| 1 | //! A run's life: made, its jobs waiting on the jobs they need, each job |
| 2 | //! skipped or expanded into its matrix and queued, started in a sandbox |
| 3 | //! when its workspace has room, reporting its steps and logs as it goes, |
| 4 | //! and finished; the run finishes with its last job. |
| 5 | |
| 6 | use g1t_actions::events::{RunInfo, runner_context}; |
| 7 | use g1t_actions::expr::{self, Scope, Status}; |
| 8 | use g1t_actions::matrix; |
| 9 | use g1t_actions::workflow::{self, Workflow}; |
| 10 | use g1t_contracts::actions::{JobCallArgs, RunActionArgs, StartJobArgs, WorkflowRun}; |
| 11 | use g1t_contracts::identity::{CreateAccessTokenArgs, CreatedAccessToken}; |
| 12 | use g1t_contracts::repos::{Repo, RepoPath}; |
| 13 | use g1t_contracts::time::rfc3339; |
| 14 | use g1t_contracts::{FailureCode, Outcome, new_id}; |
| 15 | use g1t_kit::now_ms; |
| 16 | use g1t_secrets::{random_hex, same, sha256_hex}; |
| 17 | use serde::Deserialize; |
| 18 | use serde_json::{Map, Value, json}; |
| 19 | use worker::Result; |
| 20 | |
| 21 | use crate::sync::WorkflowRow; |
| 22 | use crate::{Actions, Count, MAX_TIMEOUT_MINUTES, RUNNING_PER_WORKSPACE, SILENT_MS, SITE, check, fail, optional, repo_path}; |
| 23 | |
| 24 | /// The most log one job keeps, in bytes; past it, the log says so and stops. |
| 25 | const MAX_LOG_BYTES: usize = 4 * 1024 * 1024; |
| 26 | /// The most a single log report may add. |
| 27 | const MAX_CHUNK_BYTES: usize = 256 * 1024; |
| 28 | const MAX_ANNOTATIONS: usize = 50; |
| 29 | |
| 30 | pub struct NewRun { |
| 31 | pub repo: Repo, |
| 32 | pub path: String, |
| 33 | pub source: String, |
| 34 | pub workflow: Workflow, |
| 35 | pub info: RunInfo, |
| 36 | pub action: Option<String>, |
| 37 | pub pull: Option<u32>, |
| 38 | pub title: String, |
| 39 | pub inputs: Map<String, Value>, |
| 40 | pub event_key: String, |
| 41 | pub actor_id: Option<String>, |
| 42 | pub actor: Option<String>, |
| 43 | pub trusted: bool, |
| 44 | } |
| 45 | |
| 46 | #[derive(Clone, Deserialize)] |
| 47 | pub struct RunRow { |
| 48 | pub id: String, |
| 49 | pub workflow_id: String, |
| 50 | pub repo_id: String, |
| 51 | pub repo: String, |
| 52 | pub path: String, |
| 53 | pub name: String, |
| 54 | pub title: String, |
| 55 | pub number: u64, |
| 56 | pub attempt: u64, |
| 57 | pub event: String, |
| 58 | pub action: Option<String>, |
| 59 | pub git_ref: String, |
| 60 | pub sha: String, |
| 61 | pub pull: Option<u32>, |
| 62 | pub status: String, |
| 63 | pub conclusion: Option<String>, |
| 64 | pub error: Option<String>, |
| 65 | pub actor: Option<String>, |
| 66 | pub actor_id: Option<String>, |
| 67 | pub source: String, |
| 68 | pub info: String, |
| 69 | pub inputs: String, |
| 70 | pub trusted: u32, |
| 71 | pub concurrency_group: Option<String>, |
| 72 | pub created_at: String, |
| 73 | pub started_at: Option<String>, |
| 74 | pub finished_at: Option<String>, |
| 75 | } |
| 76 | |
| 77 | impl RunRow { |
| 78 | pub fn info(&self) -> RunInfo { |
| 79 | let mut info: RunInfo = serde_json::from_str(&self.info).unwrap_or_default(); |
| 80 | info.run_id = self.id.clone(); |
| 81 | info.run_number = self.number; |
| 82 | info.run_attempt = self.attempt; |
| 83 | info.workflow_path = self.path.clone(); |
| 84 | if info.workflow.is_empty() { |
| 85 | info.workflow = self.name.clone(); |
| 86 | } |
| 87 | info |
| 88 | } |
| 89 | |
| 90 | pub fn inputs(&self) -> Map<String, Value> { |
| 91 | serde_json::from_str(&self.inputs).unwrap_or_default() |
| 92 | } |
| 93 | |
| 94 | pub fn summary(&self) -> WorkflowRun { |
| 95 | WorkflowRun { |
| 96 | id: self.id.clone(), |
| 97 | workflow_id: self.workflow_id.clone(), |
| 98 | path: self.path.clone(), |
| 99 | name: self.name.clone(), |
| 100 | title: self.title.clone(), |
| 101 | number: self.number, |
| 102 | attempt: self.attempt, |
| 103 | event: self.event.clone(), |
| 104 | git_ref: self.git_ref.clone(), |
| 105 | sha: self.sha.clone(), |
| 106 | pull: self.pull, |
| 107 | status: self.status.clone(), |
| 108 | conclusion: self.conclusion.clone(), |
| 109 | error: self.error.clone(), |
| 110 | actor: self.actor.clone(), |
| 111 | created_at: self.created_at.clone(), |
| 112 | started_at: self.started_at.clone(), |
| 113 | finished_at: self.finished_at.clone(), |
| 114 | } |
| 115 | } |
| 116 | } |
| 117 | |
| 118 | #[derive(Clone, Deserialize)] |
| 119 | pub struct JobRow { |
| 120 | pub id: String, |
| 121 | pub run_id: String, |
| 122 | pub repo_id: String, |
| 123 | pub namespace: String, |
| 124 | pub key: String, |
| 125 | pub ordinal: u32, |
| 126 | pub name: String, |
| 127 | pub needs: String, |
| 128 | pub matrix: Option<String>, |
| 129 | pub status: String, |
| 130 | pub conclusion: Option<String>, |
| 131 | pub steps: String, |
| 132 | pub annotations: String, |
| 133 | pub outputs: String, |
| 134 | pub reason: Option<String>, |
| 135 | pub token_hash: Option<String>, |
| 136 | pub timeout_minutes: u32, |
| 137 | pub continue_on_error: u32, |
| 138 | pub max_parallel: Option<u32>, |
| 139 | pub seen_at: Option<String>, |
| 140 | pub started_at: Option<String>, |
| 141 | pub finished_at: Option<String>, |
| 142 | } |
| 143 | |
| 144 | impl JobRow { |
| 145 | pub fn needs(&self) -> Vec<String> { |
| 146 | serde_json::from_str(&self.needs).unwrap_or_default() |
| 147 | } |
| 148 | } |
| 149 | |
| 150 | /// What the jobs of one key came to, for `needs.<key>`. |
| 151 | fn key_result(rows: &[&JobRow]) -> &'static str { |
| 152 | let failed = |row: &&&JobRow| row.conclusion.as_deref() == Some("failure") && row.continue_on_error == 0; |
| 153 | if rows.iter().any(|row| failed(&row)) { |
| 154 | "failure" |
| 155 | } else if rows.iter().any(|row| row.conclusion.as_deref() == Some("cancelled")) { |
| 156 | "cancelled" |
| 157 | } else if rows.iter().all(|row| row.conclusion.as_deref() == Some("skipped")) { |
| 158 | "skipped" |
| 159 | } else { |
| 160 | "success" |
| 161 | } |
| 162 | } |
| 163 | |
| 164 | fn now() -> String { |
| 165 | rfc3339(now_ms()) |
| 166 | } |
| 167 | |
| 168 | impl Actions { |
| 169 | pub async fn run_row(&self, id: &str) -> Result<Option<RunRow>> { |
| 170 | self.db.prepare("SELECT * FROM runs WHERE id = ?").bind(&[id.into()])?.first::<RunRow>(None).await |
| 171 | } |
| 172 | |
| 173 | pub async fn job_rows(&self, run_id: &str) -> Result<Vec<JobRow>> { |
| 174 | self.db |
| 175 | .prepare("SELECT * FROM jobs WHERE run_id = ? ORDER BY rowid") |
