crates/hub/src/db/runners.rs at v0.5.0 · andymccall/oxygenhub

◆ v0.5.0 oxygenhub / crates / hub / src / db / runners.rs

15.7 KiB · Rust · last changed Alerts: a failed run tells its pusher and watchers; an erroring runner is drained 21 days ago source blame history raw

//! OxygenFlow runners (hub spec §13.2): who may claim jobs, what they are,
//! whether they are taking work, and what they have run.

use super::{now_ms, Pool};
use sqlx::FromRow;

#[derive(Clone, Debug, Default, FromRow)]
pub struct Runner {
    pub id: i64,
    pub name: String,
    pub token_hash: String,
    pub source: String,
    pub key: String,
    pub key_name: String,
    pub arch: String,
    pub os: String,
    pub version: String,
    pub labels: String,
    pub max_jobs: i64,
    pub memory_mib: Option<i64>,
    pub cpus: Option<i64>,
    pub state: String,
    pub created_ms: i64,
    pub created_by: String,
    pub last_seen_ms: Option<i64>,
    /// The owner whose repositories it serves; empty for the whole site.
    pub scope_owner: String,
    /// The one repository it serves, when it serves one.
    pub scope_repo_id: Option<i64>,
    /// Labels an administrator gave it, beside those it reports (#98).
    pub admin_labels: String,
    /// A time limit for its jobs in minutes, below the site's, when set.
    pub timeout_minutes: Option<i64>,
    /// It fetches images it lacks from the hub (oxygenflow#39).
    pub fetches_images: bool,
    /// What it measured at its last claim (oxygenflow#40): free disk (MiB),
    /// the 1-minute load ×100, its speed score and class, its kernel.
    pub free_disk_mib: Option<i64>,
    pub load_x100: Option<i64>,
    pub speed: Option<i64>,
    pub speed_class: Option<i64>,
    pub kernel: String,
    /// The tiers it offered at its last claim, comma-separated.
    pub tiers: String,
    /// The images it held then (a JSON array); empty when it did not say.
    pub images: String,
    /// BLAKE3 of what it offers; its declines hold until this changes.
    pub profile: String,
}

/// Which repositories' jobs a runner may take (oxygenhub#41).
#[derive(Clone, Debug, Default, PartialEq, Eq)]
pub struct Scope {
    /// Empty: any owner.
    pub owner: String,
    /// `None`: any of the owner's repositories.
    pub repo_id: Option<i64>,
}

impl Scope {
    pub fn is_site(&self) -> bool {
        self.owner.is_empty() && self.repo_id.is_none()
    }
}

/// A runner that has not polled for this long is shown offline.
pub const ONLINE_MS: i64 = 90_000;

impl Runner {
    pub fn online(&self) -> bool {
        self.last_seen_ms.is_some_and(|t| now_ms() - t < ONLINE_MS)
    }

    /// `online`, `offline`, `draining` or `disabled`.
    pub fn status(&self) -> &'static str {
        match self.state.as_str() {
            "disabled" => "disabled",
            "draining" => "draining",
            _ if self.online() => "online",
            _ => "offline",
        }
    }

    /// Every label it carries: what it reports, and what an administrator
    /// gave it.
    pub fn label_list(&self) -> Vec<&str> {
        let mut out: Vec<&str> = self.labels.split(',').filter(|l| !l.is_empty()).collect();
        for l in self.admin_label_list() {
            if !out.contains(&l) {
                out.push(l);
            }
        }
        out
    }

    pub fn admin_label_list(&self) -> Vec<&str> {
        self.admin_labels.split(',').filter(|l| !l.is_empty()).collect()
    }

    /// The labels it reported itself, as the admin page distinguishes them.
    pub fn reported_label_list(&self) -> Vec<&str> {
        self.labels.split(',').filter(|l| !l.is_empty()).collect()
    }

    pub fn scope(&self) -> Scope {
        Scope { owner: self.scope_owner.clone(), repo_id: self.scope_repo_id }
    }
}

pub async fn by_token(pool: &Pool, token_hash: &str) -> sqlx::Result<Option<Runner>> {
    sqlx::query_as("SELECT * FROM flow_runners WHERE token_hash = ?").bind(token_hash).fetch_optional(pool).await
}

pub async fn get(pool: &Pool, id: i64) -> sqlx::Result<Option<Runner>> {
    sqlx::query_as("SELECT * FROM flow_runners WHERE id = ?").bind(id).fetch_optional(pool).await
}

pub async fn list(pool: &Pool) -> sqlx::Result<Vec<Runner>> {
    sqlx::query_as("SELECT * FROM flow_runners ORDER BY name, id").fetch_all(pool).await
}

