| 1 |
use crate::error::Result; |
| 2 |
use crate::state::AppState; |
| 3 |
use axum::extract::{Path, Query, State, WebSocketUpgrade}; |
| 4 |
use axum::response::IntoResponse; |
| 5 |
use axum::routing::{get, post}; |
| 6 |
use axum::{Json, Router}; |
| 7 |
use serde::{Deserialize, Serialize}; |
| 8 |
use sqlx::Row; |
| 9 |
|
| 10 |
mod promotion; |
| 11 |
use promotion::{clear_partial, promote_inner, set_partial}; |
| 12 |
|
| 13 |
pub fn router(state: AppState) -> Router { |
| 14 |
let token = state.api_token.clone(); |
| 15 |
|
| 16 |
|
| 17 |
|
| 18 |
|
| 19 |
|
| 20 |
|
| 21 |
|
| 22 |
|
| 23 |
|
| 24 |
let protected = Router::new() |
| 25 |
|
| 26 |
.route("/promote/{tier}", post(promote)) |
| 27 |
.route("/rollback/{tier}", post(rollback)) |
| 28 |
.route("/rebuild", post(rebuild)) |
| 29 |
.route("/self-update", post(self_update)) |
| 30 |
.route("/confirm/{tier}", post(confirm)) |
| 31 |
.route("/backup/fetch", post(backup_fetch)) |
| 32 |
|
| 33 |
.route("/state", get(get_state)) |
| 34 |
.route("/status.json", get(get_status_json)) |
| 35 |
.route("/runs/{id}", get(get_run)) |
| 36 |
.route("/runs/{id}/wait", get(get_run_wait)) |
| 37 |
.route("/logs/{version}/{gate}", get(get_gate_log)) |
| 38 |
.route("/events", get(events_ws)) |
| 39 |
.route_layer(axum::middleware::from_fn(move |req, next| { |
| 40 |
require_bearer(token.clone(), req, next) |
| 41 |
})); |
| 42 |
|
| 43 |
Router::new().merge(protected).with_state(state) |
| 44 |
} |
| 45 |
|
| 46 |
|
| 47 |
|
| 48 |
|
| 49 |
async fn require_bearer( |
| 50 |
token: Option<std::sync::Arc<str>>, |
| 51 |
req: axum::extract::Request, |
| 52 |
next: axum::middleware::Next, |
| 53 |
) -> axum::response::Response { |
| 54 |
let Some(expected) = token.as_deref() else { |
| 55 |
return next.run(req).await; |
| 56 |
}; |
| 57 |
let path = req.uri().path().to_string(); |
| 58 |
let method = req.method().clone(); |
| 59 |
let ok = req |
| 60 |
.headers() |
| 61 |
.get(axum::http::header::AUTHORIZATION) |
| 62 |
.and_then(|h| h.to_str().ok()) |
| 63 |
.and_then(|h| h.strip_prefix("Bearer ")) |
| 64 |
.is_some_and(|t| ops_core::daemon::ct_eq(t, expected)); |
| 65 |
if ok { |
| 66 |
|
| 67 |
|
| 68 |
|
| 69 |
|
| 70 |
|
| 71 |
|
| 72 |
if method == axum::http::Method::POST { |
| 73 |
let peer = req |
| 74 |
.extensions() |
| 75 |
.get::<axum::extract::ConnectInfo<std::net::SocketAddr>>() |
| 76 |
.map_or_else(|| "unknown".into(), |ci| ci.0.to_string()); |
| 77 |
tracing::info!(path = %path, peer = %peer, "authenticated deploy mutation"); |
| 78 |
} |
| 79 |
next.run(req).await |
| 80 |
} else { |
| 81 |
tracing::warn!(path = %path, "rejected unauthenticated request"); |
| 82 |
( |
| 83 |
axum::http::StatusCode::UNAUTHORIZED, |
| 84 |
"missing or invalid bearer token\n", |
| 85 |
) |
| 86 |
.into_response() |
| 87 |
} |
| 88 |
} |
| 89 |
|
| 90 |
#[derive(Serialize)] |
| 91 |
pub(crate) struct StateView { |
| 92 |
|
| 93 |
|
| 94 |
|
| 95 |
pub(crate) sandod_version: &'static str, |
| 96 |
pub(crate) tiers: Vec<TierView>, |
| 97 |
|
| 98 |
|
| 99 |
|
| 100 |
|
| 101 |
|
| 102 |
pub(crate) build: Option<crate::runs::BuildSummary>, |
| 103 |
} |
| 104 |
|
| 105 |
#[derive(Serialize)] |
| 106 |
pub(crate) struct TierView { |
| 107 |
pub(crate) name: String, |
| 108 |
pub(crate) ord: i64, |
| 109 |
pub(crate) provisioned: bool, |
| 110 |
pub(crate) canary: String, |
| 111 |
pub(crate) current_version: Option<String>, |
| 112 |
pub(crate) previous_version: Option<String>, |
| 113 |
pub(crate) burn_in_started_at: Option<String>, |
| 114 |
|
| 115 |
|
| 116 |
|
| 117 |
pub(crate) partial_reason: Option<String>, |
| 118 |
pub(crate) nodes: Vec<String>, |
| 119 |
pub(crate) gates: Vec<GateView>, |
| 120 |
} |
| 121 |
|
| 122 |
#[derive(Serialize)] |
| 123 |
pub(crate) struct GateView { |
| 124 |
pub(crate) kind: String, |
| 125 |
pub(crate) finished_at: Option<String>, |
| 126 |
|
| 127 |
|
| 128 |
pub(crate) status: Option<String>, |
| 129 |
|
| 130 |
|
| 131 |
pub(crate) outcome: Option<serde_json::Value>, |
| 132 |
|
| 133 |
pub(crate) log_ref: Option<String>, |
| 134 |
} |
| 135 |
|
| 136 |
async fn get_state(State(s): State<AppState>) -> Result<Json<StateView>> { |
| 137 |
Ok(Json(state_view(&s).await?)) |
| 138 |
} |
| 139 |
|
| 140 |
|
| 141 |
|
| 142 |
async fn get_status_json(State(s): State<AppState>) -> Result<Json<ops_status::Payload>> { |
| 143 |
let view = state_view(&s).await?; |
| 144 |
Ok(Json(crate::status::payload(&view, chrono::Utc::now()))) |
| 145 |
} |
| 146 |
|
| 147 |
|
| 148 |
pub(crate) async fn state_view(s: &AppState) -> Result<StateView> { |
| 149 |
let rows = sqlx::query( |
| 150 |
"SELECT t.name, t.ord, t.provisioned, t.canary, |
| 151 |
ts.current_version, ts.previous_version, ts.burn_in_started_at, |
| 152 |
ts.partial_reason |
| 153 |
FROM tiers t |
| 154 |
LEFT JOIN tier_state ts ON ts.tier = t.name |
| 155 |
ORDER BY t.ord", |
| 156 |
) |
| 157 |
.fetch_all(&s.pool) |
| 158 |
.await?; |
| 159 |
|
| 160 |
let mut tiers: Vec<TierView> = Vec::with_capacity(rows.len()); |
| 161 |
for r in rows { |
| 162 |
let name: String = r.get("name"); |
| 163 |
let current_version: Option<String> = r.get("current_version"); |
| 164 |
|
| 165 |
let nodes: Vec<String> = |
| 166 |
sqlx::query_scalar("SELECT name FROM nodes WHERE tier = ? ORDER BY name") |
| 167 |
.bind(&name) |
| 168 |
.fetch_all(&s.pool) |
| 169 |
.await?; |
| 170 |
|
| 171 |
|
| 172 |
|
| 173 |
|
| 174 |
|
| 175 |
|
| 176 |
let gate_version: Option<String> = if current_version.is_some() { |
| 177 |
current_version.clone() |
| 178 |
} else { |
| 179 |
sqlx::query_scalar( |
| 180 |
"SELECT version FROM gate_runs WHERE tier = ? |
| 181 |
ORDER BY id DESC LIMIT 1", |
| 182 |
) |
| 183 |
.bind(&name) |
| 184 |
.fetch_optional(&s.pool) |
| 185 |
.await? |
| 186 |
}; |
| 187 |
|
| 188 |
let gates: Vec<GateView> = if let Some(ver) = gate_version.as_ref() { |
| 189 |
|
| 190 |
sqlx::query( |
| 191 |
"SELECT gate_kind, finished_at, status, outcome_json, log_ref |
| 192 |
FROM gate_runs g |
| 193 |
WHERE tier = ?1 AND version = ?2 |
| 194 |
AND id = (SELECT MAX(id) FROM gate_runs |
| 195 |
WHERE tier = ?1 AND version = ?2 AND gate_kind = g.gate_kind) |
| 196 |
ORDER BY gate_kind", |
| 197 |
) |
| 198 |
.bind(&name) |
| 199 |
.bind(ver) |
| 200 |
.fetch_all(&s.pool) |
| 201 |
.await? |
| 202 |
.into_iter() |
| 203 |
.map(|gr| GateView { |