| 176 | .bind(&[run_id.into()])? |
| 177 | .all() |
| 178 | .await? |
| 179 | .results::<JobRow>() |
| 180 | } |
| 181 | |
| 182 | pub async fn run_summary(&self, id: &str) -> Result<Outcome<WorkflowRun>> { |
| 183 | Ok(match self.run_row(id).await? { |
| 184 | Some(row) => Outcome::Ok(row.summary()), |
| 185 | None => fail(FailureCode::NotFound, "No such run."), |
| 186 | }) |
| 187 | } |
| 188 | |
| 189 | /// The contexts every expression outside a job's steps may use. |
| 190 | fn base_contexts(run: &RunRow, vars: &Map<String, Value>, job: &str) -> Map<String, Value> { |
| 191 | let mut contexts = Map::new(); |
| 192 | contexts.insert("github".into(), run.info().context(job, "", run.action.as_deref())); |
| 193 | contexts.insert("inputs".into(), Value::Object(run.inputs())); |
| 194 | contexts.insert("vars".into(), Value::Object(vars.clone())); |
| 195 | contexts.insert("needs".into(), json!({})); |
| 196 | contexts.insert("runner".into(), runner_context()); |
| 197 | contexts |
| 198 | } |
| 199 | |
| 200 | /// Makes a run and its jobs, and starts what can start. `None` when the |
| 201 | /// event already started this workflow. |
| 202 | pub async fn create_run(&self, new: NewRun) -> Result<Option<String>> { |
| 203 | let workflow_row = self.workflow_row(&new.repo, &new.path, &new.workflow.display_name(&new.path), &new.source).await?; |
| 204 | let numbered = self |
| 205 | .db |
| 206 | .prepare("UPDATE workflows SET run_count = run_count + 1 WHERE id = ? RETURNING run_count AS n") |
| 207 | .bind(&[workflow_row.id.as_str().into()])? |
| 208 | .first::<Count>(None) |
| 209 | .await? |
| 210 | .map_or(1, |count| count.n); |
| 211 | let id = new_id("run", now_ms()); |
| 212 | let mut info = new.info.clone(); |
| 213 | info.workflow = new.workflow.display_name(&new.path); |
| 214 | info.workflow_path = new.path.clone(); |
| 215 | info.run_id = id.clone(); |
| 216 | info.run_number = u64::from(numbered); |
| 217 | let vars = self.variables_for(&new.repo.id, &new.repo.namespace).await?; |
| 218 | |
| 219 | // run-name and the concurrency group read github, inputs and vars. |
| 220 | let mut contexts = Map::new(); |
| 221 | contexts.insert("github".into(), info.context("", "", new.action.as_deref())); |
| 222 | contexts.insert("inputs".into(), Value::Object(new.inputs.clone())); |
| 223 | contexts.insert("vars".into(), Value::Object(vars)); |
| 224 | let scope = Scope { |
| 225 | contexts: &contexts, |
| 226 | status: Status::Success, |
| 227 | hash_files: None, |
| 228 | }; |
| 229 | let title = new |
| 230 | .workflow |
| 231 | .run_name |
| 232 | .as_deref() |
| 233 | .and_then(|run_name| expr::interpolate(run_name, &scope).ok()) |
| 234 | .filter(|title| !title.trim().is_empty()) |
| 235 | .unwrap_or(new.title.clone()); |
| 236 | let group = new.workflow.concurrency.as_ref().and_then(|c| expr::interpolate(&c.group, &scope).ok()); |
| 237 | let cancel_in_progress = new |
| 238 | .workflow |
| 239 | .concurrency |
| 240 | .as_ref() |
| 241 | .and_then(|c| expr::interpolate_value(&c.cancel_in_progress, &scope).ok()) |
| 242 | .is_some_and(|value| expr::truthy(&value)); |
| 243 | |
| 244 | let inserted = self |
| 245 | .db |
| 246 | .prepare( |
| 247 | "INSERT OR IGNORE INTO runs (id, workflow_id, repo_id, repo, path, name, title, number, event, action, git_ref, sha, |
| 248 | pull, status, actor, actor_id, source, info, inputs, trusted, concurrency_group, event_key, created_at) |
| 249 | VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, 'queued', ?, ?, ?, ?, ?, ?, ?, ?, ?) RETURNING id", |
| 250 | ) |
| 251 | .bind(&[ |
| 252 | id.as_str().into(), |
| 253 | workflow_row.id.as_str().into(), |
| 254 | new.repo.id.as_str().into(), |
| 255 | format!("{}/{}", new.repo.namespace, new.repo.name).into(), |
| 256 | new.path.as_str().into(), |
| 257 | info.workflow.as_str().into(), |
| 258 | title.as_str().into(), |
| 259 | numbered.into(), |
| 260 | info.event_name.as_str().into(), |
| 261 | optional(new.action.as_deref()), |
| 262 | info.git_ref.as_str().into(), |
| 263 | info.sha.as_str().into(), |
| 264 | new.pull.map_or(worker::wasm_bindgen::JsValue::NULL, Into::into), |
| 265 | optional(new.actor.as_deref()), |
| 266 | optional(new.actor_id.as_deref()), |
| 267 | new.source.as_str().into(), |
| 268 | serde_json::to_string(&info)?.into(), |
| 269 | serde_json::to_string(&new.inputs)?.into(), |
| 270 | u32::from(new.trusted).into(), |
| 271 | optional(group.as_deref()), |
| 272 | new.event_key.as_str().into(), |
| 273 | now().into(), |
| 274 | ])? |
| 275 | .first::<Value>(None) |
| 276 | .await?; |
| 277 | if inserted.is_none() { |
| 278 | return Ok(None); |
| 279 | } |
| 280 | if let Some(run) = self.run_row(&id).await? { |
| 281 | self.report_pending(&run).await?; |
| 282 | } |
| 283 | |
| 284 | // Every job, waiting; each is expanded when the jobs it needs are done. |
| 285 | let mut statements = Vec::new(); |
| 286 | for job in &new.workflow.jobs { |
| 287 | statements.push( |
| 288 | self.db |
| 289 | .prepare( |
| 290 | "INSERT INTO jobs (id, run_id, repo_id, namespace, key, name, needs, status) VALUES (?, ?, ?, ?, ?, ?, ?, 'waiting')", |
| 291 | ) |
| 292 | .bind(&[ |
| 293 | new_id("job", now_ms()).into(), |
| 294 | id.as_str().into(), |
| 295 | new.repo.id.as_str().into(), |
| 296 | new.repo.namespace.as_str().into(), |
| 297 | job.id.as_str().into(), |
| 298 | job.name.clone().filter(|n| !expr::has_expression(n)).unwrap_or(job.id.clone()).into(), |
| 299 | serde_json::to_string(&job.needs)?.into(), |
| 300 | ])?, |
| 301 | ); |
| 302 | } |
| 303 | self.db.batch(statements).await?; |
| 304 | |
| 305 | // One run at a time per concurrency group. |
| 306 | if let Some(group) = &group { |
| 307 | let others = self |
| 308 | .db |
| 309 | .prepare("SELECT * FROM runs WHERE repo_id = ? AND concurrency_group = ? AND id != ? AND status != 'completed' ORDER BY id") |
| 310 | .bind(&[new.repo.id.as_str().into(), group.as_str().into(), id.as_str().into()])? |
| 311 | .all() |
| 312 | .await? |
| 313 | .results::<RunRow>()?; |
| 314 | for other in &others { |
| 315 | if cancel_in_progress || other.status == "pending" { |
| 316 | // A newer run replaces a waiting one, as on GitHub. |
| 317 | self.cancel_run(other, "A newer run in the same concurrency group replaced it.").await?; |
| 318 | } |
| 319 | } |
| 320 | if !cancel_in_progress && others.iter().any(|other| other.status != "pending") { |
| 321 | self.db |