/// The row of a runner configured with a hub.toml token, made on first sight.
pub async fn configured(pool: &Pool, token_hash: &str, name: &str) -> sqlx::Result<Runner> {
    sqlx::query("INSERT OR IGNORE INTO flow_runners (name, token_hash, source, created_ms) VALUES (?, ?, 'config', ?)")
        .bind(name)
        .bind(token_hash)
        .bind(now_ms())
        .execute(pool)
        .await?;
    Ok(by_token(pool, token_hash).await?.expect("just made"))
}

/// What a claim says about the runner, and that it was seen now. A
/// configured runner takes the name it claims with.
/// The administrator's settings for a runner (#98).
pub async fn set_limits(pool: &Pool, id: i64, admin_labels: &str, timeout_minutes: Option<i64>) -> sqlx::Result<()> {
    sqlx::query("UPDATE flow_runners SET admin_labels = ?, timeout_minutes = ? WHERE id = ?")
        .bind(admin_labels)
        .bind(timeout_minutes)
        .bind(id)
        .execute(pool)
        .await?;
    Ok(())
}

pub async fn seen(pool: &Pool, id: i64, name: Option<&str>, info: Option<&flow_core::api::RunnerInfo>) -> sqlx::Result<()> {
    let now = now_ms();
    match info {
        Some(i) => {
            sqlx::query(
                "UPDATE flow_runners SET arch = ?, os = ?, version = ?, labels = ?, max_jobs = ?, memory_mib = ?, cpus = ?, last_seen_ms = ? WHERE id = ?",
            )
            .bind(&i.arch)
            .bind(&i.os)
            .bind(&i.version)
            .bind(i.labels.join(","))
            .bind(i.max_jobs as i64)
            .bind(i.memory_mib.map(i64::from))
            .bind(i.cpus.map(i64::from))
            .bind(now)
            .bind(id)
            .execute(pool)
            .await?;
        }
        None => {
            sqlx::query("UPDATE flow_runners SET last_seen_ms = ? WHERE id = ?").bind(now).bind(id).execute(pool).await?;
        }
    }
    if let Some(n) = name {
        sqlx::query("UPDATE flow_runners SET name = ? WHERE id = ? AND source = 'config'").bind(n).bind(id).execute(pool).await?;
    }
    Ok(())
}

/// Keep what a runner measured about itself at a claim (oxygenflow#40).
pub async fn set_facts(pool: &Pool, id: i64, f: &flow_core::api::Facts) -> sqlx::Result<()> {
    sqlx::query("UPDATE flow_runners SET free_disk_mib = ?, load_x100 = ?, speed = ?, speed_class = ?, kernel = ? WHERE id = ?")
        .bind(f.free_disk_mib.map(|v| v.min(i64::MAX as u64) as i64))
        .bind(f.load_x100.map(i64::from))
        .bind(f.speed.map(i64::from))
        .bind(f.speed_class.map(i64::from))
        .bind(f.kernel.clone().unwrap_or_default())
        .bind(id)
        .execute(pool)
        .await?;
    Ok(())
}

impl Runner {
    /// Its facts as words for the Runners page: "speed 100 (class 3) · load
    /// 1.12 · 41 GiB free"; empty when it has sent none.
    pub fn facts_line(&self) -> String {
        let mut out = Vec::new();
        if let Some(s) = self.speed {
            out.push(format!("speed {s} (class {})", self.speed_class.unwrap_or(0)));
        }
        if let Some(l) = self.load_x100 {
            out.push(format!("load {:.2}", l as f64 / 100.0));
        }
        if let Some(d) = self.free_disk_mib {
            out.push(if d >= 1024 { format!("{} GiB free", d / 1024) } else { format!("{d} MiB free") });
        }
        out.join(" · ")
    }
}

/// What a runner offers at a claim (OxygenFlow spec §11.1): its tiers and
/// images, and the profile hash over those, its architecture, labels and
/// capacity.
pub async fn set_offer(
    pool: &Pool,
    id: i64,
    tiers: &[String],
    images: Option<&[String]>,
    fetches_images: bool,
    profile: &str,
) -> sqlx::Result<()> {
    let images = images.map(|i| serde_json::to_string(i).unwrap_or_default()).unwrap_or_default();
    sqlx::query("UPDATE flow_runners SET tiers = ?, images = ?, fetches_images = ?, profile = ? WHERE id = ?")
        .bind(tiers.join(","))
        .bind(images)
        .bind(fetches_images)
        .bind(profile)
        .bind(id)
        .execute(pool)
        .await?;
    Ok(())
}

impl Runner {
    pub fn tier_list(&self) -> Vec<&str> {
        self.tiers.split(',').filter(|t| !t.is_empty()).collect()
    }

    /// The images it holds, `None` when it does not say.
    pub fn image_list(&self) -> Option<Vec<String>> {
        if self.images.is_empty() {
            None
        } else {
            serde_json::from_str(&self.images).ok()
        }
    }
}

pub async fn set_state(pool: &Pool, id: i64, state: &str) -> sqlx::Result<()> {
    sqlx::query("UPDATE flow_runners SET state = ?, state_ms = ? WHERE id = ?")
        .bind(state)
        .bind(super::now_ms())
        .bind(id)
        .execute(pool)
        .await?;
    Ok(())
}

