,
239    /// The most jobs of this profile running at once.
240    pub slots: u32,
241    /// Whether a failed job's runner is kept as a record, to open a shell in (`runnerctl shell`).
242    pub record_failures: bool,
243    /// The least `memory` auto sizing lowers it to; `None` for no bound.
244    pub min_memory: Option<ByteSize>,
245    /// The most `memory` auto sizing raises it to; `None` for what the slice's `MemoryMax` holds.
246    pub max_memory: Option<ByteSize>,
247    /// The most `slots` auto sizing raises them to; `None` for as many as the slice's `MemoryMax` holds.
248    pub max_slots: Option<u32>,
249}
250
251/// The image a profile's runners run.
252#[derive(Debug, Clone, PartialEq, Eq)]
253pub enum Image {
254    /// `image_stage`: a stage of `image/Dockerfile` built on this host as `runnerd-runner:<stage>`.
255    Stage(String),
256    /// `image`: a reference runnerd pulls when Docker lacks it, such as
257    /// `ghcr.io/ayagmar/runnerd-runner:jvm-0.0.16`.
258    Pulled(String),
259}
260
261impl Image {
262    /// The reference Docker runs.
263    pub fn reference(&self) -> String {
264        match self {
265            Image::Stage(stage) => format!("runnerd-runner:{stage}"),
266            Image::Pulled(reference) => reference.clone(),
267        }
268    }
269}
270
271/// `[profiles.<name>]` as written, before `Profile` checks it names one image.
272#[derive(Deserialize)]
273#[serde(deny_unknown_fields)]
274struct ProfileToml {
275    image_stage: Option<String>,
276    image: Option<String>,
277    cpus: u32,
278    memory: ByteSize,
279    slots: u32,
280    #[serde(default)]
281    record_failures: bool,
282    min_memory: Option<ByteSize>,
283    max_memory: Option<ByteSize>,
284    max_slots: Option<u32>,
285}
286
287impl TryFrom<ProfileToml> for Profile {
288    type Error = String;
289
290    fn try_from(profile: ProfileToml) -> Result<Profile, String> {
291        let image = match (profile.image_stage, profile.image) {
292            (Some(stage), None) if !stage.is_empty() => Image::Stage(stage),
293            (None, Some(reference)) if !reference.is_empty() => Image::Pulled(reference),
294            (Some(_), Some(_)) => {
295                return Err(
296                    "set image_stage (a stage built from image/Dockerfile) or image (a reference to pull), \
297                     not both"
298                        .to_owned(),
299                );
300            }
301            _ => {
302                return Err(
303                    "set image_stage (a stage built from image/Dockerfile) or image (a reference to pull)"
304                        .to_owned(),
305                );
306            }
307        };
308        Ok(Profile {
309            image,
310            cpus: profile.cpus,
311            memory: profile.memory,
312            slots: profile.slots,
313            record_failures: profile.record_failures,
314            min_memory: profile.min_memory,
315            max_memory: profile.max_memory,
316            max_slots: profile.max_slots,
317        })
318    }
319}
320
321/// Whether a job asking for `wanted` may run on runners registered with `labels`: every label it asks for is one of
322/// theirs. GitHub compares labels without regard to case. A job asking for no label, as one that names only a runner
323/// group, matches no runners: nothing says which pool it is for (ADR 0058).
324pub fn labels_match(wanted: &[String], labels: &[String]) -> bool {
325    !wanted.is_empty()
326        && wanted.iter().all(|wanted| {
327            labels
328                .iter()
329                .any(|label| label.eq_ignore_ascii_case(wanted))
330        })
331}
332
333/// A repository to serve: `[[repositories]]`. Each entry is one of the repository's pools; entries with the same
334/// `name` serve it in several profiles, each with its own labels, which no two of them share (ADR 0058).
335#[derive(Debug, Deserialize)]
336#[serde(deny_unknown_fields)]
337pub struct Repository {
338    /// `owner/name` on GitHub.
339    pub name: String,
340    /// The profile the jobs this pool's labels match run in.
341    pub profile: String,
342    /// The labels its runners register with; a job runs on them when every label it asks for is one of these.
343    pub labels: Vec<String>,
344    /// How many idle runners it keeps registered for jobs not queued yet while it is active (ADR 0050); 0 turns warm
345    /// runners off. Default 1.
346    #[serde(default = "default_warm")]
347    pub warm: u32,
348    /// Whether its runners get the shared build cache, when `[caches.build]` is enabled. Default true.
349    #[serde(default = "default_true")]
350    pub build_cache: bool,
351    /// Whose runs may write to the shared build cache besides the branches' (ADR 0053), the same in every pool of
352    /// the repository. Default `"default-branch"`.
353    #[serde(default)]
354    pub build_cache_writers: BuildCacheWriters,
355}
356
357/// `build_cache_writers` of a `[[repositories]]`: whose jobs' uploads the shared build cache keeps (ADR 0053).
358#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Deserialize)]
359#[serde(rename_all = "kebab-case")]
360pub enum BuildCacheWriters {
361    /// Pushes, scheduled and manually started runs of the default branch and of `[caches.build] branches`.
362    #[default]
363    DefaultBranch,
364    /// Those, and `pull_request` runs of the repository's own branches that people, not bots, opened and pushed.
365    PullRequests,
366}
367
368fn default_true() -> bool {
369    true
370}
371
372fn default_warm() -> u32 {
373    1
374}
375
376/// `[history]`: the job history database and how long it keeps what. Every field has a default.
377#[derive(Debug, Clone, PartialEq, Eq, Deserialize)]
378#[serde(deny_unknown_fields, default)]
379pub struct History {
380    /// The SQLite database; default `$XDG_STATE_HOME/runnerd/runnerd.db`.
381    pub database: Option<PathBuf>,
382    /// How long CPU and memory samples are kept; a job's totals and peaks stay. Default 7 days.
383    #[serde(deserialize_with = "duration")]
384    pub samples: Duration,
385    /// How long a finished job's log is kept. Default 30 days.
386    #[serde(deserialize_with = "duration")]
387    pub logs: Duration,
388    /// How long events are kept for clients resuming with `Last-Event-ID`. Default 30 days.
389    #[serde(deserialize_with = "duration")]
390    pub events: Duration,
391    /// How long a finished job's record is kept, with its steps, totals and runner. Default 90 days.
392    #[serde(deserialize_with = "duration")]
393    pub jobs: Duration,
394    /// The most log stored per job; later lines are dropped. Default 50M.
395    pub log_max: ByteSize,
396}
397
398impl Default for History {
399    fn default() -> History {
400        History {
401            database: None,
402            samples: Duration::from_hours(7 * 24),
403            logs: Duration::from_hours(30 * 24),
404            events: Duration::from_hours(30 * 24),
405            jobs: Duration::from_hours(90 * 24),
406            log_max: ByteSize(50 << 20),
407        }
408    }
409}
410
411/// `[records]`: how much the records of failed jobs may keep (`record_failures` under a profile turns them on).
412/// When a new record would pass a limit, the oldest records go first. Every field has a default.
413#[derive(Debug, Clone, PartialEq, Eq, Deserialize)]
414#[serde(deny_unknown_fields, default)]
415pub struct Records {
416    /// The most disk all records may take together. Default 20G.
417    pub max_total: ByteSize,
418    /// The most records kept. Default 10.
419    pub max_count: u32,
420    /// How long a record is kept. Default 7 days.
421    #[serde(deserialize_with = "duration")]
422    pub max_age: Duration,
423    /// The disk Docker must keep free: no record is made that would leave less. Default 10G.
424    pub min_free: ByteSize,
425}
426
427impl Default for Records {
428    fn default() -> Records {
429        Records {
430            max_total: ByteSize(20 << 30),
431            max_count: 10,
432            max_age: Duration::from_hours(7 * 24),
433            min_free: ByteSize(10 << 30),
434        }
435    }
436}
437
438/// `[sizing]`: how `runnerctl sizing` turns the job history's peaks into advice, and whether the daemon applies it
439/// itself. Every field has a default.