| 322 | .prepare("UPDATE runs SET status = 'pending' WHERE id = ?") |
| 323 | .bind(&[id.as_str().into()])? |
| 324 | .run() |
| 325 | .await?; |
| 326 | return Ok(Some(id)); |
| 327 | } |
| 328 | } |
| 329 | self.advance(&id).await?; |
| 330 | Ok(Some(id)) |
| 331 | } |
| 332 | |
| 333 | /// A run that could not start, such as for a workflow file that does not read. |
| 334 | #[allow(clippy::too_many_arguments)] |
| 335 | pub async fn record_failed_run( |
| 336 | &self, |
| 337 | row: &WorkflowRow, |
| 338 | git_ref: &str, |
| 339 | sha: &str, |
| 340 | event_key: &str, |
| 341 | actor_id: Option<&str>, |
| 342 | actor: &str, |
| 343 | problem: &str, |
| 344 | ) -> Result<()> { |
| 345 | let numbered = self |
| 346 | .db |
| 347 | .prepare("UPDATE workflows SET run_count = run_count + 1 WHERE id = ? RETURNING run_count AS n") |
| 348 | .bind(&[row.id.as_str().into()])? |
| 349 | .first::<Count>(None) |
| 350 | .await? |
| 351 | .map_or(1, |count| count.n); |
| 352 | let at = now(); |
| 353 | self.db |
| 354 | .prepare( |
| 355 | "INSERT OR IGNORE INTO runs (id, workflow_id, repo_id, repo, path, name, title, number, event, git_ref, sha, status, |
| 356 | conclusion, error, actor, actor_id, source, info, event_key, created_at, finished_at) |
| 357 | VALUES (?, ?, ?, ?, ?, ?, ?, ?, 'push', ?, ?, 'completed', 'failure', ?, ?, ?, ?, '{}', ?, ?, ?)", |
| 358 | ) |
| 359 | .bind(&[ |
| 360 | new_id("run", now_ms()).into(), |
| 361 | row.id.as_str().into(), |
| 362 | row.repo_id.as_str().into(), |
| 363 | row.repo.as_str().into(), |
| 364 | row.path.as_str().into(), |
| 365 | row.path.as_str().into(), |
| 366 | "Invalid workflow file".into(), |
| 367 | numbered.into(), |
| 368 | git_ref.into(), |
| 369 | sha.into(), |
| 370 | problem.into(), |
| 371 | actor.into(), |
| 372 | optional(actor_id), |
| 373 | row.source.as_str().into(), |
| 374 | event_key.into(), |
| 375 | at.as_str().into(), |
| 376 | at.as_str().into(), |
| 377 | ])? |
| 378 | .run() |
| 379 | .await?; |
| 380 | Ok(()) |
| 381 | } |
| 382 | |
| 383 | /// Moves a run along: jobs whose needs are done are decided on, and |
| 384 | /// jobs that can start are started. |
| 385 | pub async fn advance(&self, run_id: &str) -> Result<()> { |
| 386 | // Each pass may finish jobs (skipped ones), which may free others. |
| 387 | for _ in 0..20 { |
| 388 | let Some(run) = self.run_row(run_id).await? else { return Ok(()) }; |
| 389 | if run.status == "completed" || run.status == "pending" { |
| 390 | return Ok(()); |
| 391 | } |
| 392 | let workflow = match workflow::parse(&run.source) { |
| 393 | Ok(workflow) => workflow, |
| 394 | Err(problem) => { |
| 395 | self.finish_run(&run, Some(&problem)).await?; |
| 396 | return Ok(()); |
| 397 | } |
| 398 | }; |
| 399 | let jobs = self.job_rows(run_id).await?; |
| 400 | let mut changed = false; |
| 401 | for job in &workflow.jobs { |
| 402 | let rows: Vec<&JobRow> = jobs.iter().filter(|row| row.key == job.id).collect(); |
| 403 | if rows.is_empty() || !rows.iter().all(|row| row.status == "waiting") { |
| 404 | continue; |
| 405 | } |
| 406 | let needed: Vec<(&String, Vec<&JobRow>)> = |
| 407 | job.needs.iter().map(|need| (need, jobs.iter().filter(|row| &row.key == need).collect())).collect(); |
| 408 | if !needed.iter().all(|(_, rows)| rows.iter().all(|row| row.status == "completed")) { |
| 409 | continue; |
| 410 | } |
| 411 | self.decide(&run, job, rows[0], &needed).await?; |
| 412 | changed = true; |
| 413 | } |
| 414 | if !changed { |
| 415 | break; |
| 416 | } |
| 417 | } |
| 418 | self.start_queued().await?; |
| 419 | self.finish_if_done(run_id).await |
| 420 | } |
| 421 | |
| 422 | /// Decides on one job whose needs are done: skip it, fail it, or expand |
| 423 | /// it into its matrix and queue it. |
| 424 | async fn decide(&self, run: &RunRow, job: &workflow::Job, row: &JobRow, needed: &[(&String, Vec<&JobRow>)]) -> Result<()> { |
| 425 | let vars = self.variables_for(&run.repo_id, &repo_path(&run.repo).namespace).await?; |
| 426 | let mut contexts = Self::base_contexts(run, &vars, &job.id); |
| 427 | let mut needs = Map::new(); |
| 428 | let mut status = if run.conclusion.as_deref() == Some("cancelled") { Status::Cancelled } else { Status::Success }; |
| 429 | for (key, rows) in needed { |
| 430 | let result = key_result(rows); |
| 431 | let mut outputs = Map::new(); |
| 432 | for row in rows { |
| 433 | if let Ok(Value::Object(more)) = serde_json::from_str::<Value>(&row.outputs) { |
| 434 | outputs.extend(more); |
| 435 | } |
| 436 | } |
| 437 | if result != "success" && status == Status::Success { |
| 438 | status = Status::Failure; |
| 439 | } |
| 440 | needs.insert((*key).clone(), json!({ "result": result, "outputs": outputs })); |
| 441 | } |
| 442 | contexts.insert("needs".into(), Value::Object(needs)); |
| 443 | let scope = Scope { |
| 444 | contexts: &contexts, |
| 445 | status, |
| 446 | hash_files: None, |
| 447 | }; |
| 448 | let condition = job.condition.as_deref().unwrap_or_default(); |
| 449 | match expr::condition(condition, &scope) { |
| 450 | Ok(true) => {} |
| 451 | Ok(false) => return self.skip_job(row, None).await, |
| 452 | Err(problem) => return self.fail_job(row, &format!("Its `if` does not read: {problem}")).await, |
| 453 | } |
| 454 | if job.uses.is_some() { |
| 455 | return self.fail_job(row, "Reusable workflows (`uses:` on a job) are not called on g1t yet.").await; |
| 456 | } |
| 457 | |
| 458 | // Its matrix, which may come from a needed job's outputs. |
| 459 | let combinations = match &job.matrix { |
| 460 | None => vec![Map::new()], |
| 461 | Some(matrix) => { |
| 462 | let value = match expr::interpolate_value(matrix, &scope) { |
| 463 | Ok(value) => value, |
| 464 | Err(problem) => return self.fail_job(row, &format!("Its matrix does not read: {problem}")).await, |
| 465 | }; |
| 466 | match matrix::expand(&value) { |
| 467 | Ok(combinations) if !combinations.is_empty() => combinations, |
| 468 | Ok(_) => return self.fail_job(row, "Its matrix makes no jobs.").await, |
| 469 | Err(problem) => return self.fail_job(row, &problem).await, |
| 470 | } |
| 471 | } |
| 472 | }; |
| 473 | let total = combinations.len(); |
| 474 | let raw = job.raw.as_object().cloned().unwrap_or_default(); |
| 475 | let mut statements = Vec::new(); |
| 476 | for (index, combination) in combinations.iter().enumerate() { |
| 477 | let mut contexts = contexts.clone(); |
| 478 | contexts.insert("matrix".into(), Value::Object(combination.clone())); |
| 479 | contexts.insert( |