| 204 |
kind: gr.get("gate_kind"), |
| 205 |
finished_at: gr.get("finished_at"), |
| 206 |
status: gr.get("status"), |
| 207 |
outcome: gr |
| 208 |
.get::<Option<String>, _>("outcome_json") |
| 209 |
.and_then(|s| serde_json::from_str(&s).ok()), |
| 210 |
log_ref: gr.get("log_ref"), |
| 211 |
}) |
| 212 |
.collect() |
| 213 |
} else { |
| 214 |
Vec::new() |
| 215 |
}; |
| 216 |
|
| 217 |
tiers.push(TierView { |
| 218 |
name, |
| 219 |
ord: r.get("ord"), |
| 220 |
provisioned: r.get::<i64, _>("provisioned") != 0, |
| 221 |
canary: r.get("canary"), |
| 222 |
current_version, |
| 223 |
previous_version: r.get("previous_version"), |
| 224 |
burn_in_started_at: r.get("burn_in_started_at"), |
| 225 |
partial_reason: r.get("partial_reason"), |
| 226 |
nodes, |
| 227 |
gates, |
| 228 |
}); |
| 229 |
} |
| 230 |
|
| 231 |
let build = crate::runs::latest_summary(&s.pool).await?; |
| 232 |
Ok(StateView { |
| 233 |
sandod_version: env!("CARGO_PKG_VERSION"), |
| 234 |
tiers, |
| 235 |
build, |
| 236 |
}) |
| 237 |
} |
| 238 |
|
| 239 |
#[derive(Deserialize, Default)] |
| 240 |
struct PromoteBody { |
| 241 |
|
| 242 |
|
| 243 |
#[serde(default)] |
| 244 |
version: Option<String>, |
| 245 |
#[serde(default)] |
| 246 |
hotfix: bool, |
| 247 |
#[serde(default)] |
| 248 |
reset_burn_in: bool, |
| 249 |
|
| 250 |
|
| 251 |
|
| 252 |
|
| 253 |
|
| 254 |
|
| 255 |
|
| 256 |
#[serde(default)] |
| 257 |
bears_migration: bool, |
| 258 |
} |
| 259 |
|
| 260 |
async fn promote( |
| 261 |
State(s): State<AppState>, |
| 262 |
Path(tier): Path<String>, |
| 263 |
body: Option<Json<PromoteBody>>, |
| 264 |
) -> Result<Json<serde_json::Value>> { |
| 265 |
let body = body.map(|Json(b)| b).unwrap_or_default(); |
| 266 |
|
| 267 |
|
| 268 |
|
| 269 |
|
| 270 |
|
| 271 |
|
| 272 |
|
| 273 |
|
| 274 |
|
| 275 |
tokio::spawn(promote_inner(s, tier, body)) |
| 276 |
.await |
| 277 |
.map_err(|e| { |
| 278 |
crate::error::Error::Other(anyhow::anyhow!("promote task failed to join: {e}")) |
| 279 |
})? |
| 280 |
} |
| 281 |
|
| 282 |
async fn rollback( |
| 283 |
State(s): State<AppState>, |
| 284 |
Path(tier): Path<String>, |
| 285 |
) -> Result<Json<serde_json::Value>> { |
| 286 |
|
| 287 |
let _deploy_guard = s.deploy_lock.lock().await; |
| 288 |
let tier = crate::domain::TierId::new(tier); |
| 289 |
let target = s |
| 290 |
.topo |
| 291 |
.tiers |
| 292 |
.iter() |
| 293 |
.find(|t| t.name == tier) |
| 294 |
.ok_or(crate::error::Error::NotFound)?; |
| 295 |
|
| 296 |
let row: Option<(Option<String>, Option<String>)> = |
| 297 |
sqlx::query_as("SELECT current_version, previous_version FROM tier_state WHERE tier = ?") |
| 298 |
.bind(&tier) |
| 299 |
.fetch_optional(&s.pool) |
| 300 |
.await |
| 301 |
.map_err(crate::error::Error::Db)?; |
| 302 |
let (Some(current_str), Some(previous_str)) = row.unwrap_or((None, None)) else { |
| 303 |
return Err(crate::error::Error::GateBlocked( |
| 304 |
"no previous_version to roll back to".into(), |
| 305 |
)); |
| 306 |
}; |
| 307 |
let current = crate::domain::Version::parse(¤t_str) |
| 308 |
.map_err(|e| crate::error::Error::Other(anyhow::anyhow!(e)))?; |
| 309 |
let previous = crate::domain::Version::parse(&previous_str) |
| 310 |
.map_err(|e| crate::error::Error::Other(anyhow::anyhow!(e)))?; |
| 311 |
|
| 312 |
let bin: Option<(String,)> = |
| 313 |
sqlx::query_as("SELECT artifact_path FROM versions WHERE version = ?") |
| 314 |
.bind(&previous) |
| 315 |
.fetch_optional(&s.pool) |
| 316 |
.await |
| 317 |
.map_err(crate::error::Error::Db)?; |
| 318 |
let Some((bin,)) = bin else { |
| 319 |
return Err(crate::error::Error::GateBlocked(format!( |
| 320 |
"previous version {previous} has no artifact_path; rollback impossible" |
| 321 |
))); |
| 322 |
}; |
| 323 |
let bin_path = std::path::PathBuf::from(bin); |
| 324 |
let staged_dir = bin_path |
| 325 |
.parent() |
| 326 |
.ok_or_else(|| crate::error::Error::Other(anyhow::anyhow!("artifact_path has no parent")))? |
| 327 |
.to_path_buf(); |
| 328 |
|
| 329 |
|
| 330 |
|
| 331 |
|
| 332 |
let total = target.nodes.len(); |
| 333 |
for (i, node) in target.nodes.iter().enumerate() { |
| 334 |
let executor = s |
| 335 |
.executors |
| 336 |
.get(&node.name) |
| 337 |
.cloned() |
| 338 |
.unwrap_or_else(|| crate::state::build_executor(node)); |
| 339 |
if let Err(e) = crate::deploy::deploy_node( |
| 340 |
executor.as_ref(), |
| 341 |
node, |
| 342 |
&previous_str, |
| 343 |
&staged_dir, |
| 344 |
s.cfg.primary_bin(), |
| 345 |
) |
| 346 |
.await |
| 347 |
{ |
| 348 |
set_partial( |
| 349 |
&s, |
| 350 |
&tier, |
| 351 |
&format!( |
| 352 |
"rollback to {previous} incomplete: {i}/{total} nodes rolled back, \ |
| 353 |
node {} failed; remaining nodes still on {current} — manual check needed", |
| 354 |
node.name, |
| 355 |
), |
| 356 |
) |
| 357 |
.await; |
| 358 |
tracing::error!(tier = %tier, node = %node.name, done = i, of = total, |
| 359 |
to = %previous, "rollback failed mid-fleet; tier left partial"); |
| 360 |
return Err(crate::error::Error::Other(e)); |
| 361 |
} |
| 362 |
} |
| 363 |
|
| 364 |
|
| 365 |
|
| 366 |
|
| 367 |
|
| 368 |
|
| 369 |
|
| 370 |
|
| 371 |
|
| 372 |
|
| 373 |
|
| 374 |
|
| 375 |
|
| 376 |
let now = chrono::Utc::now().to_rfc3339(); |
| 377 |
sqlx::query( |
| 378 |
"UPDATE tier_state |
| 379 |
SET current_version = ?, previous_version = NULL, |
| 380 |
burn_in_started_at = NULL, advanced_at = ? |
| 381 |
WHERE tier = ?", |
| 382 |
) |
| 383 |
.bind(&previous) |
| 384 |
.bind(&now) |
| 385 |
.bind(&tier) |
| 386 |
.execute(&s.pool) |
| 387 |
.await |
| 388 |
.map_err(crate::error::Error::Db)?; |
| 389 |
|
| 390 |
|
| 391 |
clear_partial(&s, &tier).await; |
| 392 |
|
| 393 |
tracing::warn!(tier = %tier, from = %current, to = %previous, "rollback complete"); |
| 394 |
crate::events::emit( |
| 395 |
&s.events, |
| 396 |
crate::events::Event::Rollback { |
| 397 |
tier: tier.clone(), |
| 398 |
from: current.clone(), |
| 399 |
to: previous.clone(), |
| 400 |
}, |
| 401 |
); |
| 402 |
|
| 403 |
Ok(Json(serde_json::json!({ |
| 404 |
"tier": tier, |
| 405 |
"rolled_back_from": current, |
| 406 |
"now_running": previous, |
| 407 |
}))) |
| 408 |
} |
| 409 |
|
| 410 |
#[derive(Deserialize, Default)] |
| 411 |
struct RebuildBody { |
| 412 |
|
| 413 |
#[serde(default)] |
| 414 |
sha: Option<String>, |
| 415 |
} |
| 416 |
|
| 417 |
async fn rebuild( |
| 418 |
State(s): State<AppState>, |