/// Whether runner `id`'s last `n` jobs since its state last changed all
/// ended in an infrastructure error (oxygenflow#30), with the last one's.
pub async fn erroring(pool: &Pool, id: i64, n: u32) -> sqlx::Result<Option<String>> {
    if n == 0 {
        return Ok(None);
    }
    let last: Vec<(String, String)> = sqlx::query_as(
        "SELECT j.status, j.error FROM flow_jobs j JOIN flow_runners r ON r.id = j.runner_id
         WHERE j.runner_id = ? AND j.finished_ms IS NOT NULL AND j.finished_ms >= r.state_ms
         ORDER BY j.finished_ms DESC, j.id DESC LIMIT ?",
    )
    .bind(id)
    .bind(n as i64)
    .fetch_all(pool)
    .await?;
    Ok((last.len() == n as usize && last.iter().all(|(s, _)| s == "error")).then(|| last[0].1.clone()))
}

/// Forget a runner: its credential stops working. Its jobs keep its name.
pub async fn remove(pool: &Pool, id: i64) -> sqlx::Result<()> {
    sqlx::query("DELETE FROM flow_runners WHERE id = ?").bind(id).execute(pool).await?;
    Ok(())
}

/// The key a runner signs receipts with: the one bound, or `key` (named
/// `name`) when none is yet.
pub async fn bind_key(pool: &Pool, id: i64, key: &str, name: &str) -> sqlx::Result<String> {
    sqlx::query("UPDATE flow_runners SET key = ?, key_name = ? WHERE id = ? AND key = ''")
        .bind(key)
        .bind(name)
        .bind(id)
        .execute(pool)
        .await?;
    sqlx::query_scalar("SELECT key FROM flow_runners WHERE id = ?").bind(id).fetch_one(pool).await
}

/// Keys that runners of this hub sign with.
pub async fn keys(pool: &Pool) -> sqlx::Result<Vec<String>> {
    sqlx::query_scalar("SELECT key FROM flow_runners WHERE key != ''").fetch_all(pool).await
}

/// A one-time registration token, valid for a day, for a runner of `scope`.
pub async fn add_registration(pool: &Pool, token_hash: &str, by: &str, scope: &Scope) -> sqlx::Result<()> {
    let now = now_ms();
    sqlx::query(
        "INSERT INTO flow_runner_registrations (token_hash, created_by, created_ms, expires_ms, scope_owner, scope_repo_id) VALUES (?, ?, ?, ?, ?, ?)",
    )
    .bind(token_hash)
    .bind(by)
    .bind(now)
    .bind(now + 24 * 3_600_000)
    .bind(&scope.owner)
    .bind(scope.repo_id)
    .execute(pool)
    .await?;
    Ok(())
}

/// The runners that may take a repository's jobs: the site's, its owner's
/// and its own (oxygenhub#41).
pub async fn serving(pool: &Pool, owner: &str, repo_id: i64) -> sqlx::Result<Vec<Runner>> {
    sqlx::query_as(
        "SELECT * FROM flow_runners WHERE (scope_owner = '' AND scope_repo_id IS NULL) \
         OR (scope_owner = ? AND scope_repo_id IS NULL) OR scope_repo_id = ? ORDER BY name",
    )
    .bind(owner)
    .bind(repo_id)
    .fetch_all(pool)
    .await
}

/// The runners that serve an owner: the site's, and those scoped to the
/// owner or one of its repositories.
pub async fn serving_owner(pool: &Pool, owner: &str) -> sqlx::Result<Vec<Runner>> {
    sqlx::query_as("SELECT * FROM flow_runners WHERE (scope_owner = '' AND scope_repo_id IS NULL) OR scope_owner = ? ORDER BY name")
        .bind(owner)
        .fetch_all(pool)
        .await
}

/// Change the repositories a runner serves.
pub async fn set_scope(pool: &Pool, id: i64, scope: &Scope) -> sqlx::Result<()> {
    sqlx::query("UPDATE flow_runners SET scope_owner = ?, scope_repo_id = ? WHERE id = ?")
        .bind(&scope.owner)
        .bind(scope.repo_id)
        .bind(id)
        .execute(pool)
        .await?;
    Ok(())
}

/// Registration tokens not used yet and not expired: (created by, created, expires).
pub async fn pending_registrations(pool: &Pool) -> sqlx::Result<Vec<(String, i64, i64)>> {
    sqlx::query_as("SELECT created_by, created_ms, expires_ms FROM flow_runner_registrations WHERE used_ms IS NULL AND expires_ms > ? ORDER BY created_ms DESC")
        .bind(now_ms())
        .fetch_all(pool)
        .await
}