| 480 | "strategy".into(), |
| 481 | json!({ "fail-fast": job.fail_fast, "job-index": index, "job-total": total, "max-parallel": job.max_parallel.unwrap_or(total as u32) }), |
| 482 | ); |
| 483 | let scope = Scope { |
| 484 | contexts: &contexts, |
| 485 | status: Status::Success, |
| 486 | hash_files: None, |
| 487 | }; |
| 488 | let base_name = job.name.clone().unwrap_or(job.id.clone()); |
| 489 | let name = if expr::has_expression(&base_name) { |
| 490 | expr::interpolate(&base_name, &scope).unwrap_or(base_name) |
| 491 | } else if job.matrix.is_some() { |
| 492 | matrix::job_name(&base_name, combination) |
| 493 | } else { |
| 494 | base_name |
| 495 | }; |
| 496 | let runs_on = expr::interpolate_value(&job.runs_on, &scope).unwrap_or(Value::Null); |
| 497 | let labels = match &runs_on { |
| 498 | Value::String(label) => vec![label.clone()], |
| 499 | Value::Array(labels) => labels.iter().map(expr::to_text).collect(), |
| 500 | Value::Object(spec) => spec.get("labels").map(|l| match l { |
| 501 | Value::Array(labels) => labels.iter().map(expr::to_text).collect(), |
| 502 | other => vec![expr::to_text(other)], |
| 503 | }).unwrap_or_default(), |
| 504 | _ => Vec::new(), |
| 505 | }; |
| 506 | let reason = labels |
| 507 | .iter() |
| 508 | .find(|label| { |
| 509 | let lower = label.to_ascii_lowercase(); |
| 510 | lower.contains("windows") || lower.contains("macos") |
| 511 | }) |
| 512 | .map(|label| format!("`runs-on: {label}`: g1t runs jobs on Linux only.")); |
| 513 | let timeout = raw |
| 514 | .get("timeout-minutes") |
| 515 | .and_then(|value| expr::interpolate_value(value, &scope).ok()) |
| 516 | .and_then(|value| value.as_f64().or_else(|| expr::to_text(&value).parse().ok())) |
| 517 | .map_or(MAX_TIMEOUT_MINUTES, |minutes| (minutes.ceil() as u32).clamp(1, MAX_TIMEOUT_MINUTES)); |
| 518 | let continue_on_error = raw |
| 519 | .get("continue-on-error") |
| 520 | .and_then(|value| expr::interpolate_value(value, &scope).ok()) |
| 521 | .is_some_and(|value| expr::truthy(&value)); |
| 522 | let (status, conclusion, finished) = match &reason { |
| 523 | Some(_) => ("completed", Some("failure"), Some(now())), |
| 524 | None => ("queued", None, None), |
| 525 | }; |
| 526 | let values: Vec<worker::wasm_bindgen::JsValue> = vec![ |
| 527 | name.into(), |
| 528 | serde_json::to_string(combination)?.into(), |
| 529 | status.into(), |
| 530 | optional(conclusion), |
| 531 | optional(reason.as_deref()), |
| 532 | timeout.into(), |
| 533 | u32::from(continue_on_error).into(), |
| 534 | job.max_parallel.map_or(worker::wasm_bindgen::JsValue::NULL, Into::into), |
| 535 | optional(finished.as_deref()), |
| 536 | ]; |
| 537 | if index == 0 { |
| 538 | let mut bound = values; |
| 539 | bound.push(row.id.as_str().into()); |
| 540 | statements.push( |
| 541 | self.db |
| 542 | .prepare( |
| 543 | "UPDATE jobs SET name = ?, matrix = ?, status = ?, conclusion = ?, reason = ?, timeout_minutes = ?, |
| 544 | continue_on_error = ?, max_parallel = ?, finished_at = ? WHERE id = ?", |
| 545 | ) |
| 546 | .bind(&bound)?, |
| 547 | ); |
| 548 | } else { |
| 549 | let mut bound: Vec<worker::wasm_bindgen::JsValue> = vec![ |
| 550 | new_id("job", now_ms()).into(), |
| 551 | row.run_id.as_str().into(), |
| 552 | row.repo_id.as_str().into(), |
| 553 | row.namespace.as_str().into(), |
| 554 | row.key.as_str().into(), |
| 555 | (index as u32).into(), |
| 556 | row.needs.as_str().into(), |
| 557 | ]; |
| 558 | bound.extend(values); |
| 559 | statements.push( |
| 560 | self.db |
| 561 | .prepare( |
| 562 | "INSERT INTO jobs (id, run_id, repo_id, namespace, key, ordinal, needs, name, matrix, status, conclusion, reason, |
| 563 | timeout_minutes, continue_on_error, max_parallel, finished_at) |
| 564 | VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)", |
| 565 | ) |
| 566 | .bind(&bound)?, |
| 567 | ); |
| 568 | } |
| 569 | } |
| 570 | self.db.batch(statements).await?; |
| 571 | Ok(()) |
| 572 | } |
| 573 | |
| 574 | async fn skip_job(&self, row: &JobRow, reason: Option<&str>) -> Result<()> { |
| 575 | self.db |
| 576 | .prepare("UPDATE jobs SET status = 'completed', conclusion = 'skipped', reason = ?, finished_at = ? WHERE id = ?") |
| 577 | .bind(&[optional(reason), now().into(), row.id.as_str().into()])? |
| 578 | .run() |
| 579 | .await?; |
| 580 | Ok(()) |
| 581 | } |
| 582 | |
| 583 | async fn fail_job(&self, row: &JobRow, reason: &str) -> Result<()> { |
| 584 | self.db |
| 585 | .prepare("UPDATE jobs SET status = 'completed', conclusion = 'failure', reason = ?, finished_at = ? WHERE id = ? AND status != 'completed'") |
| 586 | .bind(&[reason.into(), now().into(), row.id.as_str().into()])? |
| 587 | .run() |
| 588 | .await?; |
| 589 | Ok(()) |
| 590 | } |
| 591 | |
| 592 | /// Starts queued jobs, oldest first, while their workspace has room. |
| 593 | pub async fn start_queued(&self) -> Result<()> { |
| 594 | let queued = self |
| 595 | .db |
| 596 | .prepare("SELECT * FROM jobs WHERE status = 'queued' ORDER BY rowid LIMIT 50") |
| 597 | .all() |
| 598 | .await? |
| 599 | .results::<JobRow>()?; |
| 600 | let mut running: std::collections::HashMap<String, u32> = std::collections::HashMap::new(); |
| 601 | for job in queued { |
| 602 | let in_workspace = match running.get(&job.namespace) { |
| 603 | Some(n) => *n, |
| 604 | None => { |
| 605 | let n = self |
| 606 | .db |
| 607 | .prepare("SELECT COUNT(*) AS n FROM jobs WHERE status = 'in_progress' AND namespace = ?") |
| 608 | .bind(&[job.namespace.as_str().into()])? |
| 609 | .first::<Count>(None) |
| 610 | .await? |
| 611 | .map_or(0, |count| count.n); |
| 612 | running.insert(job.namespace.clone(), n); |
| 613 | n |
| 614 | } |
| 615 | }; |
| 616 | if in_workspace >= RUNNING_PER_WORKSPACE { |
| 617 | continue; |
| 618 | } |
| 619 | if let Some(max) = job.max_parallel { |
| 620 | let siblings = self |
| 621 | .db |
| 622 | .prepare("SELECT COUNT(*) AS n FROM jobs WHERE run_id = ? AND key = ? AND status = 'in_progress'") |
| 623 | .bind(&[job.run_id.as_str().into(), job.key.as_str().into()])? |
| 624 | .first::<Count>(None) |
| 625 | .await? |
| 626 | .map_or(0, |count| count.n); |
| 627 | if siblings >= max { |
| 628 | continue; |
| 629 | } |
| 630 | } |
| 631 | let token = random_hex(24); |
| 632 | let at = now(); |
| 633 | let claimed = self |
| 634 | .db |
| 635 | .prepare( |
| 636 | "UPDATE jobs SET status = 'in_progress', token_hash = ?, started_at = ?, seen_at = ? WHERE id = ? AND status = 'queued' RETURNING id", |