| 419 |
body: Option<Json<RebuildBody>>, |
| 420 |
) -> Result<Json<serde_json::Value>> { |
| 421 |
let body = body.map(|Json(b)| b).unwrap_or_default(); |
| 422 |
let sha = match body.sha { |
| 423 |
Some(sha) => sha, |
| 424 |
None => { |
| 425 |
|
| 426 |
|
| 427 |
|
| 428 |
|
| 429 |
|
| 430 |
|
| 431 |
let bare = std::path::Path::new(&s.topo.repo.bare_path); |
| 432 |
if let Some(upstream) = s.topo.repo.upstream.as_deref() |
| 433 |
&& let Err(e) = |
| 434 |
crate::git::fetch_upstream(bare, upstream, &s.topo.repo.branch).await |
| 435 |
{ |
| 436 |
tracing::warn!(error = %e, "pre-resolve upstream fetch failed; resolving current bare-repo branch tip"); |
| 437 |
} |
| 438 |
crate::git::resolve_ref(bare, &s.topo.repo.branch) |
| 439 |
.await |
| 440 |
.map_err(crate::error::Error::Other)? |
| 441 |
} |
| 442 |
}; |
| 443 |
|
| 444 |
|
| 445 |
|
| 446 |
let sha = crate::domain::GitSha::parse(&sha) |
| 447 |
.map_err(|e| crate::error::Error::Other(anyhow::anyhow!(e)))?; |
| 448 |
|
| 449 |
tracing::info!(sha = %sha, "rebuild requested"); |
| 450 |
crate::events::emit( |
| 451 |
&s.events, |
| 452 |
crate::events::Event::RebuildRequested { sha: sha.clone() }, |
| 453 |
); |
| 454 |
|
| 455 |
|
| 456 |
|
| 457 |
|
| 458 |
let run_id = crate::runs::create(&s.pool, sha.as_str()) |
| 459 |
.await |
| 460 |
.map_err(crate::error::Error::Other)?; |
| 461 |
|
| 462 |
|
| 463 |
|
| 464 |
|
| 465 |
|
| 466 |
let mut slot = s.active_build.lock().await; |
| 467 |
if let Some(prev) = slot.take() |
| 468 |
&& !prev.handle.is_finished() |
| 469 |
{ |
| 470 |
tracing::warn!("aborting in-flight build for newer /rebuild request"); |
| 471 |
crate::events::emit( |
| 472 |
&s.events, |
| 473 |
crate::events::Event::BuildAborted { |
| 474 |
sha_aborted: sha.clone(), |
| 475 |
}, |
| 476 |
); |
| 477 |
prev.handle.abort(); |
| 478 |
|
| 479 |
|
| 480 |
crate::runs::mark_aborted(&s.pool, prev.run_id).await.ok(); |
| 481 |
} |
| 482 |
|
| 483 |
let pool = s.pool.clone(); |
| 484 |
let cfg = s.cfg.clone(); |
| 485 |
let topo = s.topo.clone(); |
| 486 |
let events_for_task = s.events.clone(); |
| 487 |
let sha_for_task = sha.clone(); |
| 488 |
let sha_response = sha.to_string(); |
| 489 |
let pool_for_task = s.pool.clone(); |
| 490 |
let deploy_lock = s.deploy_lock.clone(); |
| 491 |
let handle = tokio::spawn(async move { |
| 492 |
if let Err(e) = crate::build::build_and_run_host( |
| 493 |
pool, |
| 494 |
cfg, |
| 495 |
topo, |
| 496 |
sha_for_task.clone(), |
| 497 |
events_for_task, |
| 498 |
run_id, |
| 499 |
deploy_lock, |
| 500 |
) |
| 501 |
.await |
| 502 |
{ |
| 503 |
tracing::error!(sha = %sha_for_task, error = %e, "rebuild pipeline failed"); |
| 504 |
|
| 505 |
|
| 506 |
|
| 507 |
crate::runs::mark_failed(&pool_for_task, run_id, &format!("{e:#}")) |
| 508 |
.await |
| 509 |
.ok(); |
| 510 |
} |
| 511 |
}); |
| 512 |
*slot = Some(crate::state::ActiveBuild { |
| 513 |
handle: handle.abort_handle(), |
| 514 |
run_id, |
| 515 |
}); |
| 516 |
|
| 517 |
Ok(Json( |
| 518 |
serde_json::json!({ "accepted": true, "sha": sha_response, "run_id": run_id.0 }), |
| 519 |
)) |
| 520 |
} |
| 521 |
|
| 522 |
|
| 523 |
|
| 524 |
async fn get_run( |
| 525 |
State(s): State<AppState>, |
| 526 |
Path(id): Path<i64>, |
| 527 |
) -> Result<Json<crate::runs::RunView>> { |
| 528 |
crate::runs::get(&s.pool, crate::domain::RunId(id)) |
| 529 |
.await |
| 530 |
.map_err(crate::error::Error::Other)? |
| 531 |
.map(Json) |
| 532 |
.ok_or(crate::error::Error::NotFound) |
| 533 |
} |
| 534 |
|
| 535 |
#[derive(Deserialize)] |
| 536 |
struct WaitParams { |
| 537 |
|
| 538 |
|
| 539 |
#[serde(default)] |
| 540 |
timeout_ms: Option<u64>, |
| 541 |
} |
| 542 |
|
| 543 |
|
| 544 |
|
| 545 |
|
| 546 |
|
| 547 |
|
| 548 |
async fn get_run_wait( |
| 549 |
State(s): State<AppState>, |
| 550 |
Path(id): Path<i64>, |
| 551 |
Query(p): Query<WaitParams>, |
| 552 |
) -> Result<Json<crate::runs::RunView>> { |
| 553 |
let run_id = crate::domain::RunId(id); |
| 554 |
let timeout = std::time::Duration::from_millis(p.timeout_ms.unwrap_or(30_000).min(120_000)); |
| 555 |
let deadline = tokio::time::Instant::now() + timeout; |
| 556 |
|
| 557 |
|
| 558 |
|
| 559 |
|
| 560 |
let tick = std::time::Duration::from_millis(750); |
| 561 |
loop { |
| 562 |
let view = crate::runs::get(&s.pool, run_id) |
| 563 |
.await |
| 564 |
.map_err(crate::error::Error::Other)? |
| 565 |
.ok_or(crate::error::Error::NotFound)?; |
| 566 |
let now = tokio::time::Instant::now(); |
| 567 |
if view.result != "building" || now >= deadline { |
| 568 |
return Ok(Json(view)); |
| 569 |
} |
| 570 |
tokio::time::sleep((deadline - now).min(tick)).await; |
| 571 |
} |
| 572 |
} |
| 573 |
|
| 574 |
#[derive(Deserialize)] |
| 575 |
struct SelfUpdateBody { |
| 576 |
|
| 577 |
|
| 578 |
sha: String, |
| 579 |
} |
| 580 |
|
| 581 |
|
| 582 |
|
| 583 |
fn self_update_unit(sha: &crate::domain::GitSha) -> String { |
| 584 |
format!("sando-update@{sha}.service") |
| 585 |
} |
| 586 |
|
| 587 |
|
| 588 |
|
| 589 |
|
| 590 |
|
| 591 |
|
| 592 |
|
| 593 |
|
| 594 |
|
| 595 |
async fn self_update( |
| 596 |
State(s): State<AppState>, |
| 597 |
crate::error::TypedBody(body): crate::error::TypedBody<SelfUpdateBody>, |
| 598 |
) -> Result<Json<serde_json::Value>> { |
| 599 |
let sha = crate::domain::GitSha::parse(&body.sha) |
| 600 |
.map_err(|e| crate::error::Error::BadRequest(format!("invalid sha: {e}")))?; |
| 601 |
|
| 602 |
let unit = self_update_unit(&sha); |
| 603 |
|
| 604 |
|
| 605 |
|
| 606 |
|
| 607 |
|
| 608 |
|
| 609 |
|
| 610 |
{ |
| 611 |
let slot = s.active_build.lock().await; |
| 612 |
if slot.as_ref().is_some_and(|b| !b.handle.is_finished()) { |
| 613 |
return Err(crate::error::Error::GateBlocked( |
| 614 |
"a server build is in flight; retry /self-update once it settles".into(), |
| 615 |
)); |
| 616 |
} |
| 617 |
|
| 618 |
|
| 619 |
|
| 620 |
|
| 621 |
if s.deploy_lock.try_lock().is_err() { |
| 622 |
return Err(crate::error::Error::GateBlocked( |
| 623 |
"a promote/rollback is in flight; retry /self-update once it settles".into(), |
| 624 |
)); |
| 625 |
} |
| 626 |
|
| 627 |
tracing::warn!(sha = %sha, unit = %unit, "self-update requested; triggering privileged updater"); |
| 628 |
|
| 629 |
|
| 630 |
|
| 631 |
let status = tokio::process::Command::new("systemctl") |
| 632 |
.args(["start", "--no-block", &unit]) |
| 633 |
.status() |