/// Use registration token `token_hash` to register a runner: `None` when
/// the token is unknown, used or expired.
#[allow(clippy::too_many_arguments)]
pub async fn register(
    pool: &Pool,
    registration: &str,
    name: &str,
    token_hash: &str,
    key: &str,
    key_name: &str,
    by_default: &str,
) -> sqlx::Result<Option<Runner>> {
    let mut tx = super::begin_write(pool).await?;
    let now = now_ms();
    let found: Option<(String, String, Option<i64>)> = sqlx::query_as(
        "SELECT created_by, scope_owner, scope_repo_id FROM flow_runner_registrations WHERE token_hash = ? AND used_ms IS NULL AND expires_ms > ?",
    )
    .bind(registration)
    .bind(now)
    .fetch_optional(&mut *tx)
    .await?;
    let Some((who, scope_owner, scope_repo_id)) = found else { return Ok(None) };
    let id: i64 = sqlx::query_scalar(
        "INSERT INTO flow_runners (name, token_hash, source, key, key_name, created_ms, created_by, scope_owner, scope_repo_id) VALUES (?, ?, 'registered', ?, ?, ?, ?, ?, ?) RETURNING id",
    )
    .bind(name)
    .bind(token_hash)
    .bind(key)
    .bind(key_name)
    .bind(now)
    .bind(if who.is_empty() { by_default } else { &who })
    .bind(scope_owner)
    .bind(scope_repo_id)
    .fetch_one(&mut *tx)
    .await?;
    sqlx::query("UPDATE flow_runner_registrations SET used_ms = ?, runner_id = ? WHERE token_hash = ?")
        .bind(now)
        .bind(id)
        .bind(registration)
        .execute(&mut *tx)
        .await?;
    tx.commit().await?;
    get(pool, id).await
}

/// A job a runner ran, for its history.
#[derive(Clone, Debug, FromRow)]
pub struct RunnerJob {
    pub owner: String,
    pub repo: String,
    pub run_number: i64,
    pub job: String,
    pub status: String,
    pub started_ms: Option<i64>,
    pub finished_ms: Option<i64>,
    pub receipt_attest: String,
}

/// Jobs runner `id` ran or is running, newest first.
pub async fn history(pool: &Pool, id: i64, limit: i64, before: Option<i64>) -> sqlx::Result<Vec<RunnerJob>> {
    sqlx::query_as(
        "SELECT p.owner AS owner, p.name AS repo, r.number AS run_number, j.name AS job, j.status AS status, j.started_ms AS started_ms, \
         j.finished_ms AS finished_ms, j.receipt_attest AS receipt_attest \
         FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id JOIN repositories p ON p.id = r.repo_id \
         WHERE j.runner_id = ? AND (? IS NULL OR j.started_ms < ?) ORDER BY j.started_ms DESC LIMIT ?",
    )
    .bind(id)
    .bind(before)
    .bind(before)
    .bind(limit)
    .fetch_all(pool)
    .await
}

/// (jobs run, passed, busy milliseconds) of runner `id`.
pub async fn totals(pool: &Pool, id: i64) -> sqlx::Result<(i64, i64, i64)> {
    sqlx::query_as(
        "SELECT COUNT(*), COALESCE(SUM(status = 'passed'), 0), COALESCE(SUM(finished_ms - started_ms), 0) \
         FROM flow_jobs WHERE runner_id = ? AND finished_ms IS NOT NULL",
    )
    .bind(id)
    .fetch_one(pool)
    .await
}

/// Jobs running on each runner now, by runner id.
pub async fn running(pool: &Pool) -> sqlx::Result<Vec<(i64, i64)>> {
    sqlx::query_as("SELECT runner_id, COUNT(*) FROM flow_jobs WHERE status = 'running' AND runner_id IS NOT NULL GROUP BY runner_id")
        .fetch_all(pool)
        .await
}

pub async fn set_job_runner(pool: &Pool, job_id: i64, runner_id: i64) -> sqlx::Result<()> {
    sqlx::query("UPDATE flow_jobs SET runner_id = ? WHERE id = ?").bind(runner_id).bind(job_id).execute(pool).await?;
    Ok(())
}

`/`	Search
`g` `c`	Code
`g` `h`	History
`g` `b`	Bookmarks
`g` `t`	Trust
`y`	Permalink to this state (files and trees)
`?`	This sheet