| 637 | ) |
| 638 | .bind(&[sha256_hex(&token).into(), at.as_str().into(), at.as_str().into(), job.id.as_str().into()])? |
| 639 | .first::<Value>(None) |
| 640 | .await?; |
| 641 | if claimed.is_none() { |
| 642 | continue; |
| 643 | } |
| 644 | running.insert(job.namespace.clone(), in_workspace + 1); |
| 645 | self.db |
| 646 | .prepare("UPDATE runs SET status = 'in_progress', started_at = COALESCE(started_at, ?) WHERE id = ? AND status = 'queued'") |
| 647 | .bind(&[at.as_str().into(), job.run_id.as_str().into()])? |
| 648 | .run() |
| 649 | .await?; |
| 650 | let run = self.run_row(&job.run_id).await?; |
| 651 | let repo: RepoPath = run.as_ref().map(|run| repo_path(&run.repo)).unwrap_or(RepoPath { |
| 652 | namespace: job.namespace.clone(), |
| 653 | name: String::new(), |
| 654 | }); |
| 655 | let started: Outcome<Value> = g1t_kit::call( |
| 656 | &self.runner, |
| 657 | "start_actions_job", |
| 658 | &StartJobArgs { |
| 659 | job: job.id.clone(), |
| 660 | token, |
| 661 | repo, |
| 662 | timeout_minutes: job.timeout_minutes, |
| 663 | }, |
| 664 | ) |
| 665 | .await |
| 666 | .unwrap_or_else(|error| fail(FailureCode::Conflict, format!("The runner could not be reached: {error}"))); |
| 667 | if let Outcome::Fail(refused) = started { |
| 668 | Box::pin(self.finish_job(&job.id, "failure", Some(&refused.message), None)).await?; |
| 669 | } |
| 670 | } |
| 671 | Ok(()) |
| 672 | } |
| 673 | |
| 674 | /// Finishes a job and moves its run along. |
| 675 | pub async fn finish_job(&self, job_id: &str, conclusion: &str, reason: Option<&str>, outputs: Option<&Map<String, Value>>) -> Result<()> { |
| 676 | let finished = self |
| 677 | .db |
| 678 | .prepare( |
| 679 | "UPDATE jobs SET status = 'completed', conclusion = ?, reason = COALESCE(?, reason), outputs = COALESCE(?, outputs), |
| 680 | finished_at = ?, token_hash = NULL WHERE id = ? AND status != 'completed' RETURNING *", |
| 681 | ) |
| 682 | .bind(&[ |
| 683 | conclusion.into(), |
| 684 | optional(reason), |
| 685 | outputs.map(|o| serde_json::to_string(o).unwrap_or_default()).as_deref().map_or(worker::wasm_bindgen::JsValue::NULL, Into::into), |
| 686 | now().into(), |
| 687 | job_id.into(), |
| 688 | ])? |
| 689 | .first::<JobRow>(None) |
| 690 | .await?; |
| 691 | let Some(job) = finished else { return Ok(()) }; |
| 692 | // Steps still marked as going are not going any more. |
| 693 | let mut steps: Vec<Value> = serde_json::from_str(&job.steps).unwrap_or_default(); |
| 694 | let mut touched = false; |
| 695 | for step in steps.iter_mut() { |
| 696 | if step["status"] != "completed" { |
| 697 | let was_running = step["status"] == "in_progress"; |
| 698 | step["status"] = json!("completed"); |
| 699 | step["conclusion"] = json!(if was_running { conclusion } else { "skipped" }); |
| 700 | touched = true; |
| 701 | } |
| 702 | } |
| 703 | if touched { |
| 704 | self.db |
| 705 | .prepare("UPDATE jobs SET steps = ? WHERE id = ?") |
| 706 | .bind(&[serde_json::to_string(&steps)?.into(), job.id.as_str().into()])? |
| 707 | .run() |
| 708 | .await?; |
| 709 | } |
| 710 | // fail-fast: one failed combination stops the rest of its matrix. |
| 711 | if conclusion == "failure" && job.continue_on_error == 0 && job.matrix.as_deref().is_some_and(|m| m != "{}") { |
| 712 | let run = self.run_row(&job.run_id).await?; |
| 713 | let fail_fast = run |
| 714 | .as_ref() |
| 715 | .and_then(|run| workflow::parse(&run.source).ok()) |
| 716 | .and_then(|workflow| workflow.jobs.into_iter().find(|j| j.id == job.key)) |
| 717 | .is_none_or(|j| j.fail_fast); |
| 718 | if fail_fast { |
| 719 | let siblings = self |
| 720 | .db |
| 721 | .prepare("SELECT * FROM jobs WHERE run_id = ? AND key = ? AND status != 'completed'") |
| 722 | .bind(&[job.run_id.as_str().into(), job.key.as_str().into()])? |
| 723 | .all() |
| 724 | .await? |
| 725 | .results::<JobRow>()?; |
| 726 | for sibling in siblings { |
| 727 | self.stop_job(&sibling, "Another job of its matrix failed, and the matrix is fail-fast.").await?; |
| 728 | } |
| 729 | } |
| 730 | } |
| 731 | Box::pin(self.advance(&job.run_id)).await |
| 732 | } |
| 733 | |
| 734 | /// Cancels a job, stopping its sandbox if it has one. |
| 735 | async fn stop_job(&self, job: &JobRow, reason: &str) -> Result<()> { |
| 736 | if job.status == "in_progress" { |
| 737 | let _: Result<Value> = g1t_kit::call(&self.runner, "stop_actions_job", &json!({ "job": job.id })).await; |
| 738 | } |
| 739 | self.db |
| 740 | .prepare("UPDATE jobs SET status = 'completed', conclusion = 'cancelled', reason = ?, finished_at = ?, token_hash = NULL WHERE id = ? AND status != 'completed'") |
| 741 | .bind(&[reason.into(), now().into(), job.id.as_str().into()])? |
| 742 | .run() |
| 743 | .await?; |
| 744 | Ok(()) |
| 745 | } |
| 746 | |
| 747 | /// Finishes the run when every job has. |
| 748 | async fn finish_if_done(&self, run_id: &str) -> Result<()> { |
| 749 | let Some(run) = self.run_row(run_id).await? else { return Ok(()) }; |
| 750 | if run.status == "completed" || run.status == "pending" { |
| 751 | return Ok(()); |
| 752 | } |
| 753 | let jobs = self.job_rows(run_id).await?; |
| 754 | if !jobs.iter().all(|job| job.status == "completed") { |
| 755 | return Ok(()); |
| 756 | } |
| 757 | self.finish_run(&run, None).await |
| 758 | } |
| 759 | |
| 760 | async fn finish_run(&self, run: &RunRow, error: Option<&str>) -> Result<()> { |
| 761 | let jobs = self.job_rows(&run.id).await?; |
| 762 | let rows: Vec<&JobRow> = jobs.iter().collect(); |
| 763 | let conclusion = if error.is_some() { |
| 764 | "failure" |
| 765 | } else if run.conclusion.as_deref() == Some("cancelled") { |
| 766 | "cancelled" |
| 767 | } else if rows.is_empty() { |
| 768 | "skipped" |
| 769 | } else { |
| 770 | key_result(&rows) |
| 771 | }; |
| 772 | let done = self |
| 773 | .db |
| 774 | .prepare("UPDATE runs SET status = 'completed', conclusion = ?, error = COALESCE(?, error), finished_at = ? WHERE id = ? AND status != 'completed' RETURNING id") |
| 775 | .bind(&[conclusion.into(), optional(error), now().into(), run.id.as_str().into()])? |
| 776 | .first::<Value>(None) |
| 777 | .await?; |
| 778 | if done.is_none() { |
| 779 | return Ok(()); |
| 780 | } |
| 781 | self.report_status(run, conclusion).await?; |
| 782 | // The next run waiting in its concurrency group. |
| 783 | if let Some(group) = &run.concurrency_group { |
| 784 | let next = self |
| 785 | .db |