440#[derive(Debug, Clone, PartialEq, Eq, Deserialize)]
441#[serde(deny_unknown_fields, default)]
442pub struct Sizing {
443    /// The percent added to the largest sampled memory peak: a container is sampled every 2 s, so a spike between
444    /// two samples goes unseen. Default 15. Advice adds at least `api::PAGE_CACHE_ROOM`.
445    pub margin: u32,
446    /// The fewest jobs a profile, or a job of a repository's workflow, needs before its peaks become advice.
447    /// Default 10.
448    pub min_jobs: u32,
449    /// Whether the daemon writes each profile's advice into the config itself, within each profile's
450    /// `min_memory`, `max_memory` and `max_slots`. Default off.
451    pub auto: bool,
452    /// The least time between a change auto sizing makes to one profile and one that lowers anything in it. Default
453    /// 24 hours.
454    #[serde(deserialize_with = "duration")]
455    pub interval: Duration,
456    /// The least time between a change auto sizing makes to one profile and one that only raises its `memory`,
457    /// `slots` or `cpus`; at most `interval`, which a longer one counts as. Default 1 hour.
458    #[serde(deserialize_with = "duration")]
459    pub raise_interval: Duration,
460    /// The most a change moves a profile's `memory` or `cpus`, in percent of them. Default 25.
461    pub max_step: u32,
462}
463
464impl Sizing {
465    /// The least time between a change to one profile and one that only raises it: `raise_interval`, at most
466    /// `interval`.
467    pub fn raise_after(&self) -> Duration {
468        self.raise_interval.min(self.interval)
469    }
470}
471
472impl Default for Sizing {
473    fn default() -> Sizing {
474        Sizing {
475            margin: 15,
476            min_jobs: 10,
477            auto: false,
478            interval: Duration::from_hours(24),
479            raise_interval: Duration::from_hours(1),
480            max_step: 25,
481        }
482    }
483}
484
485/// `[web]`: the dashboard behind a reverse proxy, such as `tailscale serve`. Read when the daemon starts.
486#[derive(Debug, Clone, Default, PartialEq, Eq, Deserialize)]
487#[serde(deny_unknown_fields, default)]
488pub struct Web {
489    /// The origins, besides the dashboard's own loopback address, whose pages may write: pause, reseed, save the
490    /// config. Default none.
491    pub origins: Vec<Origin>,
492    /// The one of `origins` that `runnerctl web` opens the dashboard on, such as a tailnet's address. Default none:
493    /// it opens the loopback address.
494    pub open: Option<Origin>,
495}
496
497impl Web {
498    /// Refuses an `open` that is not one of `origins`.
499    fn validate(&self) -> Result<(), ConfigError> {
500        match &self.open {
501            Some(open) if !self.origins.contains(open) => {
502                Err(ConfigError::OpenNotListed { open: open.clone() })
503            }
504            _ => Ok(()),
505        }
506    }
507}
508
509/// `[notify]`: where the daemon sends notifications (ADR 0051). Without targets nothing leaves the machine.
510#[derive(Debug, Clone, PartialEq, Eq, Deserialize)]
511#[serde(deny_unknown_fields, default)]
512pub struct Notify {
513    /// When the digest goes to the targets that want it, in the host's local time. Default `09:00`.
514    pub digest_at: DayTime,
515    /// The targets, each `[[notify.targets]]`.
516    pub targets: Vec<NotifyTarget>,
517}
518
519impl Default for Notify {
520    fn default() -> Notify {
521        Notify {
522            digest_at: DayTime { minutes: 9 * 60 },
523            targets: Vec::new(),
524        }
525    }
526}
527
528impl Notify {
529    /// Refuses a target without a name, with a name another has, with a URL that is not `http` or `https`, or with
530    /// a header that cannot be sent. Says which target and which field, never the URL or a header's value.
531    fn validate(&self) -> Result<(), ConfigError> {
532        let refuse = |target: &str, why: &'static str| {
533            Err(ConfigError::Notify {
534                target: target.to_owned(),
535                why,
536            })
537        };
538        for (index, target) in self.targets.iter().enumerate() {
539            let name = target.name.as_str();
540            if name.trim().is_empty() || name.chars().any(char::is_control) {
541                return refuse(name, "name is empty");
542            }
543            if self.targets[..index].iter().any(|other| other.name == name) {
544                return refuse(name, "another target has this name");
545            }
546            let url = target.url.expose();
547            if url == SECRET_SHOWN {
548                return refuse(
549                    name,
550                    "url is <secret>, which runnerd fills in only for a target of the same name that had one; write the url",
551                );
552            }
553            let Some(rest) = url
554                .strip_prefix("https://")
555                .or_else(|| url.strip_prefix("http://"))
556            else {
557                return refuse(name, "url does not start with https:// or http://");
558            };
559            let (host, path) = rest.split_once('/').unwrap_or((rest, ""));
560            if host.is_empty() || url.chars().any(|c| c.is_whitespace() || c.is_control()) {
561                return refuse(name, "url is not a URL");
562            }
563            if target.kind == NotifyKind::Ntfy
564                && (path.is_empty() || path.contains(['/', '?', '#']))
565            {
566                return refuse(
567                    name,
568                    "url is not an ntfy topic's, such as https://ntfy.sh/<topic>",
569                );
570            }
571            for (header, value) in &target.headers {
572                if value.expose() == SECRET_SHOWN {
573                    return refuse(
574                        name,
575                        "a header is <secret>, which runnerd fills in only for a target of the same name that had it; write its value",
576                    );
577                }
578                let token = |c: char| c.is_ascii_alphanumeric() || "!#$%&'*+-.^_`|~".contains(c);
579                if header.is_empty() || !header.chars().all(token) {
580                    return refuse(name, "a header's name is not a header name");
581                }
582                if value.expose().chars().any(char::is_control) {
583                    return refuse(name, "a header's value has a control character");
584                }
585            }
586            if target.events.contains(&api::NotifyEvent::Test) {
587                return refuse(
588                    name,
589                    "events lists test, which only runnerctl notify test sends",
590                );
591            }
592        }
593        Ok(())
594    }
595}
596
597/// `[[notify.targets]]`: one place notifications go.
598#[derive(Debug, Clone, PartialEq, Eq, Deserialize)]
599#[serde(deny_unknown_fields)]
600pub struct NotifyTarget {
601    /// What logs and `runnerctl notify test` call it, such as `phone`; the history keeps where it got to by it.
602    pub name: String,
603    /// What it takes.
604    pub kind: NotifyKind,
605    /// Where it is posted: the webhook's URL, Discord's included, or the ntfy topic's, such as `https://ntfy.sh/<topic>`.
606    /// A secret.
607    pub url: Secret,
608    /// Headers sent with each post, such as an ntfy access token's `Authorization`. Their values are secrets.
609    #[serde(default)]
610    pub headers: BTreeMap<String, Secret>,
611    /// What it is sent. Default `failed`, `lost`, `runner` and `insights`.
612    #[serde(default = "default_notify_events")]
613    pub events: Vec<api::NotifyEvent>,
614}
615
616fn default_notify_events() -> Vec<api::NotifyEvent> {
617    vec![
618        api::NotifyEvent::Failed,
619        api::NotifyEvent::Lost,
620        api::NotifyEvent::Runner,
621        api::NotifyEvent::Insights,
622    ]
623}
624
625/// What a notification target takes.
626#[derive(Debug, Clone, Copy, PartialEq, Eq, Deserialize)]
627#[serde(rename_all = "kebab-case")]
628pub enum NotifyKind {
629    /// `api::Notification` as JSON.