| 634 |
.await |
| 635 |
.map_err(|e| crate::error::Error::Other(anyhow::anyhow!("spawning systemctl: {e}")))?; |
| 636 |
if !status.success() { |
| 637 |
return Err(crate::error::Error::Other(anyhow::anyhow!( |
| 638 |
"systemctl start {unit} exited {status}; is sando-update@.service installed and the sando-user polkit rule in place?" |
| 639 |
))); |
| 640 |
} |
| 641 |
} |
| 642 |
Ok(Json( |
| 643 |
serde_json::json!({ "accepted": true, "sha": sha.to_string(), "unit": unit }), |
| 644 |
)) |
| 645 |
} |
| 646 |
|
| 647 |
async fn confirm( |
| 648 |
State(s): State<AppState>, |
| 649 |
Path(tier): Path<String>, |
| 650 |
) -> Result<Json<serde_json::Value>> { |
| 651 |
|
| 652 |
|
| 653 |
|
| 654 |
|
| 655 |
|
| 656 |
|
| 657 |
|
| 658 |
let _deploy_guard = s.deploy_lock.lock().await; |
| 659 |
let tier = crate::domain::TierId::new(tier); |
| 660 |
let target = s |
| 661 |
.topo |
| 662 |
.tiers |
| 663 |
.iter() |
| 664 |
.find(|t| t.name == tier) |
| 665 |
.ok_or(crate::error::Error::NotFound)?; |
| 666 |
|
| 667 |
let version_str: Option<String> = |
| 668 |
sqlx::query_scalar("SELECT current_version FROM tier_state WHERE tier = ?") |
| 669 |
.bind(&target.name) |
| 670 |
.fetch_optional(&s.pool) |
| 671 |
.await |
| 672 |
.map_err(crate::error::Error::Db)? |
| 673 |
.flatten(); |
| 674 |
let version_str = version_str.ok_or_else(|| { |
| 675 |
crate::error::Error::GateBlocked(format!( |
| 676 |
"tier {tier} has no current_version; nothing to confirm" |
| 677 |
)) |
| 678 |
})?; |
| 679 |
let version = crate::domain::Version::parse(&version_str) |
| 680 |
.map_err(|e| crate::error::Error::Other(anyhow::anyhow!(e)))?; |
| 681 |
|
| 682 |
let now = chrono::Utc::now().to_rfc3339(); |
| 683 |
let outcome = |
| 684 |
crate::outcome::GateOutcome::passed(crate::outcome::PassNote::OperatorConfirmed { |
| 685 |
at: chrono::Utc::now(), |
| 686 |
}); |
| 687 |
let outcome_json = serde_json::to_string(&outcome) |
| 688 |
.map_err(|e| crate::error::Error::Other(anyhow::anyhow!(e)))?; |
| 689 |
sqlx::query( |
| 690 |
"INSERT INTO gate_runs (version, tier, gate_kind, started_at, finished_at, status, outcome_json) |
| 691 |
VALUES (?, ?, 'manual_confirm', ?, ?, 'passed', ?)", |
| 692 |
) |
| 693 |
.bind(&version).bind(&target.name).bind(&now).bind(&now).bind(&outcome_json) |
| 694 |
.execute(&s.pool).await.map_err(crate::error::Error::Db)?; |
| 695 |
|
| 696 |
tracing::info!(tier = %tier, version = %version, "manual_confirm recorded"); |
| 697 |
crate::events::emit( |
| 698 |
&s.events, |
| 699 |
crate::events::Event::ManualConfirm { |
| 700 |
tier: tier.clone(), |
| 701 |
version: version.clone(), |
| 702 |
}, |
| 703 |
); |
| 704 |
|
| 705 |
Ok(Json( |
| 706 |
serde_json::json!({ "tier": tier, "version": version }), |
| 707 |
)) |
| 708 |
} |
| 709 |
|
| 710 |
async fn backup_fetch(State(s): State<AppState>) -> Result<Json<serde_json::Value>> { |
| 711 |
let fb = crate::backup::fetch(&s.pool, &s.cfg, &s.topo) |
| 712 |
.await |
| 713 |
.map_err(crate::error::Error::Other)?; |
| 714 |
crate::events::emit( |
| 715 |
&s.events, |
| 716 |
crate::events::Event::BackupFetched { |
| 717 |
source: fb.source.clone(), |
| 718 |
byte_size: fb.byte_size.unwrap_or(0), |
| 719 |
}, |
| 720 |
); |
| 721 |
Ok(Json(serde_json::json!({ |
| 722 |
"source": fb.source, |
| 723 |
"local_path": fb.local_path, |
| 724 |
"byte_size": fb.byte_size, |
| 725 |
}))) |
| 726 |
} |
| 727 |
|
| 728 |
async fn get_gate_log( |
| 729 |
State(s): State<AppState>, |
| 730 |
Path((version, gate)): Path<(String, String)>, |
| 731 |
) -> Result<axum::response::Response> { |
| 732 |
|
| 733 |
|
| 734 |
|
| 735 |
fn safe(seg: &str) -> bool { |
| 736 |
!seg.is_empty() && !seg.contains('/') && !seg.contains('\\') && seg != "." && seg != ".." |
| 737 |
} |
| 738 |
if !safe(&version) { |
| 739 |
return Err(crate::error::Error::NotFound); |
| 740 |
} |
| 741 |
|
| 742 |
|
| 743 |
|
| 744 |
|
| 745 |
let kind: crate::domain::GateKind = gate.parse().map_err(|_| crate::error::Error::NotFound)?; |
| 746 |
let path = s |
| 747 |
.cfg |
| 748 |
.logs_root |
| 749 |
.join(&version) |
| 750 |
.join(format!("{}.log", kind.as_str())); |
| 751 |
|
| 752 |
|
| 753 |
|
| 754 |
const MAX_LOG_BYTES: u64 = 8 * 1024 * 1024; |
| 755 |
use tokio::io::{AsyncReadExt, AsyncSeekExt}; |
| 756 |
let mut file = match tokio::fs::File::open(&path).await { |
| 757 |
Ok(f) => f, |
| 758 |
Err(e) if e.kind() == std::io::ErrorKind::NotFound => { |
| 759 |
return Err(crate::error::Error::NotFound); |
| 760 |
} |
| 761 |
Err(e) => return Err(crate::error::Error::Other(e.into())), |
| 762 |
}; |
| 763 |
let len = file |
| 764 |
.metadata() |
| 765 |
.await |
| 766 |
.map_err(|e| crate::error::Error::Other(e.into()))? |
| 767 |
.len(); |
| 768 |
let body: Vec<u8> = if len <= MAX_LOG_BYTES { |
| 769 |
let mut buf = Vec::with_capacity(len as usize); |
| 770 |
file.read_to_end(&mut buf) |
| 771 |
.await |
| 772 |
.map_err(|e| crate::error::Error::Other(e.into()))?; |
| 773 |
buf |
| 774 |
} else { |
| 775 |
file.seek(std::io::SeekFrom::End(-(MAX_LOG_BYTES as i64))) |
| 776 |
.await |
| 777 |
.map_err(|e| crate::error::Error::Other(e.into()))?; |
| 778 |
let mut buf = format!( |
| 779 |
"[log truncated: showing the last {} MiB of {len} bytes]\n", |
| 780 |
MAX_LOG_BYTES / 1024 / 1024, |
| 781 |
) |
| 782 |
.into_bytes(); |
| 783 |
file.read_to_end(&mut buf) |
| 784 |
.await |
| 785 |
.map_err(|e| crate::error::Error::Other(e.into()))?; |
| 786 |
buf |
| 787 |
}; |
| 788 |
Ok(( |
| 789 |
[( |
| 790 |
axum::http::header::CONTENT_TYPE, |
| 791 |
"text/plain; charset=utf-8", |
| 792 |
)], |
| 793 |
body, |
| 794 |
) |
| 795 |
.into_response()) |
| 796 |
} |
| 797 |
|
| 798 |
async fn events_ws(ws: WebSocketUpgrade, State(s): State<AppState>) -> impl IntoResponse { |
| 799 |
use axum::extract::ws::Message; |
| 800 |
use tokio::sync::broadcast::error::RecvError; |
| 801 |
|
| 802 |
ws.on_upgrade(move |mut socket| async move { |
| 803 |
|
| 804 |
|
| 805 |
|
| 806 |
|
| 807 |
let mut status_rx = s.events.subscribe_status(); |
| 808 |
let mut logs_rx = s.events.subscribe_logs(); |
| 809 |
loop { |
| 810 |
let recv = tokio::select! { |
| 811 |
r = status_rx.recv() => r, |
| 812 |
r = logs_rx.recv() => r, |
| 813 |
}; |
| 814 |
match recv { |
| 815 |
Ok(env) => { |
| 816 |
let json = match serde_json::to_string(&env) { |
| 817 |
Ok(s) => s, |