1	//! OxygenFlow runners (hub spec §13.2): who may claim jobs, what they are,
2	//! whether they are taking work, and what they have run.
3
4	use super::{now_ms, Pool};
5	use sqlx::FromRow;
6
7	#[derive(Clone, Debug, Default, FromRow)]
8	pub struct Runner {
9	pub id: i64,
10	pub name: String,
11	pub token_hash: String,
12	pub source: String,
13	pub key: String,
14	pub key_name: String,
15	pub arch: String,
16	pub os: String,
17	pub version: String,
18	pub labels: String,
19	pub max_jobs: i64,
20	pub memory_mib: Option<i64>,
21	pub cpus: Option<i64>,
22	pub state: String,
23	pub created_ms: i64,
24	pub created_by: String,
25	pub last_seen_ms: Option<i64>,
26	/// The owner whose repositories it serves; empty for the whole site.
27	pub scope_owner: String,
28	/// The one repository it serves, when it serves one.
29	pub scope_repo_id: Option<i64>,
30	/// Labels an administrator gave it, beside those it reports (#98).
31	pub admin_labels: String,
32	/// A time limit for its jobs in minutes, below the site's, when set.
33	pub timeout_minutes: Option<i64>,
34	/// It fetches images it lacks from the hub (oxygenflow#39).
35	pub fetches_images: bool,
36	/// What it measured at its last claim (oxygenflow#40): free disk (MiB),
37	/// the 1-minute load ×100, its speed score and class, its kernel.
38	pub free_disk_mib: Option<i64>,
39	pub load_x100: Option<i64>,
40	pub speed: Option<i64>,
41	pub speed_class: Option<i64>,
42	pub kernel: String,
43	/// The tiers it offered at its last claim, comma-separated.
44	pub tiers: String,
45	/// The images it held then (a JSON array); empty when it did not say.
46	pub images: String,
47	/// BLAKE3 of what it offers; its declines hold until this changes.
48	pub profile: String,
49	}
50
51	/// Which repositories' jobs a runner may take (oxygenhub#41).
52	#[derive(Clone, Debug, Default, PartialEq, Eq)]
53	pub struct Scope {
54	/// Empty: any owner.
55	pub owner: String,
56	/// `None`: any of the owner's repositories.
57	pub repo_id: Option<i64>,
58	}
59
60	impl Scope {
61	pub fn is_site(&self) -> bool {
62	self.owner.is_empty() && self.repo_id.is_none()
63	}
64	}
65
66	/// A runner that has not polled for this long is shown offline.
67	pub const ONLINE_MS: i64 = 90_000;
68
69	impl Runner {
70	pub fn online(&self) -> bool {
71	self.last_seen_ms.is_some_and(\|t\| now_ms() - t < ONLINE_MS)
72	}
73
74	/// `online`, `offline`, `draining` or `disabled`.
75	pub fn status(&self) -> &'static str {
76	match self.state.as_str() {
77	"disabled" => "disabled",
78	"draining" => "draining",
79	_ if self.online() => "online",
80	_ => "offline",
81	}
82	}
83
84	/// Every label it carries: what it reports, and what an administrator
85	/// gave it.
86	pub fn label_list(&self) -> Vec<&str> {
87	let mut out: Vec<&str> = self.labels.split(',').filter(\|l\| !l.is_empty()).collect();
88	for l in self.admin_label_list() {
89	if !out.contains(&l) {
90	out.push(l);
91	}
92	}
93	out
94	}
95
96	pub fn admin_label_list(&self) -> Vec<&str> {
97	self.admin_labels.split(',').filter(\|l\| !l.is_empty()).collect()
98	}
99
100	/// The labels it reported itself, as the admin page distinguishes them.
101	pub fn reported_label_list(&self) -> Vec<&str> {
102	self.labels.split(',').filter(\|l\| !l.is_empty()).collect()
103	}
104
105	pub fn scope(&self) -> Scope {
106	Scope { owner: self.scope_owner.clone(), repo_id: self.scope_repo_id }
107	}
108	}
109
110	pub async fn by_token(pool: &Pool, token_hash: &str) -> sqlx::Result<Option<Runner>> {
111	sqlx::query_as("SELECT * FROM flow_runners WHERE token_hash = ?").bind(token_hash).fetch_optional(pool).await
112	}
113
114	pub async fn get(pool: &Pool, id: i64) -> sqlx::Result<Option<Runner>> {
115	sqlx::query_as("SELECT * FROM flow_runners WHERE id = ?").bind(id).fetch_optional(pool).await
116	}
117
118	pub async fn list(pool: &Pool) -> sqlx::Result<Vec<Runner>> {
119	sqlx::query_as("SELECT * FROM flow_runners ORDER BY name, id").fetch_all(pool).await
120	}
121