630    Webhook,
631    /// An ntfy topic: the title, message, priority and link, published as ntfy's JSON.
632    Ntfy,
633    /// A Discord channel's webhook: the title, message and link as one embed, coloured by event, that pings no one.
634    Discord,
635}
636
637/// A value no `Debug` output, log line or error shows: a notification target's URL or header.
638#[derive(Clone, PartialEq, Eq, Deserialize)]
639#[serde(transparent)]
640pub struct Secret(String);
641
642impl Secret {
643    /// The value, for the request that sends it.
644    pub fn expose(&self) -> &str {
645        &self.0
646    }
647}
648
649impl From<&str> for Secret {
650    fn from(value: &str) -> Secret {
651        Secret(value.to_owned())
652    }
653}
654
655impl fmt::Debug for Secret {
656    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
657        f.write_str("<secret>")
658    }
659}
660
661/// A time of day, `HH:MM`, 24-hour.
662#[derive(Debug, Clone, Copy, PartialEq, Eq)]
663pub struct DayTime {
664    /// Minutes after midnight.
665    pub minutes: u32,
666}
667
668impl FromStr for DayTime {
669    type Err = String;
670
671    fn from_str(text: &str) -> Result<DayTime, String> {
672        let invalid = || format!("invalid time of day {text:?}, expected HH:MM such as \"09:00\"");
673        let (hour, minute) = text.split_once(':').ok_or_else(invalid)?;
674        let digits = |part: &str| part.len() == 2 && part.chars().all(|c| c.is_ascii_digit());
675        if !digits(hour) || !digits(minute) {
676            return Err(invalid());
677        }
678        let (hour, minute): (u32, u32) = (
679            hour.parse().map_err(|_| invalid())?,
680            minute.parse().map_err(|_| invalid())?,
681        );
682        if hour > 23 || minute > 59 {
683            return Err(invalid());
684        }
685        Ok(DayTime {
686            minutes: hour * 60 + minute,
687        })
688    }
689}
690
691impl fmt::Display for DayTime {
692    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
693        write!(f, "{:02}:{:02}", self.minutes / 60, self.minutes % 60)
694    }
695}
696
697impl<'de> Deserialize<'de> for DayTime {
698    fn deserialize<D: Deserializer<'de>>(deserializer: D) -> Result<Self, D::Error> {
699        String::deserialize(deserializer)?
700            .parse()
701            .map_err(de::Error::custom)
702    }
703}
704
705/// An origin as a browser sends it in `Origin`: `https://host[:port]`, or `http://` for a loopback host, in lower
706/// case, without the scheme's default port, and an address written as the URL standard serializes it. Nothing after
707/// the port: no path, query or user.
708#[derive(Debug, Clone, PartialEq, Eq)]
709pub struct Origin(String);
710
711impl Origin {
712    /// The origin as a browser sends it, such as `https://host.example.ts.net:8443`.
713    pub fn as_str(&self) -> &str {
714        &self.0
715    }
716}
717
718impl FromStr for Origin {
719    type Err = String;
720
721    fn from_str(s: &str) -> Result<Origin, String> {
722        let invalid = |why: &str| {
723            format!(
724                "invalid origin {s:?}: {why}; write the scheme, host and port only, such as \
725                 \"https://host.example.ts.net:8443\""
726            )
727        };
728        let (scheme, rest) = s
729            .split_once("://")
730            .ok_or_else(|| invalid("it has no scheme"))?;
731        let scheme = scheme.to_ascii_lowercase();
732        if rest.contains(['/', '?', '#']) {
733            return Err(invalid("it has a path"));
734        }
735        if rest.contains('@') {
736            return Err(invalid("it names a user"));
737        }
738        let (host, port) = match rest.strip_prefix('[') {
739            Some(v6) => {
740                let (address, after) = v6
741                    .split_once(']')
742                    .ok_or_else(|| invalid("its IPv6 address has no closing ]"))?;
743                let port = match after {
744                    "" => None,
745                    _ => Some(after.strip_prefix(':').ok_or_else(|| invalid("its port"))?),
746                };
747                (format!("[{address}]"), port)
748            }
749            None => match rest.split_once(':') {
750                Some((host, port)) => (host.to_owned(), Some(port)),
751                None => (rest.to_owned(), None),
752            },
753        };
754        let host = host.to_ascii_lowercase();
755        let not_a_host = || invalid("its host is not a name or an address");
756        // An address is kept as a browser serializes it, or it would never match the `Origin` the browser sends.
757        let (host, loopback) =
758            if let Some(v6) = host.strip_prefix('[').and_then(|v6| v6.strip_suffix(']')) {
759                let address: Ipv6Addr = v6.parse().map_err(|_| not_a_host())?;
760                (format!("[{}]", ipv6_origin(address)), address.is_loopback())
761            } else {
762                if host.is_empty()
763                    || !host
764                        .chars()
765                        .all(|c| c.is_ascii_alphanumeric() || c == '-' || c == '.')
766                {
767                    return Err(not_a_host());
768                }
769                // As the URL standard reads a host: one whose last label is a number is an IPv4 address.
770                let name = host.strip_suffix('.').unwrap_or(&host);
771                let last = name.rsplit('.').next().unwrap_or_default();
772                let numeric = !last.is_empty()
773                    && (last.bytes().all(|byte| byte.is_ascii_digit())
774                        || last
775                            .strip_prefix("0x")
776                            .is_some_and(|hex| hex.bytes().all(|byte| byte.is_ascii_hexdigit())));
777                if numeric {
778                    let address: Ipv4Addr = name.parse().map_err(|_| {
779                        invalid("its IPv4 address is not four decimal numbers, such as 192.0.2.1")
780                    })?;
781                    (address.to_string(), address.is_loopback())
782                } else {
783                    let loopback = host == "localhost";
784                    (host, loopback)
785                }
786            };
787        let port = port
788            .map(|port| {
789                Some(port)
790                    .filter(|port| port.bytes().all(|byte| byte.is_ascii_digit()))
791                    .and_then(|port| port.parse::<u16>().ok())
792                    .filter(|port| *port != 0)
793                    .ok_or_else(|| invalid("its port is not a number from 1 to 65535"))
794            })
795            .transpose()?;
796        let default_port = match scheme.as_str() {
797            "https" => 443,
798            "http" if loopback => 80,
799            "http" => {
800                return Err(invalid(
801                    "http is for a loopback host only, since the session cookie would cross the network in \
802                     the clear; use https",
803                ));
804            }
805            _ => return Err(invalid("its scheme is not https")),
806        };
807        Ok(Origin(match port {
808            Some(port) if port != default_port => format!("{scheme}://{host}:{port}"),
809            _ => format!("{scheme}://{host}"),
810        }))
811    }
812}
813
814/// An IPv6 address as the URL standard serializes it: lower-case hex pieces, the first longest run of two or more
815/// zero pieces as `::`, and no dotted IPv4 tail.
816fn ipv6_origin(address: Ipv6Addr) -> String {
817    let pieces = address.segments();
818    let (mut zeros_at, mut zeros) = (0, 0);
819    let mut at = 0;
820    while at < pieces.len() {
821        let run = pieces[at..].iter().take_while(|piece| **piece == 0).count();
822        if run > zeros {
823            (zeros_at, zeros) = (at, run);
824        }
825        at += run.max(1);
826    }
827    let hex = |pieces: &[u16]| {
828        pieces
829            .iter()
830            .map(|piece| format!("{piece:x}"))
831            .collect::<Vec<_>>()
832            .join(":")
833    };
834    if zeros < 2 {
835        return hex(&pieces);
836    }
837    format!(
838        "{}::{}",
839        hex(&pieces[..zeros_at]),
840        hex(&pieces[zeros_at + zeros..])