| 786 | .prepare("SELECT * FROM runs WHERE repo_id = ? AND concurrency_group = ? AND status = 'pending' ORDER BY id LIMIT 1") |
| 787 | .bind(&[run.repo_id.as_str().into(), group.as_str().into()])? |
| 788 | .first::<RunRow>(None) |
| 789 | .await?; |
| 790 | if let Some(next) = next { |
| 791 | self.db.prepare("UPDATE runs SET status = 'queued' WHERE id = ?").bind(&[next.id.as_str().into()])?.run().await?; |
| 792 | Box::pin(self.advance(&next.id)).await?; |
| 793 | } |
| 794 | } |
| 795 | Ok(()) |
| 796 | } |
| 797 | |
| 798 | /// Tells the pull request (or commit) how the run went, as a status. |
| 799 | async fn report_status(&self, run: &RunRow, conclusion: &str) -> Result<()> { |
| 800 | let state = match conclusion { |
| 801 | "success" | "skipped" => "success", |
| 802 | "cancelled" => "error", |
| 803 | _ => "failure", |
| 804 | }; |
| 805 | let _: Result<Value> = g1t_kit::call( |
| 806 | &self.work, |
| 807 | "set_commit_status", |
| 808 | &json!({ |
| 809 | "repoId": run.repo_id, |
| 810 | "sha": run.sha, |
| 811 | "context": format!("{} / {}", run.name, run.event), |
| 812 | "state": state, |
| 813 | "description": format!("{} {}", run.name, match conclusion { |
| 814 | "success" => "passed", |
| 815 | "skipped" => "was skipped", |
| 816 | "cancelled" => "was cancelled", |
| 817 | _ => "failed", |
| 818 | }), |
| 819 | "targetUrl": format!("{SITE}/{}/actions/runs/{}", run.repo, run.id), |
| 820 | }), |
| 821 | ) |
| 822 | .await; |
| 823 | Ok(()) |
| 824 | } |
| 825 | |
| 826 | /// Tells the pull request a run has started on its head. |
| 827 | pub async fn report_pending(&self, run: &RunRow) -> Result<()> { |
| 828 | let _: Result<Value> = g1t_kit::call( |
| 829 | &self.work, |
| 830 | "set_commit_status", |
| 831 | &json!({ |
| 832 | "repoId": run.repo_id, |
| 833 | "sha": run.sha, |
| 834 | "context": format!("{} / {}", run.name, run.event), |
| 835 | "state": "pending", |
| 836 | "description": format!("{} is running", run.name), |
| 837 | "targetUrl": format!("{SITE}/{}/actions/runs/{}", run.repo, run.id), |
| 838 | }), |
| 839 | ) |
| 840 | .await; |
| 841 | Ok(()) |
| 842 | } |
| 843 | |
| 844 | /// Cancels a run: its waiting and queued jobs, and stops its running ones. |
| 845 | pub async fn cancel_run(&self, run: &RunRow, reason: &str) -> Result<()> { |
| 846 | self.db |
| 847 | .prepare("UPDATE runs SET conclusion = 'cancelled' WHERE id = ? AND status != 'completed'") |
| 848 | .bind(&[run.id.as_str().into()])? |
| 849 | .run() |
| 850 | .await?; |
| 851 | for job in self.job_rows(&run.id).await?.iter().filter(|job| job.status != "completed") { |
| 852 | self.stop_job(job, reason).await?; |
| 853 | } |
| 854 | if run.status == "pending" { |
| 855 | self.db.prepare("UPDATE runs SET status = 'queued' WHERE id = ?").bind(&[run.id.as_str().into()])?.run().await?; |
| 856 | } |
| 857 | self.advance(&run.id).await |
| 858 | } |
| 859 | |
| 860 | pub async fn cancel(&self, a: RunActionArgs) -> Result<Outcome<WorkflowRun>> { |
| 861 | if let Some(Outcome::Fail(refused)) = Self::member(&a.actor, &a.repo) { |
| 862 | return Ok(Outcome::Fail(refused)); |
| 863 | } |
| 864 | let run = check!(self.run_in(&a.repo, &a.id).await?); |
| 865 | if run.status == "completed" { |
| 866 | return Ok(fail(FailureCode::Conflict, "The run has already finished.")); |
| 867 | } |
| 868 | self.cancel_run(&run, &format!("{} cancelled the run.", a.actor.username)).await?; |
| 869 | self.run_summary(&run.id).await |
| 870 | } |
| 871 | |
| 872 | /// Runs again: every job, or with `failed_only` those that did not |
| 873 | /// succeed and the jobs that need them. |
| 874 | pub async fn rerun(&self, a: RunActionArgs) -> Result<Outcome<WorkflowRun>> { |
| 875 | if let Some(Outcome::Fail(refused)) = Self::member(&a.actor, &a.repo) { |
| 876 | return Ok(Outcome::Fail(refused)); |
| 877 | } |
| 878 | let run = check!(self.run_in(&a.repo, &a.id).await?); |
| 879 | if run.status != "completed" { |
| 880 | return Ok(fail(FailureCode::Conflict, "The run is still going: cancel it first.")); |
| 881 | } |
| 882 | if run.error.is_some() { |
| 883 | return Ok(fail(FailureCode::Conflict, "This run never started: fix the workflow file and push again.")); |
| 884 | } |
| 885 | let jobs = self.job_rows(&run.id).await?; |
| 886 | let workflow = workflow::parse(&run.source).ok(); |
| 887 | // Which keys run again: failed ones and, transitively, those needing them. |
| 888 | let mut again: Vec<String> = Vec::new(); |
| 889 | for key in workflow.as_ref().map(|w| w.job_order()).unwrap_or_default() { |
| 890 | let rows: Vec<&JobRow> = jobs.iter().filter(|j| j.key == key).collect(); |
| 891 | let failed = rows.iter().any(|row| row.conclusion.as_deref() != Some("success")); |
| 892 | let needs_again = rows.first().is_some_and(|row| row.needs().iter().any(|need| again.contains(need))); |
| 893 | if !a.failed_only || failed || needs_again { |
| 894 | again.push(key.to_owned()); |
| 895 | } |
| 896 | } |
| 897 | if again.is_empty() { |
| 898 | return Ok(fail(FailureCode::Conflict, "Every job succeeded: there is nothing to run again.")); |
| 899 | } |
| 900 | let mut statements = Vec::new(); |
| 901 | for key in &again { |
| 902 | statements.push(self.db.prepare("DELETE FROM logs WHERE job_id IN (SELECT id FROM jobs WHERE run_id = ? AND key = ?)").bind(&[run.id.as_str().into(), key.as_str().into()])?); |
| 903 | statements.push(self.db.prepare("DELETE FROM jobs WHERE run_id = ? AND key = ? AND ordinal > 0").bind(&[run.id.as_str().into(), key.as_str().into()])?); |
| 904 | statements.push( |
| 905 | self.db |
| 906 | .prepare( |
| 907 | "UPDATE jobs SET status = 'waiting', conclusion = NULL, steps = '[]', annotations = '[]', outputs = '{}', reason = NULL, |
| 908 | matrix = NULL, token_hash = NULL, seen_at = NULL, started_at = NULL, finished_at = NULL WHERE run_id = ? AND key = ?", |
| 909 | ) |
| 910 | .bind(&[run.id.as_str().into(), key.as_str().into()])?, |
| 911 | ); |
| 912 | } |
| 913 | statements.push( |
| 914 | self.db |
| 915 | .prepare("UPDATE runs SET status = 'queued', conclusion = NULL, attempt = attempt + 1, started_at = NULL, finished_at = NULL WHERE id = ?") |
| 916 | .bind(&[run.id.as_str().into()])?, |
| 917 | ); |
| 918 | self.db.batch(statements).await?; |
| 919 | if let Some(run) = self.run_row(&run.id).await? { |
| 920 | self.report_pending(&run).await?; |
| 921 | } |
| 922 | self.advance(&run.id).await?; |
| 923 | self.run_summary(&run.id).await |
| 924 | } |
| 925 | |