122	/// The row of a runner configured with a hub.toml token, made on first sight.
123	pub async fn configured(pool: &Pool, token_hash: &str, name: &str) -> sqlx::Result<Runner> {
124	sqlx::query("INSERT OR IGNORE INTO flow_runners (name, token_hash, source, created_ms) VALUES (?, ?, 'config', ?)")
125	.bind(name)
126	.bind(token_hash)
127	.bind(now_ms())
128	.execute(pool)
129	.await?;
130	Ok(by_token(pool, token_hash).await?.expect("just made"))
131	}
132
133	/// What a claim says about the runner, and that it was seen now. A
134	/// configured runner takes the name it claims with.
135	/// The administrator's settings for a runner (#98).
136	pub async fn set_limits(pool: &Pool, id: i64, admin_labels: &str, timeout_minutes: Option<i64>) -> sqlx::Result<()> {
137	sqlx::query("UPDATE flow_runners SET admin_labels = ?, timeout_minutes = ? WHERE id = ?")
138	.bind(admin_labels)
139	.bind(timeout_minutes)
140	.bind(id)
141	.execute(pool)
142	.await?;
143	Ok(())
144	}
145
146	pub async fn seen(pool: &Pool, id: i64, name: Option<&str>, info: Option<&flow_core::api::RunnerInfo>) -> sqlx::Result<()> {
147	let now = now_ms();
148	match info {
149	Some(i) => {
150	sqlx::query(
151	"UPDATE flow_runners SET arch = ?, os = ?, version = ?, labels = ?, max_jobs = ?, memory_mib = ?, cpus = ?, last_seen_ms = ? WHERE id = ?",
152	)
153	.bind(&i.arch)
154	.bind(&i.os)
155	.bind(&i.version)
156	.bind(i.labels.join(","))
157	.bind(i.max_jobs as i64)
158	.bind(i.memory_mib.map(i64::from))
159	.bind(i.cpus.map(i64::from))
160	.bind(now)
161	.bind(id)
162	.execute(pool)
163	.await?;
164	}
165	None => {
166	sqlx::query("UPDATE flow_runners SET last_seen_ms = ? WHERE id = ?").bind(now).bind(id).execute(pool).await?;
167	}
168	}
169	if let Some(n) = name {
170	sqlx::query("UPDATE flow_runners SET name = ? WHERE id = ? AND source = 'config'").bind(n).bind(id).execute(pool).await?;
171	}
172	Ok(())
173	}
174
175	/// Keep what a runner measured about itself at a claim (oxygenflow#40).
176	pub async fn set_facts(pool: &Pool, id: i64, f: &flow_core::api::Facts) -> sqlx::Result<()> {
177	sqlx::query("UPDATE flow_runners SET free_disk_mib = ?, load_x100 = ?, speed = ?, speed_class = ?, kernel = ? WHERE id = ?")
178	.bind(f.free_disk_mib.map(\|v\| v.min(i64::MAX as u64) as i64))
179	.bind(f.load_x100.map(i64::from))
180	.bind(f.speed.map(i64::from))
181	.bind(f.speed_class.map(i64::from))
182	.bind(f.kernel.clone().unwrap_or_default())
183	.bind(id)
184	.execute(pool)
185	.await?;
186	Ok(())
187	}
188
189	impl Runner {
190	/// Its facts as words for the Runners page: "speed 100 (class 3) · load
191	/// 1.12 · 41 GiB free"; empty when it has sent none.
192	pub fn facts_line(&self) -> String {
193	let mut out = Vec::new();
194	if let Some(s) = self.speed {
195	out.push(format!("speed {s} (class {})", self.speed_class.unwrap_or(0)));
196	}
197	if let Some(l) = self.load_x100 {
198	out.push(format!("load {:.2}", l as f64 / 100.0));
199	}
200	if let Some(d) = self.free_disk_mib {
201	out.push(if d >= 1024 { format!("{} GiB free", d / 1024) } else { format!("{d} MiB free") });
202	}
203	out.join(" · ")
204	}
205	}
206
207	/// What a runner offers at a claim (OxygenFlow spec §11.1): its tiers and
208	/// images, and the profile hash over those, its architecture, labels and
209	/// capacity.
210	pub async fn set_offer(
211	pool: &Pool,
212	id: i64,
213	tiers: &[String],
214	images: Option<&[String]>,
215	fetches_images: bool,
216	profile: &str,
217	) -> sqlx::Result<()> {
218	let images = images.map(\|i\| serde_json::to_string(i).unwrap_or_default()).unwrap_or_default();
219	sqlx::query("UPDATE flow_runners SET tiers = ?, images = ?, fetches_images = ?, profile = ? WHERE id = ?")
220	.bind(tiers.join(","))
221	.bind(images)
222	.bind(fetches_images)
223	.bind(profile)
224	.bind(id)
225	.execute(pool)
226	.await?;
227	Ok(())
228	}
229
230	impl Runner {
231	pub fn tier_list(&self) -> Vec<&str> {
232	self.tiers.split(',').filter(\|t\| !t.is_empty()).collect()
233	}
234