841    )
842}
843
844impl fmt::Display for Origin {
845    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
846        f.write_str(&self.0)
847    }
848}
849
850impl<'de> Deserialize<'de> for Origin {
851    fn deserialize<D: Deserializer<'de>>(deserializer: D) -> Result<Self, D::Error> {
852        String::deserialize(deserializer)?
853            .parse()
854            .map_err(de::Error::custom)
855    }
856}
857
858/// Why a config cannot be used.
859#[derive(Debug, thiserror::Error)]
860pub enum ConfigError {
861    /// The file cannot be read.
862    #[error("cannot read {}: {source}", path.display())]
863    Read {
864        /// The config file.
865        path: PathBuf,
866        /// Why reading it failed.
867        source: std::io::Error,
868    },
869    /// The file is not valid TOML or does not match the config's shape.
870    #[error(transparent)]
871    Parse(#[from] toml::de::Error),
872    /// A repository names a profile `[profiles]` does not define.
873    #[error(
874        "repository {repository} uses profile {profile:?}, which is not defined under [profiles]; {}",
875        defined_profiles(.defined)
876    )]
877    UnknownProfile {
878        /// The repository's `owner/name`.
879        repository: String,
880        /// The profile it names.
881        profile: String,
882        /// The profiles `[profiles]` defines.
883        defined: Vec<String>,
884    },
885    /// A repository has no labels, or an empty one.
886    #[error("repository {repository} needs at least one non-empty label")]
887    MissingLabels {
888        /// The repository's `owner/name`.
889        repository: String,
890    },
891    /// Two `[[repositories]]` entries of one repository use the same profile.
892    #[error(
893        "repository {repository} has two entries with profile {profile:?}; give each of its pools its own profile"
894    )]
895    PoolProfileTwice {
896        /// The repository's `owner/name`.
897        repository: String,
898        /// The profile both entries name.
899        profile: String,
900    },
901    /// Two `[[repositories]]` entries of one repository share a label, so a job asking only for labels they share
902    /// could run in either.
903    #[error(
904        "repository {repository}'s pools in {first} and {second} both have the label {label:?}, so a job asking for \
905         it could land in either; give each pool labels of its own, such as only {second:?}'s own label for the \
906         lighter one"
907    )]
908    PoolLabelsOverlap {
909        /// The repository's `owner/name`.
910        repository: String,
911        /// The profile of the first entry.
912        first: String,
913        /// The profile of the later entry.
914        second: String,
915        /// A label both have.
916        label: String,
917    },
918    /// Two `[[repositories]]` entries of one repository say differently whose runs write to the build cache.
919    #[error(
920        "repository {repository}'s entries set build_cache_writers differently; whose runs may write to the build \
921         cache is the repository's, so set the same in each"
922    )]
923    PoolWritersDiffer {
924        /// The repository's `owner/name`.
925        repository: String,
926    },
927    /// `unsafe-dind` without `allow_unsafe = true`.
928    #[error(
929        "isolation \"unsafe-dind\" runs runner containers privileged, so a job can take over the host; \
930         set allow_unsafe = true under [host] to use it anyway"
931    )]
932    UnsafeNotAllowed,
933    /// A limit is 0.
934    #[error("{field} is 0; set it above 0")]
935    ZeroLimit {
936        /// The setting, such as `profiles.app.cpus`.
937        field: String,
938    },
939    /// `cpu_overcommit` is not a number above 0 and at most `MAX_CPU_OVERCOMMIT`.
940    #[error(
941        "host.cpu_overcommit is {value}; set it above 0 and at most {MAX_CPU_OVERCOMMIT}, such as 1.0"
942    )]
943    CpuOvercommit {
944        /// The value set.
945        value: f64,
946    },
947    /// `[web] open` is not one of `[web] origins`.
948    #[error(
949        "web.open is {open}, which is not one of [web] origins; add it to origins, or remove open"
950    )]
951    OpenNotListed {
952        /// The origin `open` names.
953        open: Origin,
954    },
955    /// A profile's `memory` or `slots` are outside its `min_memory`, `max_memory` or `max_slots`, or its
956    /// `min_memory` is above its `max_memory`.
957    #[error("profiles.{profile}: {why}")]
958    SizingBounds {
959        /// The profile.
960        profile: String,
961        /// Which bound, such as `memory is above its max_memory`.
962        why: &'static str,
963    },
964    /// A `[[notify.targets]]` entry cannot be used. Never says its URL or a header's value.
965    #[error("notify target {target:?}: {why}")]
966    Notify {
967        /// The target's name.
968        target: String,
969        /// What is wrong, such as `url does not start with https:// or http://`.
970        why: &'static str,
971    },
972    /// The caps add up past `u64::MAX` bytes.
973    #[error("slots × memory over all profiles plus the mirror is too large to add up")]
974    BudgetOverflow,
975    /// `admission = "measured"` with a slice that has no `MemoryMax` to admit against.
976    #[error(
977        "admission = \"measured\" admits runners against the slice {slice}'s MemoryMax, and it has none; install \
978         a unit file for it (`runnerd doctor` prints one)"
979    )]
980    MeasuredWithoutMemoryMax {
981        /// The slice runners run in.
982        slice: String,
983    },
984    /// Under `admission = "measured"`, one runner of a profile and the mirror do not fit `MemoryMax`.
985    #[error(
986        "one runner of {profile} and the mirror take {needs}, the slice {slice}'s MemoryMax is {memory_max}; lower \
987         its `memory` or raise the slice's MemoryMax"
988    )]
989    ProfileOverBudget {
990        /// The profile.
991        profile: String,
992        /// Its `memory` and the mirror's.
993        needs: ByteSize,
994        /// The slice runners run in.
995        slice: String,
996        /// The slice's `MemoryMax`.
997        memory_max: ByteSize,
998    },
999    /// The caps add up to more than the slice's `MemoryMax`.
1000    #[error(
1001        "slots × memory over all profiles plus the mirror is {total}, \
1002         the slice {slice}'s MemoryMax is {memory_max}; lower `slots` or `memory`, or raise the slice's \
1003         MemoryMax (`runnerd doctor` prints its unit)"
1004    )]
1005    OverBudget {
1006        /// What every runner and the mirror may use at once.
1007        total: ByteSize,
1008        /// The slice runners run in.
1009        slice: String,
1010        /// The slice's `MemoryMax`.
1011        memory_max: ByteSize,
1012    },
1013}
1014
1015fn defined_profiles(defined: &[String]) -> String {
1016    if defined.is_empty() {
1017        "it defines none".to_owned()
1018    } else {
1019        format!("defined: {}", defined.join(", "))
1020    }
1021}
1022
1023/// What a valid config should still fix.
1024#[derive(Debug, Clone, PartialEq, Eq)]
1025pub enum Warning {
1026    /// The slice has no unit file, so systemd creates it without limits.
1027    UnlimitedSlice {
1028        /// The slice runners run in.
1029        slice: String,
1030        /// What every runner and the mirror may use at once.
1031        total: ByteSize,
1032    },
1033}
1034
1035impl fmt::Display for Warning {
1036    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1037        match self {
1038            Warning::UnlimitedSlice { slice, total } => write!(
1039                f,
1040                "the slice {slice} has no MemoryMax, so nothing holds runners to their {total} of caps; \
1041                 install a unit file for it"
1042            ),
1043        }
1044    }
1045}
1046
1047impl Config {
1048    /// Reads and parses a config file.
1049    pub fn load(path: &Path) -> Result<Config, ConfigError> {
1050        let text = std::fs::read_to_string(path).map_err(|source| ConfigError::Read {
1051            path: path.to_owned(),
1052            source,
1053        })?;
1054        Config::parse(&text)
1055    }
1056
1057    /// Parses a config from TOML text.
1058    pub fn parse(text: &str) -> Result<Config, ConfigError> {
1059        Ok(toml::from_str(text)?)
1060    }
1061
1062    /// Parses `text` and validates it against the `MemoryMax` that `memory_max` reads for the slice it names:
1063    /// what `runnerd check-config` prints and `POST /v1/config/check` answers. Returns the config if it is valid.