| 818 |
Err(e) => { |
| 819 |
tracing::warn!(error = %e, "events ws: serialize failed"); |
| 820 |
continue; |
| 821 |
} |
| 822 |
}; |
| 823 |
if socket.send(Message::Text(json.into())).await.is_err() { |
| 824 |
break; |
| 825 |
} |
| 826 |
} |
| 827 |
Err(RecvError::Lagged(n)) => { |
| 828 |
let _ = socket |
| 829 |
.send(Message::Text( |
| 830 |
format!(r#"{{"kind":"lagged","skipped":{n}}}"#).into(), |
| 831 |
)) |
| 832 |
.await; |
| 833 |
} |
| 834 |
Err(RecvError::Closed) => break, |
| 835 |
} |
| 836 |
} |
| 837 |
}) |
| 838 |
} |
| 839 |
|
| 840 |
#[cfg(test)] |
| 841 |
mod tests { |
| 842 |
use super::promotion::{rollback_deployed_nodes, unsatisfied_gates}; |
| 843 |
use super::*; |
| 844 |
use crate::config::Config; |
| 845 |
use crate::topology::{BackupConfig, CanaryPolicy, Gate, Node, RepoConfig, Tier, Topology}; |
| 846 |
use async_trait::async_trait; |
| 847 |
use axum::body::Body; |
| 848 |
use axum::http::{Request, StatusCode}; |
| 849 |
use http_body_util::BodyExt; |
| 850 |
use ops_exec::{CapabilitySet, Executor, LogSink, RunOutput, Step, SyncOpts}; |
| 851 |
use sqlx::SqlitePool; |
| 852 |
use sqlx::sqlite::SqlitePoolOptions; |
| 853 |
use std::os::unix::process::ExitStatusExt; |
| 854 |
use std::path::PathBuf; |
| 855 |
use std::sync::{Arc, Mutex as StdMutex}; |
| 856 |
use tower::ServiceExt; |
| 857 |
|
| 858 |
async fn fresh_pool() -> SqlitePool { |
| 859 |
let pool = SqlitePoolOptions::new() |
| 860 |
.max_connections(1) |
| 861 |
.connect("sqlite::memory:") |
| 862 |
.await |
| 863 |
.unwrap(); |
| 864 |
sqlx::migrate!("./migrations").run(&pool).await.unwrap(); |
| 865 |
pool |
| 866 |
} |
| 867 |
|
| 868 |
|
| 869 |
|
| 870 |
|
| 871 |
fn test_topo() -> Topology { |
| 872 |
Topology { |
| 873 |
repo: RepoConfig { |
| 874 |
bare_path: "/tmp/test.git".into(), |
| 875 |
branch: "main".into(), |
| 876 |
upstream: None, |
| 877 |
}, |
| 878 |
backup: BackupConfig { |
| 879 |
source: "file:///tmp/test-backup.sql".into(), |
| 880 |
local_path: "/tmp/local-backup.sql".into(), |
| 881 |
}, |
| 882 |
tiers: vec![ |
| 883 |
Tier { |
| 884 |
name: "host".into(), |
| 885 |
provisioned: true, |
| 886 |
gates: vec![], |
| 887 |
canary: CanaryPolicy::Sequential, |
| 888 |
nodes: vec![], |
| 889 |
}, |
| 890 |
Tier { |
| 891 |
name: "a".into(), |
| 892 |
provisioned: true, |
| 893 |
gates: vec![Gate::BootSmoke], |
| 894 |
canary: CanaryPolicy::Sequential, |
| 895 |
nodes: vec![Node { |
| 896 |
name: "a-local".into(), |
| 897 |
ssh_target: "local".into(), |
| 898 |
release_root: "/tmp/a-node".into(), |
| 899 |
service_name: "makenotwork.service".into(), |
| 900 |
health_url: None, |
| 901 |
config_check_env_file: None, |
| 902 |
actuate: crate::topology::default_actuate(), |
| 903 |
observe: crate::topology::default_observe(), |
| 904 |
companions: Vec::new(), |
| 905 |
}], |
| 906 |
}, |
| 907 |
], |
| 908 |
} |
| 909 |
} |
| 910 |
|
| 911 |
fn test_cfg() -> Config { |
| 912 |
Config { |
| 913 |
listen: "127.0.0.1:0".into(), |
| 914 |
db_path: PathBuf::from(":memory:"), |
| 915 |
topology_path: PathBuf::from("/tmp/test-sando.toml"), |
| 916 |
build_host: "test-host".into(), |
| 917 |
workdir: PathBuf::from("/tmp/sando-work"), |
| 918 |
release_root: PathBuf::from("/tmp/sando-releases"), |
| 919 |
scratch_db_url: None, |
| 920 |
scratch_owner_role: "makenotwork".into(), |
| 921 |
boot_smoke_port: 18181, |
| 922 |
code_smoke_port: 18182, |
| 923 |
bin_names: vec!["makenotwork".into()], |
| 924 |
logs_root: PathBuf::from("/tmp/sando-logs"), |
| 925 |
release_contents: vec![], |
| 926 |
cargo_target_dir: None, |
| 927 |
gate_timeout_secs: 2400, |
| 928 |
companions: Vec::new(), |
| 929 |
test_targets: vec![crate::config::TestTarget { |
| 930 |
dir: PathBuf::from("server"), |
| 931 |
features: vec!["fast-tests".into()], |
| 932 |
all_features: false, |
| 933 |
scratch_db: true, |
| 934 |
}], |
| 935 |
} |
| 936 |
} |
| 937 |
|
| 938 |
async fn test_state() -> AppState { |
| 939 |
let pool = fresh_pool().await; |
| 940 |
|
| 941 |
for (i, name) in ["host", "a"].iter().enumerate() { |
| 942 |
sqlx::query( |
| 943 |
"INSERT INTO tiers (name, ord, provisioned, canary) VALUES (?, ?, 1, 'sequential')", |
| 944 |
) |
| 945 |
.bind(name) |
| 946 |
.bind(i as i64) |
| 947 |
.execute(&pool) |
| 948 |
.await |
| 949 |
.unwrap(); |
| 950 |
sqlx::query("INSERT INTO tier_state (tier) VALUES (?)") |
| 951 |
.bind(name) |
| 952 |
.execute(&pool) |
| 953 |
.await |
| 954 |
.unwrap(); |
| 955 |
} |
| 956 |
|
| 957 |
|
| 958 |
let topo = test_topo(); |
| 959 |
let executors = Arc::new(crate::state::build_executors(&topo)); |
| 960 |
AppState { |
| 961 |
pool, |
| 962 |
topo: Arc::new(topo), |
| 963 |
cfg: Arc::new(test_cfg()), |
| 964 |
active_build: Arc::new(tokio::sync::Mutex::new(None)), |
| 965 |
deploy_lock: Arc::new(tokio::sync::Mutex::new(())), |
| 966 |
events: crate::events::channel(), |
| 967 |
executors, |
| 968 |
api_token: None, |
| 969 |
} |
| 970 |
} |
| 971 |
|
| 972 |
async fn body_string(resp: axum::response::Response) -> String { |
| 973 |
let bytes = resp.into_body().collect().await.unwrap().to_bytes(); |
| 974 |
String::from_utf8(bytes.to_vec()).unwrap() |
| 975 |
} |
| 976 |
|
| 977 |
|
| 978 |
async fn seed(pool: &SqlitePool, tier: &str, version: &str) { |
| 979 |
sqlx::query("INSERT INTO tiers (name, ord, provisioned, canary) VALUES (?, 0, 1, 'sequential') ON CONFLICT DO NOTHING") |
| 980 |
.bind(tier).execute(pool).await.unwrap(); |
| 981 |
sqlx::query("INSERT INTO versions (version, git_sha, built_at, artifact_path) VALUES (?, 'sha', datetime('now'), '/tmp/x') ON CONFLICT DO NOTHING") |
| 982 |
.bind(version).execute(pool).await.unwrap(); |
| 983 |
sqlx::query("INSERT INTO tier_state (tier, current_version) VALUES (?, NULL) ON CONFLICT DO NOTHING") |
| 984 |
.bind(tier).execute(pool).await.unwrap(); |
| 985 |
} |
| 986 |
|
| 987 |
async fn insert_gate(pool: &SqlitePool, tier: &str, version: &str, kind: &str, passed: i64) { |
| 988 |
let status = if passed == 1 { "passed" } else { "failed" }; |
| 989 |
sqlx::query( |
| 990 |
"INSERT INTO gate_runs (version, tier, gate_kind, started_at, finished_at, status) \ |
| 991 |
VALUES (?, ?, ?, datetime('now'), datetime('now'), ?)", |
| 992 |
) |
| 993 |
.bind(version) |
| 994 |
.bind(tier) |
| 995 |
.bind(kind) |
| 996 |
.bind(status) |
| 997 |
.execute(pool) |
| 998 |
.await |
| 999 |
.unwrap(); |