| 926 | pub async fn run_in(&self, repo: &RepoPath, id: &str) -> Result<Outcome<RunRow>> { |
| 927 | let row = self |
| 928 | .db |
| 929 | .prepare("SELECT * FROM runs WHERE id = ? AND lower(repo) = lower(?)") |
| 930 | .bind(&[id.into(), format!("{}/{}", repo.namespace, repo.name).into()])? |
| 931 | .first::<RunRow>(None) |
| 932 | .await?; |
| 933 | Ok(row.map_or_else(|| fail(FailureCode::NotFound, "No such run."), Outcome::Ok)) |
| 934 | } |
| 935 | |
| 936 | // --- The sandbox's side ----------------------------------------------------- |
| 937 | |
| 938 | async fn job_for_token(&self, a: &JobCallArgs) -> Result<Outcome<JobRow>> { |
| 939 | let job = self.db.prepare("SELECT * FROM jobs WHERE id = ?").bind(&[a.job.as_str().into()])?.first::<JobRow>(None).await?; |
| 940 | Ok(match job { |
| 941 | Some(job) if job.status == "in_progress" && job.token_hash.as_deref().is_some_and(|hash| same(hash, &sha256_hex(&a.token))) => { |
| 942 | Outcome::Ok(job) |
| 943 | } |
| 944 | _ => fail(FailureCode::Unauthenticated, "That job is not running, or the token is not its."), |
| 945 | }) |
| 946 | } |
| 947 | |
| 948 | /// `job_spec`: everything the sandbox needs to run the job. |
| 949 | pub async fn job_spec(&self, a: JobCallArgs) -> Result<Outcome<Value>> { |
| 950 | let job = check!(self.job_for_token(&a).await?); |
| 951 | let Some(run) = self.run_row(&job.run_id).await? else { |
| 952 | return Ok(fail(FailureCode::NotFound, "No such run.")); |
| 953 | }; |
| 954 | let Ok(workflow) = workflow::parse(&run.source) else { |
| 955 | return Ok(fail(FailureCode::Invalid, "The workflow no longer reads.")); |
| 956 | }; |
| 957 | let Some(spec) = workflow.jobs.iter().find(|j| j.id == job.key) else { |
| 958 | return Ok(fail(FailureCode::NotFound, "The job is not in the workflow.")); |
| 959 | }; |
| 960 | let repo = repo_path(&run.repo); |
| 961 | let trusted = run.trusted != 0; |
| 962 | // GITHUB_TOKEN: the workspace's, for as long as the job may run. |
| 963 | let token = if trusted { |
| 964 | match self.workspace_actor(&repo.namespace).await? { |
| 965 | Some(workspace) => { |
| 966 | let created: CreatedAccessToken = g1t_kit::call( |
| 967 | &self.identity, |
| 968 | "create_access_token", |
| 969 | &CreateAccessTokenArgs { |
| 970 | user: workspace, |
| 971 | name: format!("GITHUB_TOKEN for {} run {}", run.repo, run.number), |
| 972 | ttl_seconds: Some(u64::from(job.timeout_minutes) * 60 + 600), |
| 973 | }, |
| 974 | ) |
| 975 | .await?; |
| 976 | created.token |
| 977 | } |
| 978 | None => String::new(), |
| 979 | } |
| 980 | } else { |
| 981 | String::new() |
| 982 | }; |
| 983 | let mut secrets = if trusted { self.secrets_for(&run.repo_id, &repo.namespace).await? } else { Map::new() }; |
| 984 | secrets.insert("GITHUB_TOKEN".into(), Value::String(token.clone())); |
| 985 | let masks: Vec<String> = secrets.values().filter_map(|v| v.as_str()).filter(|v| v.len() >= 4).map(str::to_owned).collect(); |
| 986 | let vars = self.variables_for(&run.repo_id, &repo.namespace).await?; |
| 987 | |
| 988 | let jobs = self.job_rows(&run.id).await?; |
| 989 | let mut needs = Map::new(); |
| 990 | for need in &spec.needs { |
| 991 | let rows: Vec<&JobRow> = jobs.iter().filter(|row| &row.key == need).collect(); |
| 992 | let mut outputs = Map::new(); |
| 993 | for row in &rows { |
| 994 | if let Ok(Value::Object(more)) = serde_json::from_str::<Value>(&row.outputs) { |
| 995 | outputs.extend(more); |
| 996 | } |
| 997 | } |
| 998 | needs.insert(need.clone(), json!({ "result": key_result(&rows), "outputs": outputs })); |
| 999 | } |
| 1000 | let siblings = jobs.iter().filter(|row| row.key == job.key).count(); |
| 1001 | let matrix: Value = job.matrix.as_deref().and_then(|m| serde_json::from_str(m).ok()).unwrap_or(json!({})); |
| 1002 | let info = run.info(); |
| 1003 | let mut github = info.context(&job.key, &token, run.action.as_deref()); |
| 1004 | github["token"] = json!(token); |
| 1005 | |
| 1006 | // Where to check out: a pull request's fork, or the repository. |
| 1007 | let clone_url = match run.pull { |
| 1008 | Some(number) if run.event.starts_with("pull_request") && run.event != "pull_request_target" => { |
| 1009 | let located: Outcome<g1t_contracts::work::PullDetail> = g1t_kit::call( |
| 1010 | &self.work, |
| 1011 | "get_pull", |
| 1012 | &g1t_contracts::work::ViewArgs { |
| 1013 | repo: repo.clone(), |
| 1014 | number, |
| 1015 | viewer: self.workspace_actor(&repo.namespace).await?, |
| 1016 | after_seq: 0, |
| 1017 | }, |
| 1018 | ) |
| 1019 | .await?; |
| 1020 | match located { |
| 1021 | Outcome::Ok(detail) => match detail.pull.fork { |
| 1022 | Some(fork) => format!("{SITE}/{}/{}.git", fork.namespace, fork.name), |
| 1023 | None => format!("{SITE}/{}.git", run.repo), |
| 1024 | }, |
| 1025 | Outcome::Fail(_) => format!("{SITE}/{}.git", run.repo), |
| 1026 | } |
| 1027 | } |
| 1028 | _ => format!("{SITE}/{}.git", run.repo), |
| 1029 | }; |
| 1030 | |
| 1031 | Ok(Outcome::Ok(json!({ |
| 1032 | "job": job.id, |
| 1033 | "run": run.id, |
| 1034 | "key": job.key, |
| 1035 | "name": job.name, |
| 1036 | "spec": spec.raw, |
| 1037 | "workflow": { |
| 1038 | "env": workflow.env, |
| 1039 | "defaults": workflow.raw.get("defaults").cloned().unwrap_or(Value::Null), |
| 1040 | }, |
| 1041 | "github": github, |
| 1042 | "variables": info.variables(&job.key), |
| 1043 | "event": info.event, |
| 1044 | "contexts": { |
| 1045 | "vars": vars, |
| 1046 | "secrets": secrets, |
| 1047 | "inputs": run.inputs(), |
| 1048 | "matrix": matrix, |
| 1049 | "needs": needs, |
| 1050 | "strategy": { |
| 1051 | "fail-fast": spec.fail_fast, |
| 1052 | "job-index": job.ordinal, |
| 1053 | "job-total": siblings, |
| 1054 | "max-parallel": spec.max_parallel.unwrap_or(siblings as u32), |
| 1055 | }, |
| 1056 | "runner": runner_context(), |
| 1057 | }, |
| 1058 | "checkout": { |
| 1059 | "repository": run.repo, |
| 1060 | "url": clone_url, |
| 1061 | "sha": run.sha, |
| 1062 | "ref": run.git_ref, |
| 1063 | "token": token, |
| 1064 | }, |
| 1065 | "timeoutMinutes": job.timeout_minutes, |
| 1066 | "masks": masks, |
| 1067 | }))) |
| 1068 | } |
| 1069 | |
| 1070 | /// `job_report`: the sandbox telling how the job is going. |
| 1071 | pub async fn job_report(&self, a: JobCallArgs) -> Result<Outcome<Value>> { |
| 1072 | let job = check!(self.job_for_token(&a).await?); |
| 1073 | let report = &a.report; |
| 1074 | let at = now(); |
| 1075 | match report["kind"].as_str().unwrap_or_default() { |