1064    pub fn check(
1065        text: &str,
1066        memory_max: impl FnOnce(&str) -> std::io::Result<Option<ByteSize>>,
1067    ) -> (Option<Config>, api::ConfigCheck) {
1068        let config = match Config::parse(text) {
1069            Ok(config) => config,
1070            Err(error) => return (None, budget::invalid(error.to_string(), None, None)),
1071        };
1072        let max = match memory_max(&config.host.slice) {
1073            Ok(max) => max,
1074            Err(error) => {
1075                return (
1076                    None,
1077                    budget::invalid(error.to_string(), Some(&config), None),
1078                );
1079            }
1080        };
1081        let check = config.check_against(max);
1082        if check.valid {
1083            (Some(config), check)
1084        } else {
1085            (None, check)
1086        }
1087    }
1088
1089    /// The mirror's memory cap, 0 when it is disabled.
1090    pub fn mirror_memory(&self) -> u64 {
1091        if self.caches.mirror.enabled {
1092            self.caches.mirror.memory.0
1093        } else {
1094            0
1095        }
1096    }
1097
1098    /// The memory every runner and the mirror may use at once.
1099    pub fn memory_caps(&self) -> Result<ByteSize, ConfigError> {
1100        let mirror = self.mirror_memory();
1101        self.profiles
1102            .values()
1103            .try_fold(mirror, |total, p| {
1104                p.memory
1105                    .0
1106                    .checked_mul(u64::from(p.slots))?
1107                    .checked_add(total)
1108            })
1109            .map(ByteSize)
1110            .ok_or(ConfigError::BudgetOverflow)
1111    }
1112
1113    /// The memory `[host] admission` must fit in `MemoryMax`: every cap at once under caps; under measured, the
1114    /// largest single runner with the mirror, since `slots` there only bounds how many run at once.
1115    pub fn memory_needed(&self) -> Result<ByteSize, ConfigError> {
1116        match self.host.admission {
1117            Admission::Caps => self.memory_caps(),
1118            Admission::Measured => {
1119                let largest = self
1120                    .profiles
1121                    .values()
1122                    .map(|p| p.memory.0)
1123                    .max()
1124                    .unwrap_or(0);
1125                largest
1126                    .checked_add(self.mirror_memory())
1127                    .map(ByteSize)
1128                    .ok_or(ConfigError::BudgetOverflow)
1129            }
1130        }
1131    }
1132
1133    /// Refuses a profile outside the bounds auto sizing keeps it in, or whose bounds cross.
1134    fn check_sizing_bounds(&self) -> Result<(), ConfigError> {
1135        for (name, p) in &self.profiles {
1136            let memory = p.memory.0;
1137            let field = if p
1138                .min_memory
1139                .zip(p.max_memory)
1140                .is_some_and(|(min, max)| min.0 > max.0)
1141            {
1142                "min_memory is above its max_memory"
1143            } else if p.min_memory.is_some_and(|min| memory < min.0) {
1144                "memory is below its min_memory"
1145            } else if p.max_memory.is_some_and(|max| memory > max.0) {
1146                "memory is above its max_memory"
1147            } else if p.max_slots.is_some_and(|max| p.slots > max) {
1148                "slots are above its max_slots"
1149            } else {
1150                continue;
1151            };
1152            return Err(ConfigError::SizingBounds {
1153                profile: name.clone(),
1154                why: field,
1155            });
1156        }
1157        Ok(())
1158    }
1159
1160    /// Checks that each repository's pools tell its jobs apart: each in its own profile, with labels no other of its
1161    /// pools has, so at most one pool matches a job; and that they agree on whose runs write to the build cache.
1162    fn check_pools(&self) -> Result<(), ConfigError> {
1163        for (at, first) in self.repositories.iter().enumerate() {
1164            for second in self.repositories[at + 1..]
1165                .iter()
1166                .filter(|second| second.name == first.name)
1167            {
1168                if second.profile == first.profile {
1169                    return Err(ConfigError::PoolProfileTwice {
1170                        repository: first.name.clone(),
1171                        profile: first.profile.clone(),
1172                    });
1173                }
1174                if let Some(label) = first.labels.iter().find(|label| {
1175                    second
1176                        .labels
1177                        .iter()
1178                        .any(|other| other.eq_ignore_ascii_case(label))
1179                }) {
1180                    return Err(ConfigError::PoolLabelsOverlap {
1181                        repository: first.name.clone(),
1182                        first: first.profile.clone(),
1183                        second: second.profile.clone(),
1184                        label: label.clone(),
1185                    });
1186                }
1187                if second.build_cache_writers != first.build_cache_writers {
1188                    return Err(ConfigError::PoolWritersDiffer {
1189                        repository: first.name.clone(),
1190                    });
1191                }
1192            }
1193        }
1194        Ok(())
1195    }
1196
1197    /// The pool of `repository` whose labels match a job asking for `labels`: every label it asks for is one of
1198    /// the pool's. `None` when no pool of it does.
1199    pub fn pool(&self, repository: &str, labels: &[String]) -> Option<&Repository> {
1200        self.repositories
1201            .iter()
1202            .find(|pool| pool.name == repository && labels_match(labels, &pool.labels))
1203    }
1204
1205    /// The pool of `repository` in `profile`, whose runners start in that profile's slots.
1206    pub fn pool_in(&self, repository: &str, profile: &str) -> Option<&Repository> {
1207        self.repositories
1208            .iter()
1209            .find(|pool| pool.name == repository && pool.profile == profile)
1210    }
1211
1212    /// The repositories served, each once, in the order the config first names them.
1213    pub fn repository_names(&self) -> Vec<String> {
1214        let mut names: Vec<String> = Vec::new();
1215        for repository in &self.repositories {
1216            if !names.contains(&repository.name) {
1217                names.push(repository.name.clone());
1218            }
1219        }
1220        names
1221    }
1222
1223    /// Checks the config against the slice's `MemoryMax` (`None` when the slice is unlimited).
1224    pub fn validate(&self, memory_max: Option<ByteSize>) -> Result<Vec<Warning>, ConfigError> {
1225        if self.host.isolation == Isolation::UnsafeDind && !self.host.allow_unsafe {
1226            return Err(ConfigError::UnsafeNotAllowed);
1227        }
1228        let zero_limits = self.profiles.iter().flat_map(|(name, p)| {
1229            [
1230                (p.memory.0 == 0, format!("profiles.{name}.memory")),
1231                (p.cpus == 0, format!("profiles.{name}.cpus")),
1232                (p.slots == 0, format!("profiles.{name}.slots")),
1233                (p.max_slots == Some(0), format!("profiles.{name}.max_slots")),
1234            ]
1235        });
1236        let mirror = &self.caches.mirror;
1237        let mirror_zero_limits = [
1238            (
1239                mirror.enabled && mirror.memory.0 == 0,
1240                "caches.mirror.memory".to_owned(),
1241            ),
1242            (
1243                mirror.enabled && mirror.cpus == 0,
1244                "caches.mirror.cpus".to_owned(),
1245            ),
1246            (
1247                self.host.poll_interval.is_zero(),
1248                "host.poll_interval".to_owned(),
1249            ),
1250            (self.records.max_count == 0, "records.max_count".to_owned()),
1251            (self.sizing.min_jobs == 0, "sizing.min_jobs".to_owned()),
1252            (self.sizing.interval.is_zero(), "sizing.interval".to_owned()),
1253            (
1254                self.sizing.raise_interval.is_zero(),
1255                "sizing.raise_interval".to_owned(),
1256            ),
1257            (self.sizing.max_step == 0, "sizing.max_step".to_owned()),
1258            (
1259                self.records.max_total.0 == 0,
1260                "records.max_total".to_owned(),
1261            ),
1262        ];
1263        if let Some((_, field)) = zero_limits
1264            .chain(mirror_zero_limits)
1265            .find(|(zero, _)| *zero)
1266        {
1267            return Err(ConfigError::ZeroLimit { field });
1268        }
1269        self.check_sizing_bounds()?;
1270        self.caches.build.validate()?;
1271        self.host.check_cpu_overcommit()?;
1272        self.web.validate()?;
1273        self.notify.validate()?;
1274        for repository in &self.repositories {
1275            if !self.profiles.contains_key(&repository.profile) {
1276                return Err(ConfigError::UnknownProfile {
1277                    repository: repository.name.clone(),
1278                    profile: repository.profile.clone(),
1279                    defined: self.profiles.keys().cloned().collect(),
1280                });
1281            }
1282            if repository.labels.is_empty() || repository.labels.iter().any(|l| l.trim().is_empty())
1283            {
1284                return Err(ConfigError::MissingLabels {
1285                    repository: repository.name.clone(),
1286                });
1287            }
1288        }
1289        self.check_pools()?;
1290        self.check_budget(memory_max)
1291    }
1292
1293    /// Holds what `[host] admission` needs to the slice's `MemoryMax`; warns when the slice has none to hold caps to.