| 1000 |
} |
| 1001 |
|
| 1002 |
|
| 1003 |
|
| 1004 |
fn tid(s: &str) -> crate::domain::TierId { |
| 1005 |
crate::domain::TierId::new(s) |
| 1006 |
} |
| 1007 |
|
| 1008 |
#[tokio::test] |
| 1009 |
async fn unsatisfied_gates_empty_when_no_configured_gates() { |
| 1010 |
|
| 1011 |
let pool = fresh_pool().await; |
| 1012 |
seed(&pool, "host", "0.8.12").await; |
| 1013 |
let pending = unsatisfied_gates(&pool, &tid("host"), &[], "0.8.12", None, false) |
| 1014 |
.await |
| 1015 |
.unwrap(); |
| 1016 |
assert_eq!(pending, Vec::<String>::new()); |
| 1017 |
} |
| 1018 |
|
| 1019 |
#[tokio::test] |
| 1020 |
async fn unsatisfied_gates_flags_configured_gate_that_never_ran() { |
| 1021 |
|
| 1022 |
|
| 1023 |
|
| 1024 |
|
| 1025 |
let pool = fresh_pool().await; |
| 1026 |
seed(&pool, "a", "0.8.12").await; |
| 1027 |
let pending = |
| 1028 |
unsatisfied_gates(&pool, &tid("a"), &[Gate::BootSmoke], "0.8.12", None, false) |
| 1029 |
.await |
| 1030 |
.unwrap(); |
| 1031 |
assert_eq!(pending, vec!["boot_smoke".to_string()]); |
| 1032 |
} |
| 1033 |
|
| 1034 |
#[tokio::test] |
| 1035 |
async fn unsatisfied_gates_flags_failed_kind() { |
| 1036 |
let pool = fresh_pool().await; |
| 1037 |
seed(&pool, "host", "0.8.12").await; |
| 1038 |
insert_gate(&pool, "host", "0.8.12", "cargo_test", 0).await; |
| 1039 |
insert_gate(&pool, "host", "0.8.12", "boot_smoke", 1).await; |
| 1040 |
let pending = unsatisfied_gates( |
| 1041 |
&pool, |
| 1042 |
&tid("host"), |
| 1043 |
&[Gate::CargoTest, Gate::BootSmoke], |
| 1044 |
"0.8.12", |
| 1045 |
None, |
| 1046 |
false, |
| 1047 |
) |
| 1048 |
.await |
| 1049 |
.unwrap(); |
| 1050 |
assert_eq!(pending, vec!["cargo_test".to_string()]); |
| 1051 |
} |
| 1052 |
|
| 1053 |
#[tokio::test] |
| 1054 |
async fn unsatisfied_gates_latest_row_wins() { |
| 1055 |
|
| 1056 |
|
| 1057 |
let pool = fresh_pool().await; |
| 1058 |
seed(&pool, "host", "0.8.12").await; |
| 1059 |
insert_gate(&pool, "host", "0.8.12", "cargo_test", 0).await; |
| 1060 |
insert_gate(&pool, "host", "0.8.12", "cargo_test", 1).await; |
| 1061 |
let pending = unsatisfied_gates( |
| 1062 |
&pool, |
| 1063 |
&tid("host"), |
| 1064 |
&[Gate::CargoTest], |
| 1065 |
"0.8.12", |
| 1066 |
None, |
| 1067 |
false, |
| 1068 |
) |
| 1069 |
.await |
| 1070 |
.unwrap(); |
| 1071 |
assert!(pending.is_empty()); |
| 1072 |
} |
| 1073 |
|
| 1074 |
async fn insert_confirm( |
| 1075 |
pool: &SqlitePool, |
| 1076 |
tier: &str, |
| 1077 |
version: &str, |
| 1078 |
at: chrono::DateTime<chrono::Utc>, |
| 1079 |
) { |
| 1080 |
sqlx::query( |
| 1081 |
"INSERT INTO gate_runs (version, tier, gate_kind, started_at, finished_at, status) \ |
| 1082 |
VALUES (?, ?, 'manual_confirm', ?, ?, 'passed')", |
| 1083 |
) |
| 1084 |
.bind(version) |
| 1085 |
.bind(tier) |
| 1086 |
.bind(at.to_rfc3339()) |
| 1087 |
.bind(at.to_rfc3339()) |
| 1088 |
.execute(pool) |
| 1089 |
.await |
| 1090 |
.unwrap(); |
| 1091 |
} |
| 1092 |
|
| 1093 |
#[tokio::test] |
| 1094 |
async fn unsatisfied_gates_manual_confirm_requires_fresh_confirmation() { |
| 1095 |
|
| 1096 |
|
| 1097 |
|
| 1098 |
let pool = fresh_pool().await; |
| 1099 |
seed(&pool, "a", "0.8.12").await; |
| 1100 |
let landed = chrono::Utc::now(); |
| 1101 |
sqlx::query("UPDATE tier_state SET burn_in_started_at = ? WHERE tier = 'a'") |
| 1102 |
.bind(landed.to_rfc3339()) |
| 1103 |
.execute(&pool) |
| 1104 |
.await |
| 1105 |
.unwrap(); |
| 1106 |
|
| 1107 |
|
| 1108 |
insert_confirm(&pool, "a", "0.8.12", landed - chrono::Duration::hours(1)).await; |
| 1109 |
let pending = unsatisfied_gates( |
| 1110 |
&pool, |
| 1111 |
&tid("a"), |
| 1112 |
&[Gate::ManualConfirm], |
| 1113 |
"0.8.12", |
| 1114 |
None, |
| 1115 |
false, |
| 1116 |
) |
| 1117 |
.await |
| 1118 |
.unwrap(); |
| 1119 |
assert_eq!( |
| 1120 |
pending, |
| 1121 |
vec!["manual_confirm".to_string()], |
| 1122 |
"stale confirm must not satisfy" |
| 1123 |
); |
| 1124 |
|
| 1125 |
|
| 1126 |
insert_confirm(&pool, "a", "0.8.12", landed + chrono::Duration::minutes(5)).await; |
| 1127 |
let pending = unsatisfied_gates( |
| 1128 |
&pool, |
| 1129 |
&tid("a"), |
| 1130 |
&[Gate::ManualConfirm], |
| 1131 |
"0.8.12", |
| 1132 |
None, |
| 1133 |
false, |
| 1134 |
) |
| 1135 |
.await |
| 1136 |
.unwrap(); |
| 1137 |
assert!(pending.is_empty(), "fresh confirm satisfies"); |
| 1138 |
} |
| 1139 |
|
| 1140 |
#[tokio::test] |
| 1141 |
async fn unsatisfied_gates_manual_confirm_fails_closed_without_baseline() { |
| 1142 |
|
| 1143 |
|
| 1144 |
let pool = fresh_pool().await; |
| 1145 |
seed(&pool, "a", "0.8.12").await; |
| 1146 |
insert_confirm(&pool, "a", "0.8.12", chrono::Utc::now()).await; |
| 1147 |
let pending = unsatisfied_gates( |
| 1148 |
&pool, |
| 1149 |
&tid("a"), |
| 1150 |
&[Gate::ManualConfirm], |
| 1151 |
"0.8.12", |
| 1152 |
None, |
| 1153 |
false, |
| 1154 |
) |
| 1155 |
.await |
| 1156 |
.unwrap(); |
| 1157 |
assert_eq!( |
| 1158 |
pending, |
| 1159 |
vec!["manual_confirm".to_string()], |
| 1160 |
"no baseline -> fail closed" |
| 1161 |
); |
| 1162 |
} |
| 1163 |
|
| 1164 |
#[tokio::test] |
| 1165 |
async fn unsatisfied_gates_hotfix_skips_only_burn_in() { |
| 1166 |
|
| 1167 |
|
| 1168 |
|
| 1169 |
|
| 1170 |
let pool = fresh_pool().await; |
| 1171 |
seed(&pool, "a", "0.8.12").await; |
| 1172 |
insert_gate(&pool, "a", "0.8.12", "cargo_test", 0).await; |
| 1173 |
let gates = [Gate::BurnIn { hours: 48 }, Gate::CargoTest]; |
| 1174 |
|
| 1175 |
let normal = unsatisfied_gates(&pool, &tid("a"), &gates, "0.8.12", None, false) |
| 1176 |
.await |
| 1177 |
.unwrap(); |
| 1178 |
assert_eq!( |
| 1179 |
normal, |
| 1180 |
vec!["burn_in".to_string(), "cargo_test".to_string()] |
| 1181 |
); |
| 1182 |
|
| 1183 |
let with_hotfix = unsatisfied_gates(&pool, &tid("a"), &gates, "0.8.12", None, true) |
| 1184 |
.await |
| 1185 |
.unwrap(); |
| 1186 |
assert_eq!(with_hotfix, vec!["cargo_test".to_string()]); |
| 1187 |
} |
| 1188 |
|
| 1189 |
#[tokio::test] |
| 1190 |
async fn unsatisfied_gates_burn_in_passes_when_window_elapsed() { |
| 1191 |
|
| 1192 |
|
| 1193 |
let pool = fresh_pool().await; |
| 1194 |
seed(&pool, "a", "0.8.12").await; |
| 1195 |
sqlx::query("UPDATE tier_state SET burn_in_started_at = ? WHERE tier = 'a'") |
| 1196 |
.bind((chrono::Utc::now() - chrono::Duration::hours(50)).to_rfc3339()) |
| 1197 |
.execute(&pool) |
| 1198 |
.await |
| 1199 |
.unwrap(); |
| 1200 |
let pending = unsatisfied_gates( |