235	/// The images it holds, `None` when it does not say.
236	pub fn image_list(&self) -> Option<Vec<String>> {
237	if self.images.is_empty() {
238	None
239	} else {
240	serde_json::from_str(&self.images).ok()
241	}
242	}
243	}
244
245	pub async fn set_state(pool: &Pool, id: i64, state: &str) -> sqlx::Result<()> {
246	sqlx::query("UPDATE flow_runners SET state = ?, state_ms = ? WHERE id = ?")
247	.bind(state)
248	.bind(super::now_ms())
249	.bind(id)
250	.execute(pool)
251	.await?;
252	Ok(())
253	}
254
255	/// Whether runner `id`'s last `n` jobs since its state last changed all
256	/// ended in an infrastructure error (oxygenflow#30), with the last one's.
257	pub async fn erroring(pool: &Pool, id: i64, n: u32) -> sqlx::Result<Option<String>> {
258	if n == 0 {
259	return Ok(None);
260	}
261	let last: Vec<(String, String)> = sqlx::query_as(
262	"SELECT j.status, j.error FROM flow_jobs j JOIN flow_runners r ON r.id = j.runner_id
263	WHERE j.runner_id = ? AND j.finished_ms IS NOT NULL AND j.finished_ms >= r.state_ms
264	ORDER BY j.finished_ms DESC, j.id DESC LIMIT ?",
265	)
266	.bind(id)
267	.bind(n as i64)
268	.fetch_all(pool)
269	.await?;
270	Ok((last.len() == n as usize && last.iter().all(\|(s, _)\| s == "error")).then(\|\| last[0].1.clone()))
271	}
272
273	/// Forget a runner: its credential stops working. Its jobs keep its name.
274	pub async fn remove(pool: &Pool, id: i64) -> sqlx::Result<()> {
275	sqlx::query("DELETE FROM flow_runners WHERE id = ?").bind(id).execute(pool).await?;
276	Ok(())
277	}
278
279	/// The key a runner signs receipts with: the one bound, or `key` (named
280	/// `name`) when none is yet.
281	pub async fn bind_key(pool: &Pool, id: i64, key: &str, name: &str) -> sqlx::Result<String> {
282	sqlx::query("UPDATE flow_runners SET key = ?, key_name = ? WHERE id = ? AND key = ''")
283	.bind(key)
284	.bind(name)
285	.bind(id)
286	.execute(pool)
287	.await?;
288	sqlx::query_scalar("SELECT key FROM flow_runners WHERE id = ?").bind(id).fetch_one(pool).await
289	}
290
291	/// Keys that runners of this hub sign with.
292	pub async fn keys(pool: &Pool) -> sqlx::Result<Vec<String>> {
293	sqlx::query_scalar("SELECT key FROM flow_runners WHERE key != ''").fetch_all(pool).await
294	}
295
296	/// A one-time registration token, valid for a day, for a runner of `scope`.
297	pub async fn add_registration(pool: &Pool, token_hash: &str, by: &str, scope: &Scope) -> sqlx::Result<()> {
298	let now = now_ms();
299	sqlx::query(
300	"INSERT INTO flow_runner_registrations (token_hash, created_by, created_ms, expires_ms, scope_owner, scope_repo_id) VALUES (?, ?, ?, ?, ?, ?)",
301	)
302	.bind(token_hash)
303	.bind(by)
304	.bind(now)
305	.bind(now + 24 * 3_600_000)
306	.bind(&scope.owner)
307	.bind(scope.repo_id)
308	.execute(pool)
309	.await?;
310	Ok(())
311	}
312
313	/// The runners that may take a repository's jobs: the site's, its owner's
314	/// and its own (oxygenhub#41).
315	pub async fn serving(pool: &Pool, owner: &str, repo_id: i64) -> sqlx::Result<Vec<Runner>> {
316	sqlx::query_as(
317	"SELECT * FROM flow_runners WHERE (scope_owner = '' AND scope_repo_id IS NULL) \
318	OR (scope_owner = ? AND scope_repo_id IS NULL) OR scope_repo_id = ? ORDER BY name",
319	)
320	.bind(owner)
321	.bind(repo_id)
322	.fetch_all(pool)
323	.await
324	}
325
326	/// The runners that serve an owner: the site's, and those scoped to the
327	/// owner or one of its repositories.
328	pub async fn serving_owner(pool: &Pool, owner: &str) -> sqlx::Result<Vec<Runner>> {
329	sqlx::query_as("SELECT * FROM flow_runners WHERE (scope_owner = '' AND scope_repo_id IS NULL) OR scope_owner = ? ORDER BY name")
330	.bind(owner)
331	.fetch_all(pool)
332	.await
333	}
334
335	/// Change the repositories a runner serves.
336	pub async fn set_scope(pool: &Pool, id: i64, scope: &Scope) -> sqlx::Result<()> {
337	sqlx::query("UPDATE flow_runners SET scope_owner = ?, scope_repo_id = ? WHERE id = ?")
338	.bind(&scope.owner)
339	.bind(scope.repo_id)
340	.bind(id)