1294    fn check_budget(&self, memory_max: Option<ByteSize>) -> Result<Vec<Warning>, ConfigError> {
1295        let total = self.memory_caps()?;
1296        if self.host.admission == Admission::Measured {
1297            let Some(memory_max) = memory_max else {
1298                return Err(ConfigError::MeasuredWithoutMemoryMax {
1299                    slice: self.host.slice.clone(),
1300                });
1301            };
1302            let mirror = self.mirror_memory();
1303            if let Some((profile, needs)) = self
1304                .profiles
1305                .iter()
1306                .map(|(name, profile)| (name, profile.memory.0.saturating_add(mirror)))
1307                .find(|(_, needs)| *needs > memory_max.0)
1308            {
1309                return Err(ConfigError::ProfileOverBudget {
1310                    profile: profile.clone(),
1311                    needs: ByteSize(needs),
1312                    slice: self.host.slice.clone(),
1313                    memory_max,
1314                });
1315            }
1316            // Without a profile, the mirror alone must fit: the caps sum, which is all `memory_needed` holds.
1317            let needed = self.memory_needed()?;
1318            if needed > memory_max {
1319                return Err(ConfigError::OverBudget {
1320                    total: needed,
1321                    slice: self.host.slice.clone(),
1322                    memory_max,
1323                });
1324            }
1325            return Ok(Vec::new());
1326        }
1327        match memory_max {
1328            Some(memory_max) if total > memory_max => Err(ConfigError::OverBudget {
1329                total,
1330                slice: self.host.slice.clone(),
1331                memory_max,
1332            }),
1333            Some(_) => Ok(Vec::new()),
1334            None => Ok(vec![Warning::UnlimitedSlice {
1335                slice: self.host.slice.clone(),
1336                total,
1337            }]),
1338        }
1339    }
1340}
1341
1342/// Why `resize_profile` cannot change a config.
1343#[derive(Debug, thiserror::Error)]
1344pub enum ResizeError {
1345    /// The text is not valid TOML.
1346    #[error(transparent)]
1347    Parse(#[from] toml_edit::TomlError),
1348    /// `[profiles]` does not define the profile.
1349    #[error("the config has no profile {0:?} under [profiles]")]
1350    NoProfile(String),
1351}
1352
1353/// `text` with the profile's `memory` and `slots` replaced, and its `cpus` when `resize` has them, everything else as
1354/// written, comments included: what applying a sizing suggestion sends to `PUT /v1/config`, which validates it.
1355pub fn resize_profile(
1356    text: &str,
1357    profile: &str,
1358    resize: api::Resize,
1359) -> Result<String, ResizeError> {
1360    let mut document: toml_edit::DocumentMut = text.parse()?;
1361    let table = document
1362        .get_mut("profiles")
1363        .and_then(|profiles| profiles.get_mut(profile))
1364        .and_then(toml_edit::Item::as_table_like_mut)
1365        .ok_or_else(|| ResizeError::NoProfile(profile.to_owned()))?;
1366    let values = [
1367        (
1368            "memory",
1369            toml_edit::Value::from(ByteSize(resize.memory).to_string()),
1370        ),
1371        ("slots", toml_edit::Value::from(i64::from(resize.slots))),
1372    ];
1373    let cpus = resize
1374        .cpus
1375        .map(|cpus| ("cpus", toml_edit::Value::from(i64::from(cpus))));
1376    for (key, mut value) in values.into_iter().chain(cpus) {
1377        // The comment after the old value stays after the new one.
1378        if let Some(old) = table.get(key).and_then(toml_edit::Item::as_value) {
1379            *value.decor_mut() = old.decor().clone();
1380        }
1381        table.insert(key, toml_edit::Item::Value(value));
1382    }
1383    Ok(document.to_string())
1384}
1385
1386/// What `GET /v1/config` shows in place of a notification target's URL and header values (ADR 0051).
1387pub const SECRET_SHOWN: &str = "<secret>";
1388
1389/// `text` with each `[[notify.targets]]` entry's `url` and header values shown as [`SECRET_SHOWN`], everything else
1390/// as written: what `GET /v1/config` answers, so no client, the dashboard's browsers included, holds them. `None`
1391/// when `text` is not TOML.
1392pub fn mask_secrets(text: &str) -> Option<String> {
1393    let mut document: toml_edit::DocumentMut = text.parse().ok()?;
1394    let mut masked = false;
1395    for target in notify_targets(&mut document) {
1396        for (_, value) in secrets_of(target) {
1397            replace_value(value, toml_edit::Value::from(SECRET_SHOWN));
1398            masked = true;
1399        }
1400    }
1401    Some(if masked {
1402        document.to_string()
1403    } else {
1404        text.to_owned()
1405    })
1406}
1407
1408/// `text` with each [`SECRET_SHOWN`] in a target's `url` or headers replaced by the value `stored` has for the target
1409/// of the same name, everything else as written: what `PUT /v1/config` and `POST /v1/config/check` read, so a config
1410/// shown masked saves with its secrets. A value with nothing to keep stays [`SECRET_SHOWN`], which `Config::validate`
1411/// refuses. A text that does not parse as a config is answered as sent: a TOML error quotes the line it is on, and
1412/// that line must hold the mask, not the value put back. One that parses still does with the values put back, as they
1413/// are strings where strings were, and `Config::validate` names a target's field, never its value.
1414pub fn keep_secrets(text: &str, stored: &str) -> String {
1415    if !text.contains(SECRET_SHOWN) || Config::parse(text).is_err() {
1416        return text.to_owned();
1417    }
1418    let (Ok(mut document), Ok(mut old)) = (
1419        text.parse::<toml_edit::DocumentMut>(),
1420        stored.parse::<toml_edit::DocumentMut>(),
1421    ) else {
1422        return text.to_owned();
1423    };
1424    let mut kept: BTreeMap<String, Vec<(SecretField, toml_edit::Value)>> = BTreeMap::new();
1425    for target in notify_targets(&mut old) {
1426        let Some(name) = target_name(target) else {
1427            continue;
1428        };
1429        let secrets = secrets_of(target)
1430            .into_iter()
1431            .map(|(field, value)| (field, value.clone()))
1432            .collect();
1433        kept.insert(name, secrets);
1434    }
1435    for target in notify_targets(&mut document) {
1436        let Some(secrets) = target_name(target).and_then(|name| kept.get(&name)) else {
1437            continue;
1438        };
1439        for (field, value) in secrets_of(target) {
1440            if value.as_str() != Some(SECRET_SHOWN) {
1441                continue;
1442            }
1443            if let Some((_, old)) = secrets.iter().find(|(kept, _)| *kept == field) {
1444                replace_value(value, old.clone());
1445            }
1446        }
1447    }
1448    document.to_string()
1449}
1450
1451/// Where a target keeps a secret.