| 1201 |
&pool, |
| 1202 |
&tid("a"), |
| 1203 |
&[Gate::BurnIn { hours: 48 }], |
| 1204 |
"0.8.12", |
| 1205 |
None, |
| 1206 |
false, |
| 1207 |
) |
| 1208 |
.await |
| 1209 |
.unwrap(); |
| 1210 |
assert!(pending.is_empty(), "50h elapsed satisfies a 48h burn-in"); |
| 1211 |
} |
| 1212 |
|
| 1213 |
#[tokio::test] |
| 1214 |
async fn unsatisfied_gates_ignores_other_tiers_and_versions() { |
| 1215 |
let pool = fresh_pool().await; |
| 1216 |
seed(&pool, "host", "0.8.12").await; |
| 1217 |
seed(&pool, "host", "0.8.11").await; |
| 1218 |
seed(&pool, "a", "0.8.12").await; |
| 1219 |
|
| 1220 |
|
| 1221 |
insert_gate(&pool, "host", "0.8.12", "cargo_test", 0).await; |
| 1222 |
insert_gate(&pool, "a", "0.8.12", "cargo_test", 0).await; |
| 1223 |
insert_gate(&pool, "host", "0.8.11", "cargo_test", 0).await; |
| 1224 |
|
| 1225 |
let pending = unsatisfied_gates( |
| 1226 |
&pool, |
| 1227 |
&tid("host"), |
| 1228 |
&[Gate::CargoTest], |
| 1229 |
"0.8.12", |
| 1230 |
None, |
| 1231 |
false, |
| 1232 |
) |
| 1233 |
.await |
| 1234 |
.unwrap(); |
| 1235 |
assert_eq!(pending, vec!["cargo_test".to_string()]); |
| 1236 |
} |
| 1237 |
|
| 1238 |
#[tokio::test] |
| 1239 |
async fn unsatisfied_gates_null_status_is_treated_as_failing() { |
| 1240 |
|
| 1241 |
|
| 1242 |
|
| 1243 |
let pool = fresh_pool().await; |
| 1244 |
seed(&pool, "host", "0.8.12").await; |
| 1245 |
sqlx::query( |
| 1246 |
"INSERT INTO gate_runs (version, tier, gate_kind, started_at) \ |
| 1247 |
VALUES ('0.8.12', 'host', 'cargo_test', datetime('now'))", |
| 1248 |
) |
| 1249 |
.execute(&pool) |
| 1250 |
.await |
| 1251 |
.unwrap(); |
| 1252 |
|
| 1253 |
let pending = unsatisfied_gates( |
| 1254 |
&pool, |
| 1255 |
&tid("host"), |
| 1256 |
&[Gate::CargoTest], |
| 1257 |
"0.8.12", |
| 1258 |
None, |
| 1259 |
false, |
| 1260 |
) |
| 1261 |
.await |
| 1262 |
.unwrap(); |
| 1263 |
assert_eq!(pending, vec!["cargo_test".to_string()]); |
| 1264 |
} |
| 1265 |
|
| 1266 |
|
| 1267 |
|
| 1268 |
#[tokio::test] |
| 1269 |
async fn confirm_rejects_when_tier_has_no_current_version() { |
| 1270 |
|
| 1271 |
|
| 1272 |
let state = test_state().await; |
| 1273 |
let app = router(state.clone()); |
| 1274 |
let resp = app |
| 1275 |
.oneshot( |
| 1276 |
Request::builder() |
| 1277 |
.method("POST") |
| 1278 |
.uri("/confirm/a") |
| 1279 |
.body(Body::empty()) |
| 1280 |
.unwrap(), |
| 1281 |
) |
| 1282 |
.await |
| 1283 |
.unwrap(); |
| 1284 |
assert_eq!(resp.status(), StatusCode::CONFLICT); |
| 1285 |
let body = body_string(resp).await; |
| 1286 |
assert!(body.contains("no current_version"), "got: {body}"); |
| 1287 |
} |
| 1288 |
|
| 1289 |
#[tokio::test] |
| 1290 |
async fn confirm_accepts_when_current_version_set_and_inserts_row() { |
| 1291 |
let state = test_state().await; |
| 1292 |
|
| 1293 |
sqlx::query("INSERT INTO versions (version, git_sha, built_at, artifact_path) VALUES ('0.8.12','sha',datetime('now'),'/tmp/x')") |
| 1294 |
.execute(&state.pool).await.unwrap(); |
| 1295 |
sqlx::query("UPDATE tier_state SET current_version = '0.8.12' WHERE tier = 'a'") |
| 1296 |
.execute(&state.pool) |
| 1297 |
.await |
| 1298 |
.unwrap(); |
| 1299 |
|
| 1300 |
let app = router(state.clone()); |
| 1301 |
let resp = app |
| 1302 |
.oneshot( |
| 1303 |
Request::builder() |
| 1304 |
.method("POST") |
| 1305 |
.uri("/confirm/a") |
| 1306 |
.body(Body::empty()) |
| 1307 |
.unwrap(), |
| 1308 |
) |
| 1309 |
.await |
| 1310 |
.unwrap(); |
| 1311 |
assert_eq!(resp.status(), StatusCode::OK); |
| 1312 |
let body = body_string(resp).await; |
| 1313 |
assert!(body.contains("\"tier\":\"a\"")); |
| 1314 |
assert!(body.contains("\"version\":\"0.8.12\"")); |
| 1315 |
|
| 1316 |
|
| 1317 |
let count: (i64,) = sqlx::query_as( |
| 1318 |
"SELECT COUNT(*) FROM gate_runs WHERE tier='a' AND gate_kind='manual_confirm' AND status='passed'", |
| 1319 |
) |
| 1320 |
.fetch_one(&state.pool) |
| 1321 |
.await |
| 1322 |
.unwrap(); |
| 1323 |
assert_eq!(count.0, 1); |
| 1324 |
} |
| 1325 |
|
| 1326 |
#[tokio::test] |
| 1327 |
async fn confirm_404s_for_unknown_tier() { |
| 1328 |
let state = test_state().await; |
| 1329 |
let app = router(state); |
| 1330 |
let resp = app |
| 1331 |
.oneshot( |
| 1332 |
Request::builder() |
| 1333 |
.method("POST") |
| 1334 |
.uri("/confirm/zzzz") |
| 1335 |
.body(Body::empty()) |
| 1336 |
.unwrap(), |
| 1337 |
) |
| 1338 |
.await |
| 1339 |
.unwrap(); |
| 1340 |
assert_eq!(resp.status(), StatusCode::NOT_FOUND); |
| 1341 |
} |
| 1342 |
|
| 1343 |
#[tokio::test] |
| 1344 |
async fn get_run_404s_for_unknown_id() { |
| 1345 |
let state = test_state().await; |
| 1346 |
let app = router(state); |
| 1347 |
let resp = app |
| 1348 |
.oneshot( |
| 1349 |
Request::builder() |
| 1350 |
.uri("/runs/999") |
| 1351 |
.body(Body::empty()) |
| 1352 |
.unwrap(), |
| 1353 |
) |
| 1354 |
.await |
| 1355 |
.unwrap(); |
| 1356 |
assert_eq!(resp.status(), StatusCode::NOT_FOUND); |
| 1357 |
} |
| 1358 |
|
| 1359 |
#[tokio::test] |
| 1360 |
async fn get_run_returns_view_with_gates() { |
| 1361 |
let state = test_state().await; |
| 1362 |
|
| 1363 |
let run_id = crate::runs::create(&state.pool, "abc1234def") |
| 1364 |
.await |
| 1365 |
.unwrap(); |
| 1366 |
let ver: crate::domain::Version = "0.10.2".parse().unwrap(); |
| 1367 |
seed(&state.pool, "host", "0.10.2").await; |
| 1368 |
crate::runs::set_version(&state.pool, run_id, &ver) |
| 1369 |
.await |
| 1370 |
.unwrap(); |
| 1371 |
insert_gate(&state.pool, "host", "0.10.2", "cargo_test", 0).await; |
| 1372 |
insert_gate(&state.pool, "host", "0.10.2", "boot_smoke", 1).await; |
| 1373 |
|
| 1374 |
let app = router(state); |
| 1375 |
let resp = app |
| 1376 |
.oneshot( |
| 1377 |
Request::builder() |
| 1378 |
.uri(format!("/runs/{}", run_id.0)) |
| 1379 |
.body(Body::empty()) |
| 1380 |
.unwrap(), |
| 1381 |
) |
| 1382 |
.await |
| 1383 |
.unwrap(); |
| 1384 |
assert_eq!(resp.status(), StatusCode::OK); |
| 1385 |
let v: serde_json::Value = serde_json::from_str(&body_string(resp).await).unwrap(); |
| 1386 |
assert_eq!(v["run_id"], run_id.0); |
| 1387 |
assert_eq!(v["sha"], "abc1234def"); |
| 1388 |
assert_eq!(v["version"], "0.10.2"); |
| 1389 |
assert_eq!(v["result"], "building"); |
| 1390 |
|
| 1391 |
assert_eq!(v["gates"].as_array().unwrap().len(), 2); |
| 1392 |
assert_eq!(v["gates"][0]["kind"], "boot_smoke"); |
| 1393 |
assert_eq!(v["gates"][0]["status"], "passed"); |
| 1394 |