341	.execute(pool)
342	.await?;
343	Ok(())
344	}
345
346	/// Registration tokens not used yet and not expired: (created by, created, expires).
347	pub async fn pending_registrations(pool: &Pool) -> sqlx::Result<Vec<(String, i64, i64)>> {
348	sqlx::query_as("SELECT created_by, created_ms, expires_ms FROM flow_runner_registrations WHERE used_ms IS NULL AND expires_ms > ? ORDER BY created_ms DESC")
349	.bind(now_ms())
350	.fetch_all(pool)
351	.await
352	}
353
354	/// Use registration token `token_hash` to register a runner: `None` when
355	/// the token is unknown, used or expired.
356	#[allow(clippy::too_many_arguments)]
357	pub async fn register(
358	pool: &Pool,
359	registration: &str,
360	name: &str,
361	token_hash: &str,
362	key: &str,
363	key_name: &str,
364	by_default: &str,
365	) -> sqlx::Result<Option<Runner>> {
366	let mut tx = super::begin_write(pool).await?;
367	let now = now_ms();
368	let found: Option<(String, String, Option<i64>)> = sqlx::query_as(
369	"SELECT created_by, scope_owner, scope_repo_id FROM flow_runner_registrations WHERE token_hash = ? AND used_ms IS NULL AND expires_ms > ?",
370	)
371	.bind(registration)
372	.bind(now)
373	.fetch_optional(&mut *tx)
374	.await?;
375	let Some((who, scope_owner, scope_repo_id)) = found else { return Ok(None) };
376	let id: i64 = sqlx::query_scalar(
377	"INSERT INTO flow_runners (name, token_hash, source, key, key_name, created_ms, created_by, scope_owner, scope_repo_id) VALUES (?, ?, 'registered', ?, ?, ?, ?, ?, ?) RETURNING id",
378	)
379	.bind(name)
380	.bind(token_hash)
381	.bind(key)
382	.bind(key_name)
383	.bind(now)
384	.bind(if who.is_empty() { by_default } else { &who })
385	.bind(scope_owner)
386	.bind(scope_repo_id)
387	.fetch_one(&mut *tx)
388	.await?;
389	sqlx::query("UPDATE flow_runner_registrations SET used_ms = ?, runner_id = ? WHERE token_hash = ?")
390	.bind(now)
391	.bind(id)
392	.bind(registration)
393	.execute(&mut *tx)
394	.await?;
395	tx.commit().await?;
396	get(pool, id).await
397	}
398
399	/// A job a runner ran, for its history.
400	#[derive(Clone, Debug, FromRow)]
401	pub struct RunnerJob {
402	pub owner: String,
403	pub repo: String,
404	pub run_number: i64,
405	pub job: String,
406	pub status: String,
407	pub started_ms: Option<i64>,
408	pub finished_ms: Option<i64>,
409	pub receipt_attest: String,
410	}
411
412	/// Jobs runner `id` ran or is running, newest first.
413	pub async fn history(pool: &Pool, id: i64, limit: i64, before: Option<i64>) -> sqlx::Result<Vec<RunnerJob>> {
414	sqlx::query_as(
415	"SELECT p.owner AS owner, p.name AS repo, r.number AS run_number, j.name AS job, j.status AS status, j.started_ms AS started_ms, \
416	j.finished_ms AS finished_ms, j.receipt_attest AS receipt_attest \
417	FROM flow_jobs j JOIN flow_runs r ON r.id = j.run_id JOIN repositories p ON p.id = r.repo_id \
418	WHERE j.runner_id = ? AND (? IS NULL OR j.started_ms < ?) ORDER BY j.started_ms DESC LIMIT ?",
419	)
420	.bind(id)
421	.bind(before)
422	.bind(before)
423	.bind(limit)
424	.fetch_all(pool)
425	.await
426	}
427
428	/// (jobs run, passed, busy milliseconds) of runner `id`.
429	pub async fn totals(pool: &Pool, id: i64) -> sqlx::Result<(i64, i64, i64)> {
430	sqlx::query_as(
431	"SELECT COUNT(*), COALESCE(SUM(status = 'passed'), 0), COALESCE(SUM(finished_ms - started_ms), 0) \
432	FROM flow_jobs WHERE runner_id = ? AND finished_ms IS NOT NULL",
433	)
434	.bind(id)
435	.fetch_one(pool)
436	.await
437	}
438
439	/// Jobs running on each runner now, by runner id.
440	pub async fn running(pool: &Pool) -> sqlx::Result<Vec<(i64, i64)>> {
441	sqlx::query_as("SELECT runner_id, COUNT(*) FROM flow_jobs WHERE status = 'running' AND runner_id IS NOT NULL GROUP BY runner_id")
442	.fetch_all(pool)
443	.await
444	}
445
446	pub async fn set_job_runner(pool: &Pool, job_id: i64, runner_id: i64) -> sqlx::Result<()> {
447	sqlx::query("UPDATE flow_jobs SET runner_id = ? WHERE id = ?").bind(runner_id).bind(job_id).execute(pool).await?;
448	Ok(())
449	}

Keyboard