1452#[derive(Debug, PartialEq, Eq)]
1453enum SecretField {
1454    Url,
1455    Header(String),
1456}
1457
1458/// The tables of `[[notify.targets]]`, or of an inline `targets = [...]`.
1459fn notify_targets(document: &mut toml_edit::DocumentMut) -> Vec<&mut dyn toml_edit::TableLike> {
1460    match document
1461        .get_mut("notify")
1462        .and_then(|notify| notify.get_mut("targets"))
1463    {
1464        Some(toml_edit::Item::ArrayOfTables(tables)) => tables
1465            .iter_mut()
1466            .map(|table| table as &mut dyn toml_edit::TableLike)
1467            .collect(),
1468        Some(toml_edit::Item::Value(toml_edit::Value::Array(array))) => array
1469            .iter_mut()
1470            .filter_map(toml_edit::Value::as_inline_table_mut)
1471            .map(|table| table as &mut dyn toml_edit::TableLike)
1472            .collect(),
1473        _ => Vec::new(),
1474    }
1475}
1476
1477fn target_name(target: &dyn toml_edit::TableLike) -> Option<String> {
1478    target
1479        .get("name")
1480        .and_then(toml_edit::Item::as_str)
1481        .map(str::to_owned)
1482}
1483
1484/// A target's `url` and header values.
1485fn secrets_of(target: &mut dyn toml_edit::TableLike) -> Vec<(SecretField, &mut toml_edit::Value)> {
1486    let mut found = Vec::new();
1487    for (key, item) in target.iter_mut() {
1488        match key.get() {
1489            "url" => found.extend(item.as_value_mut().map(|value| (SecretField::Url, value))),
1490            "headers" => {
1491                let Some(headers) = item.as_table_like_mut() else {
1492                    continue;
1493                };
1494                for (name, item) in headers.iter_mut() {
1495                    let field = SecretField::Header(name.get().to_owned());
1496                    found.extend(item.as_value_mut().map(|value| (field, value)));
1497                }
1498            }
1499            _ => {}
1500        }
1501    }
1502    found
1503}
1504
1505/// Puts `new` in place of `value`, with the spaces and comment around it as they were.
1506fn replace_value(value: &mut toml_edit::Value, mut new: toml_edit::Value) {
1507    *new.decor_mut() = value.decor().clone();
1508    *value = new;
1509}
1510
1511/// A config text's revision: a hash of its bytes (64-bit FNV-1a, in hex), the same in every runnerd. `PUT
1512/// /v1/config?revision=` refuses a text sent against another revision than the daemon's.
1513pub fn revision(text: &str) -> String {
1514    let hash = text.bytes().fold(0xcbf2_9ce4_8422_2325_u64, |hash, byte| {
1515        (hash ^ u64::from(byte)).wrapping_mul(0x0100_0000_01b3)
1516    });
1517    format!("{hash:016x}")
1518}
1519
1520/// Reads a slice's `MemoryMax` from systemd; `None` when it is unlimited.
1521pub fn slice_memory_max(slice: &str) -> std::io::Result<Option<ByteSize>> {
1522    let output = Command::new("systemctl")
1523        .args(["show", slice, "--property=MemoryMax", "--value"])
1524        .output()?;
1525    if !output.status.success() {
1526        return Err(std::io::Error::other(format!(
1527            "systemctl show {slice} failed: {}",
1528            String::from_utf8_lossy(&output.stderr).trim()
1529        )));
1530    }
1531    let value = String::from_utf8_lossy(&output.stdout);
1532    parse_memory_max(value.trim()).ok_or_else(|| {
1533        std::io::Error::other(format!(
1534            "systemctl show {slice}: unexpected MemoryMax {value:?}"
1535        ))
1536    })
1537}
1538
1539/// The CPU budget, in millicores, of `millicores` of CPUs overcommitted `overcommit` times.
1540pub fn cpu_budget(millicores: u64, overcommit: f64) -> u64 {
1541    #[expect(
1542        clippy::cast_precision_loss,
1543        clippy::cast_possible_truncation,
1544        clippy::cast_sign_loss,
1545        reason = "a host's millicores are far below 2^52, and the config check holds the factor above 0 and at \
1546                  most 16"
1547    )]
1548    let budget = (millicores as f64 * overcommit.max(0.0)).round() as u64;
1549    budget
1550}
1551
1552/// Reads the CPUs a slice's processes may use from systemd, in millicores: its `EffectiveCPUs` (or `AllowedCPUs`,
1553/// else the host's CPUs), less when its `CPUQuota` is lower.
1554pub fn slice_cpus(slice: &str) -> std::io::Result<u64> {
1555    let output = Command::new("systemctl")
1556        .args([
1557            "show",
1558            slice,
1559            "--property=EffectiveCPUs",
1560            "--property=AllowedCPUs",
1561            "--property=CPUQuotaPerSecUSec",
1562        ])
1563        .output()?;
1564    if !output.status.success() {
1565        return Err(std::io::Error::other(format!(
1566            "systemctl show {slice} failed: {}",
1567            String::from_utf8_lossy(&output.stderr).trim()
1568        )));
1569    }
1570    let host = std::thread::available_parallelism().map_or(1, std::num::NonZero::get) as u64;
1571    parse_slice_cpus(&String::from_utf8_lossy(&output.stdout), host).ok_or_else(|| {
1572        std::io::Error::other(format!(
1573            "systemctl show {slice}: unexpected CPU properties {:?}",
1574            String::from_utf8_lossy(&output.stdout).trim()
1575        ))
1576    })
1577}
1578
1579/// The CPUs, in millicores, that `systemctl show`'s `EffectiveCPUs`, `AllowedCPUs` and `CPUQuotaPerSecUSec` lines
1580/// leave a slice on a host with `host` CPUs; `None` when a value does not parse.
1581pub fn parse_slice_cpus(properties: &str, host: u64) -> Option<u64> {
1582    let value = |name: &str| {
1583        properties.lines().find_map(|line| {
1584            line.strip_prefix(name)
1585                .and_then(|rest| rest.strip_prefix('='))
1586                .map(str::trim)
1587        })
1588    };
1589    let set = [value("EffectiveCPUs"), value("AllowedCPUs")]
1590        .into_iter()
1591        .flatten()
1592        .find(|cpus| !cpus.is_empty());
1593    let cpus = match set {
1594        Some(set) => cpu_count(set)?,
1595        None => host,
1596    };
1597    let quota = match value("CPUQuotaPerSecUSec") {
1598        None | Some("" | "infinity") => u64::MAX,
1599        Some(quota) => quota_millicores(quota)?,
1600    };
1601    Some(cpus.saturating_mul(1000).min(quota))
1602}
1603
1604/// How many CPUs a set names, as systemd prints it (`0-3 8-11`) or as it is written (`0-3,8,10-11`).
1605fn cpu_count(set: &str) -> Option<u64> {
1606    set.split([',', ' '])
1607        .filter(|range| !range.is_empty())
1608        .try_fold(0u64, |count, range| {
1609            let (first, last) = range.split_once('-').unwrap_or((range, range));
1610            let (first, last): (u64, u64) = (first.parse().ok()?, last.parse().ok()?);
1611            Some(count + last.checked_sub(first)? + 1)
1612        })
1613}
1614
1615/// A `CPUQuotaPerSecUSec` as systemd prints a timespan, such as `800ms`, `2.500000s` or `1min 4s`, as the
1616/// millicores it allows: a second of CPU time each second is 1000.
1617fn quota_millicores(quota: &str) -> Option<u64> {
1618    // Each unit's length in milliseconds, as systemd's `format_timespan` names them.