| 1076 | "steps" => { |
| 1077 | // The list can grow as the job goes (post steps), so steps |
| 1078 | // already reported keep where they stand. |
| 1079 | let known: Vec<Value> = serde_json::from_str(&job.steps).unwrap_or_default(); |
| 1080 | let steps: Vec<Value> = report["steps"] |
| 1081 | .as_array() |
| 1082 | .map(|names| { |
| 1083 | names |
| 1084 | .iter() |
| 1085 | .enumerate() |
| 1086 | .map(|(i, name)| match known.get(i) { |
| 1087 | Some(step) if step["status"] != "queued" => step.clone(), |
| 1088 | _ => json!({ "number": i + 1, "name": expr::to_text(name), "status": "queued", "conclusion": null, "startedAt": null, "finishedAt": null }), |
| 1089 | }) |
| 1090 | .collect() |
| 1091 | }) |
| 1092 | .unwrap_or_default(); |
| 1093 | self.db |
| 1094 | .prepare("UPDATE jobs SET steps = ?, seen_at = ? WHERE id = ?") |
| 1095 | .bind(&[serde_json::to_string(&steps)?.into(), at.as_str().into(), job.id.as_str().into()])? |
| 1096 | .run() |
| 1097 | .await?; |
| 1098 | } |
| 1099 | "step" => { |
| 1100 | let number = report["number"].as_u64().unwrap_or(0) as usize; |
| 1101 | let mut steps: Vec<Value> = serde_json::from_str(&job.steps).unwrap_or_default(); |
| 1102 | if let Some(step) = number.checked_sub(1).and_then(|i| steps.get_mut(i)) { |
| 1103 | let status = report["status"].as_str().unwrap_or("in_progress"); |
| 1104 | step["status"] = json!(status); |
| 1105 | if status == "in_progress" { |
| 1106 | step["startedAt"] = json!(at); |
| 1107 | } |
| 1108 | if status == "completed" { |
| 1109 | step["finishedAt"] = json!(at); |
| 1110 | step["conclusion"] = report["conclusion"].clone(); |
| 1111 | } |
| 1112 | if let Some(name) = report["name"].as_str() { |
| 1113 | step["name"] = json!(name); |
| 1114 | } |
| 1115 | } |
| 1116 | self.db |
| 1117 | .prepare("UPDATE jobs SET steps = ?, seen_at = ? WHERE id = ?") |
| 1118 | .bind(&[serde_json::to_string(&steps)?.into(), at.as_str().into(), job.id.as_str().into()])? |
| 1119 | .run() |
| 1120 | .await?; |
| 1121 | } |
| 1122 | "log" => { |
| 1123 | let mut text = report["text"].as_str().unwrap_or_default().to_owned(); |
| 1124 | if text.len() > MAX_CHUNK_BYTES { |
| 1125 | let mut cut = MAX_CHUNK_BYTES; |
| 1126 | while !text.is_char_boundary(cut) { |
| 1127 | cut -= 1; |
| 1128 | } |
| 1129 | text.truncate(cut); |
| 1130 | } |
| 1131 | #[derive(Deserialize)] |
| 1132 | struct Size { |
| 1133 | n: Option<f64>, |
| 1134 | seq: Option<f64>, |
| 1135 | } |
| 1136 | let size = self |
| 1137 | .db |
| 1138 | .prepare("SELECT SUM(LENGTH(text)) AS n, MAX(seq) AS seq FROM logs WHERE job_id = ?") |
| 1139 | .bind(&[job.id.as_str().into()])? |
| 1140 | .first::<Size>(None) |
| 1141 | .await?; |
| 1142 | let (used, seq) = size.map_or((0, 0.0), |s| (s.n.unwrap_or(0.0) as usize, s.seq.unwrap_or(0.0))); |
| 1143 | if used < MAX_LOG_BYTES { |
| 1144 | if used + text.len() >= MAX_LOG_BYTES { |
| 1145 | text.push_str("\n… The log reached its limit of 4 MB; the rest is not kept.\n"); |
| 1146 | } |
| 1147 | self.db |
| 1148 | .prepare("INSERT INTO logs (job_id, seq, step, text) VALUES (?, ?, ?, ?)") |
| 1149 | .bind(&[job.id.as_str().into(), // Numbers go to D1 as f64: a u64 would be a BigInt, which it refuses. |
| 1150 | (seq + 1.0).into(), (report["step"].as_u64().unwrap_or(0) as u32).into(), text.into()])? |
| 1151 | .run() |
| 1152 | .await?; |
| 1153 | } |
| 1154 | self.db.prepare("UPDATE jobs SET seen_at = ? WHERE id = ?").bind(&[at.into(), job.id.as_str().into()])?.run().await?; |
| 1155 | } |
| 1156 | "annotation" => { |
| 1157 | let mut annotations: Vec<Value> = serde_json::from_str(&job.annotations).unwrap_or_default(); |
| 1158 | if annotations.len() < MAX_ANNOTATIONS { |
| 1159 | annotations.push(json!({ |
| 1160 | "level": report["level"].as_str().unwrap_or("notice"), |
| 1161 | "message": report["message"].as_str().unwrap_or_default().chars().take(4000).collect::<String>(), |
| 1162 | "title": report["title"], |
| 1163 | "file": report["file"], |
| 1164 | "line": report["line"], |
| 1165 | })); |
| 1166 | self.db |
| 1167 | .prepare("UPDATE jobs SET annotations = ?, seen_at = ? WHERE id = ?") |
| 1168 | .bind(&[serde_json::to_string(&annotations)?.into(), at.as_str().into(), job.id.as_str().into()])? |
| 1169 | .run() |
| 1170 | .await?; |
| 1171 | } |
| 1172 | } |
| 1173 | "done" => { |
| 1174 | let conclusion = report["conclusion"] |
| 1175 | .as_str() |
| 1176 | .filter(|c| matches!(*c, "success" | "failure" | "cancelled")) |
| 1177 | .unwrap_or("failure"); |
| 1178 | let outputs = report["outputs"].as_object().cloned(); |
| 1179 | Box::pin(self.finish_job(&job.id, conclusion, report["reason"].as_str(), outputs.as_ref())).await?; |
| 1180 | } |
| 1181 | other => return Ok(fail(FailureCode::Invalid, format!("There is no report called `{other}`."))), |
| 1182 | } |
| 1183 | Ok(Outcome::Ok(json!({ "ok": true }))) |
| 1184 | } |
| 1185 | |
| 1186 | // --- Every minute --------------------------------------------------------------- |
| 1187 | |
| 1188 | pub async fn on_minute(&self, now_ms: u64) -> Result<()> { |
| 1189 | let minute = now_ms / 60_000 * 60_000; |
| 1190 | if let Err(error) = self.run_schedules(minute).await { |
| 1191 | worker::console_error!("actions: schedules failed: {error}"); |
| 1192 | } |
| 1193 | // Jobs whose sandbox went quiet or ran past their time. |
| 1194 | let running = self.db.prepare("SELECT * FROM jobs WHERE status = 'in_progress'").all().await?.results::<JobRow>()?; |
| 1195 | for job in running { |
| 1196 | // Times in g1t's format compare as text. |
| 1197 | let before = |ms: u64| rfc3339(now_ms.saturating_sub(ms)); |
| 1198 | let silent = job.seen_at.as_deref().is_some_and(|seen| seen < before(SILENT_MS).as_str()); |
| 1199 | let limit = (u64::from(job.timeout_minutes) * 60 + 120) * 1000; |
| 1200 | let over = job.started_at.as_deref().is_some_and(|started| started < before(limit).as_str()); |
| 1201 | if over { |
| 1202 | let reason = format!("It ran longer than its time limit of {} minutes.", job.timeout_minutes); |
| 1203 | let _: Result<Value> = g1t_kit::call(&self.runner, "stop_actions_job", &json!({ "job": job.id })).await; |
| 1204 | self.finish_job(&job.id, "failure", Some(&reason), None).await?; |
| 1205 | } else if silent { |
| 1206 | self.finish_job(&job.id, "failure", Some("The runner stopped answering."), None).await?; |
| 1207 | } |
| 1208 | } |
| 1209 | self.start_queued().await |
| 1210 | } |
| 1211 | } |
| 1212 | |