1619    const UNITS: [(&str, f64); 10] = [
1620        ("y", 31_557_600_000.0),
1621        ("M", 2_629_800_000.0),
1622        ("w", 604_800_000.0),
1623        ("d", 86_400_000.0),
1624        ("h", 3_600_000.0),
1625        ("min", 60_000.0),
1626        ("s", 1000.0),
1627        ("ms", 1.0),
1628        ("us", 0.001),
1629        ("µs", 0.001),
1630    ];
1631    let millicores = quota.split_whitespace().try_fold(0.0, |total, part| {
1632        let at = part.find(|c: char| !(c.is_ascii_digit() || c == '.'))?;
1633        let (number, unit) = part.split_at(at);
1634        let per_unit = UNITS
1635            .iter()
1636            .find_map(|(name, per_unit)| (*name == unit).then_some(*per_unit))?;
1637        Some(total + number.parse::<f64>().ok()? * per_unit)
1638    })?;
1639    #[allow(
1640        clippy::cast_possible_truncation,
1641        clippy::cast_sign_loss,
1642        reason = "a quota is a few thousand millicores per CPU, checked finite and not negative first"
1643    )]
1644    (!quota.trim().is_empty() && millicores.is_finite() && millicores >= 0.0)
1645        .then(|| millicores.round() as u64)
1646}
1647
1648/// Parses a `MemoryMax` value as `systemctl show` prints it: bytes, or `infinity`.
1649pub fn parse_memory_max(value: &str) -> Option<Option<ByteSize>> {
1650    if value == "infinity" {
1651        Some(None)
1652    } else {
1653        value.parse().ok().map(|bytes| Some(ByteSize(bytes)))
1654    }
1655}
1656
1657/// A size in bytes, written with a binary unit as Docker and systemd do: `512M`, `9G`, `32.5G`.
1658#[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord)]
1659pub struct ByteSize(pub u64);
1660
1661const UNITS: [(char, u64); 4] = [
1662    ('T', 1 << 40),
1663    ('G', 1 << 30),
1664    ('M', 1 << 20),
1665    ('K', 1 << 10),
1666];
1667
1668impl FromStr for ByteSize {
1669    type Err = String;
1670
1671    #[allow(
1672        clippy::cast_precision_loss,
1673        clippy::cast_possible_truncation,
1674        clippy::cast_sign_loss,
1675        reason = "sizes in a config are far below 2^52 bytes, and the result is checked against u64's range first"
1676    )]
1677    fn from_str(s: &str) -> Result<Self, Self::Err> {
1678        let invalid = || {
1679            format!(
1680                "invalid size {s:?}, expected a number and a unit (K, M, G or T), such as \"9G\""
1681            )
1682        };
1683        let unit = s.chars().last().ok_or_else(invalid)?.to_ascii_uppercase();
1684        let (_, factor) = UNITS.iter().find(|(u, _)| *u == unit).ok_or_else(invalid)?;
1685        let number = &s[..s.len() - 1];
1686        if number.is_empty() || !number.chars().all(|c| c.is_ascii_digit() || c == '.') {
1687            return Err(invalid());
1688        }
1689        let number: f64 = number.parse().map_err(|_| invalid())?;
1690        let bytes = (number * *factor as f64).round();
1691        // `u64::MAX as f64` is 2^64, the first value out of range.
1692        if bytes >= u64::MAX as f64 {
1693            return Err(format!("size {s:?} is too large"));
1694        }
1695        Ok(ByteSize(bytes as u64))
1696    }
1697}
1698
1699impl fmt::Display for ByteSize {
1700    #[allow(
1701        clippy::cast_precision_loss,
1702        reason = "shown with two decimals, far coarser than f64's precision"
1703    )]
1704    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1705        let Some((unit, factor)) = UNITS.iter().find(|(_, factor)| self.0 >= *factor) else {
1706            return write!(f, "{}B", self.0);
1707        };
1708        let number = format!("{:.2}", self.0 as f64 / *factor as f64);
1709        let number = number.trim_end_matches('0').trim_end_matches('.');
1710        write!(f, "{number}{unit}")
1711    }
1712}
1713
1714impl ByteSize {
1715    /// The size rounded to the nearest, for a label too narrow for its `Display`: two significant digits below 10
1716    /// of its unit, whole units above, and the next unit up once it rounds to 1024: `3.3G` for 3.25G, `12G`,
1717    /// `512M`, and `1G` for 1023.5M. Every client's budget bar labels a block with it.
1718    #[allow(
1719        clippy::cast_precision_loss,
1720        reason = "shown with one decimal at most, far coarser than f64's precision"
1721    )]
1722    pub fn short(self) -> String {
1723        let Some(index) = UNITS.iter().position(|(_, factor)| self.0 >= *factor) else {
1724            return format!("{}B", self.0);
1725        };
1726        let (mut unit, factor) = UNITS[index];
1727        let mut value = self.0 as f64 / factor as f64;
1728        if value.round() >= 1024.0
1729            && let Some(larger) = index.checked_sub(1)
1730        {
1731            unit = UNITS[larger].0;
1732            value /= 1024.0;
1733        }
1734        // Halves round up, as `f64::round` does and formatting, which rounds them to even, does not.
1735        let tenths = (value * 10.0).round() / 10.0;
1736        if tenths < 10.0 {
1737            let number = format!("{tenths:.1}");
1738            format!("{}{unit}", number.trim_end_matches(".0"))
1739        } else {
1740            format!("{:.0}{unit}", value.round())
1741        }
1742    }
1743}
1744
1745impl<'de> Deserialize<'de> for ByteSize {
1746    fn deserialize<D: Deserializer<'de>>(deserializer: D) -> Result<Self, D::Error> {
1747        String::deserialize(deserializer)?
1748            .parse()
1749            .map_err(de::Error::custom)
1750    }
1751}
1752
1753/// Parses a duration as Go and the registry write it, `30s`, `90m`, `168h`, or in days: `7d`.
1754pub fn parse_duration(s: &str) -> Result<Duration, String> {
1755    let invalid = || {
1756        format!(
1757            "invalid duration {s:?}, expected a number and s, m, h or d, such as \"168h\" or \"7d\""
1758        )
1759    };
1760    let (number, factor) = match s.chars().last() {
1761        Some('s') => (&s[..s.len() - 1], 1),
1762        Some('m') => (&s[..s.len() - 1], 60),
1763        Some('h') => (&s[..s.len() - 1], 3600),
1764        Some('d') => (&s[..s.len() - 1], 86400),
1765        _ => return Err(invalid()),
1766    };
1767    let number: u64 = number.parse().map_err(|_| invalid())?;
1768    number
1769        .checked_mul(factor)
1770        .map(Duration::from_secs)
1771        .ok_or_else(invalid)
1772}
1773
1774/// Parses an age as the clients' `--since` and `--older-than` take it: `30m`, `12h`, `7d`, `2w`.
1775pub fn parse_age(text: &str) -> Result<Duration, String> {
1776    let error = || "expected a number and s, m, h, d or w, such as 12h".to_owned();
1777    let split = text.find(|c: char| !c.is_ascii_digit()).ok_or_else(error)?;
1778    let (number, unit) = text.split_at(split);
1779    let number: u64 = number.parse().map_err(|_| error())?;
1780    let unit = match unit {
1781        "s" => 1,
1782        "m" => 60,
1783        "h" => 3600,
1784        "d" => 86_400,
1785        "w" => 7 * 86_400,
1786        _ => return Err(error()),
1787    };
1788    Ok(Duration::from_secs(number.saturating_mul(unit)))
1789}
1790
1791fn duration<'de, D: Deserializer<'de>>(deserializer: D) -> Result<Duration, D::Error> {
1792    parse_duration(&String::deserialize(deserializer)?).map_err(de::Error::custom)
1793}
1794
1795#[cfg(test)]
1796mod tests;