assert_eq!(v["gates"][1]["kind"], "cargo_test"); |
| 1395 |
assert_eq!(v["gates"][1]["status"], "failed"); |
| 1396 |
} |
| 1397 |
|
| 1398 |
#[tokio::test] |
| 1399 |
async fn get_run_wait_returns_immediately_when_settled() { |
| 1400 |
let state = test_state().await; |
| 1401 |
let run_id = crate::runs::create(&state.pool, "abc1234def") |
| 1402 |
.await |
| 1403 |
.unwrap(); |
| 1404 |
crate::runs::mark_passed(&state.pool, run_id).await.unwrap(); |
| 1405 |
|
| 1406 |
let app = router(state); |
| 1407 |
|
| 1408 |
let resp = app |
| 1409 |
.oneshot( |
| 1410 |
Request::builder() |
| 1411 |
.uri(format!("/runs/{}/wait?timeout_ms=60000", run_id.0)) |
| 1412 |
.body(Body::empty()) |
| 1413 |
.unwrap(), |
| 1414 |
) |
| 1415 |
.await |
| 1416 |
.unwrap(); |
| 1417 |
assert_eq!(resp.status(), StatusCode::OK); |
| 1418 |
let v: serde_json::Value = serde_json::from_str(&body_string(resp).await).unwrap(); |
| 1419 |
assert_eq!(v["result"], "passed"); |
| 1420 |
} |
| 1421 |
|
| 1422 |
#[tokio::test] |
| 1423 |
async fn get_run_wait_returns_building_at_timeout() { |
| 1424 |
let state = test_state().await; |
| 1425 |
let run_id = crate::runs::create(&state.pool, "abc1234def") |
| 1426 |
.await |
| 1427 |
.unwrap(); |
| 1428 |
|
| 1429 |
let app = router(state); |
| 1430 |
|
| 1431 |
|
| 1432 |
let resp = app |
| 1433 |
.oneshot( |
| 1434 |
Request::builder() |
| 1435 |
.uri(format!("/runs/{}/wait?timeout_ms=0", run_id.0)) |
| 1436 |
.body(Body::empty()) |
| 1437 |
.unwrap(), |
| 1438 |
) |
| 1439 |
.await |
| 1440 |
.unwrap(); |
| 1441 |
assert_eq!(resp.status(), StatusCode::OK); |
| 1442 |
let v: serde_json::Value = serde_json::from_str(&body_string(resp).await).unwrap(); |
| 1443 |
assert_eq!(v["result"], "building"); |
| 1444 |
} |
| 1445 |
|
| 1446 |
#[tokio::test] |
| 1447 |
async fn get_run_wait_404s_for_unknown_id() { |
| 1448 |
let state = test_state().await; |
| 1449 |
let app = router(state); |
| 1450 |
let resp = app |
| 1451 |
.oneshot( |
| 1452 |
Request::builder() |
| 1453 |
.uri("/runs/999/wait?timeout_ms=0") |
| 1454 |
.body(Body::empty()) |
| 1455 |
.unwrap(), |
| 1456 |
) |
| 1457 |
.await |
| 1458 |
.unwrap(); |
| 1459 |
assert_eq!(resp.status(), StatusCode::NOT_FOUND); |
| 1460 |
} |
| 1461 |
|
| 1462 |
#[test] |
| 1463 |
fn self_update_unit_maps_sha_to_instance() { |
| 1464 |
let sha = crate::domain::GitSha::parse("abc1234def5678").unwrap(); |
| 1465 |
assert_eq!( |
| 1466 |
self_update_unit(&sha), |
| 1467 |
"sando-update@abc1234def5678.service" |
| 1468 |
); |
| 1469 |
} |
| 1470 |
|
| 1471 |
#[tokio::test] |
| 1472 |
async fn self_update_rejects_bad_sha_with_400() { |
| 1473 |
|
| 1474 |
|
| 1475 |
let state = test_state().await; |
| 1476 |
let app = router(state); |
| 1477 |
let resp = app |
| 1478 |
.oneshot( |
| 1479 |
Request::builder() |
| 1480 |
.method("POST") |
| 1481 |
.uri("/self-update") |
| 1482 |
.header("Content-Type", "application/json") |
| 1483 |
.body(Body::from(r#"{"sha":"not-a-sha!"}"#)) |
| 1484 |
.unwrap(), |
| 1485 |
) |
| 1486 |
.await |
| 1487 |
.unwrap(); |
| 1488 |
assert_eq!(resp.status(), StatusCode::BAD_REQUEST); |
| 1489 |
} |
| 1490 |
|
| 1491 |
|
| 1492 |
|
| 1493 |
#[tokio::test] |
| 1494 |
async fn promote_to_first_tier_is_rejected() { |
| 1495 |
|
| 1496 |
let state = test_state().await; |
| 1497 |
let app = router(state); |
| 1498 |
let resp = app |
| 1499 |
.oneshot( |
| 1500 |
Request::builder() |
| 1501 |
.method("POST") |
| 1502 |
.uri("/promote/host") |
| 1503 |
.body(Body::empty()) |
| 1504 |
.unwrap(), |
| 1505 |
) |
| 1506 |
.await |
| 1507 |
.unwrap(); |
| 1508 |
assert_eq!(resp.status(), StatusCode::CONFLICT); |
| 1509 |
let body = body_string(resp).await; |
| 1510 |
assert!( |
| 1511 |
body.contains("cannot /promote to the first tier"), |
| 1512 |
"got: {body}" |
| 1513 |
); |
| 1514 |
} |
| 1515 |
|
| 1516 |
#[tokio::test] |
| 1517 |
async fn promote_without_body_and_no_predecessor_version_errors() { |
| 1518 |
|
| 1519 |
|
| 1520 |
let state = test_state().await; |
| 1521 |
let app = router(state); |
| 1522 |
let resp = app |
| 1523 |
.oneshot( |
| 1524 |
Request::builder() |
| 1525 |
.method("POST") |
| 1526 |
.uri("/promote/a") |
| 1527 |
.body(Body::empty()) |
| 1528 |
.unwrap(), |
| 1529 |
) |
| 1530 |
.await |
| 1531 |
.unwrap(); |
| 1532 |
assert_eq!(resp.status(), StatusCode::CONFLICT); |
| 1533 |
let body = body_string(resp).await; |
| 1534 |
assert!( |
| 1535 |
body.contains("no version specified") || body.contains("no current_version"), |
| 1536 |
"got: {body}" |
| 1537 |
); |
| 1538 |
} |
| 1539 |
|
| 1540 |
#[tokio::test] |
| 1541 |
async fn promote_blocked_when_predecessor_gate_never_ran() { |
| 1542 |
|
| 1543 |
|
| 1544 |
|
| 1545 |
|
| 1546 |
let pool = fresh_pool().await; |
| 1547 |
for (i, name) in ["host", "a"].iter().enumerate() { |
| 1548 |
sqlx::query( |
| 1549 |
"INSERT INTO tiers (name, ord, provisioned, canary) VALUES (?, ?, 1, 'sequential')", |
| 1550 |
) |
| 1551 |
.bind(name) |
| 1552 |
.bind(i as i64) |
| 1553 |
.execute(&pool) |
| 1554 |
.await |
| 1555 |
.unwrap(); |
| 1556 |
sqlx::query("INSERT INTO tier_state (tier) VALUES (?)") |
| 1557 |
.bind(name) |
| 1558 |
.execute(&pool) |
| 1559 |
.await |
| 1560 |
.unwrap(); |
| 1561 |
} |
| 1562 |
let mut topo = test_topo(); |
| 1563 |
topo.tiers[0].gates = vec![Gate::BootSmoke]; |
| 1564 |
let executors = Arc::new(crate::state::build_executors(&topo)); |
| 1565 |
let state = AppState { |
| 1566 |
pool, |
| 1567 |
topo: Arc::new(topo), |
| 1568 |
cfg: Arc::new(test_cfg()), |
| 1569 |
active_build: Arc::new(tokio::sync::Mutex::new(None)), |
| 1570 |
deploy_lock: Arc::new(tokio::sync::Mutex::new(())), |
| 1571 |
events: crate::events::channel(), |
| 1572 |
executors, |
| 1573 |
api_token: None, |
| 1574 |
}; |
| 1575 |
sqlx::query("INSERT INTO versions (version, git_sha, built_at, artifact_path) VALUES ('0.8.12','sha',datetime('now'),'/tmp/x')") |
| 1576 |
.execute(&state.pool).await.unwrap(); |
| 1577 |
sqlx::query("UPDATE tier_state SET current_version = '0.8.12' WHERE tier = 'host'") |
| 1578 |
.execute(&state.pool) |
| 1579 |
.await |
| 1580 |
.unwrap(); |
| 1581 |
|
| 1582 |
let app = router(state); |
| 1583 |
let resp = app |
| 1584 |
.oneshot( |
| 1585 |
Request::builder() |
| 1586 |
.method("POST") |
| 1587 |
.uri("/promote/a") |
| 1588 |
.body(Body::empty()) |
| 1589 |
.unwrap(), |
| 1590 |
) |
| 1591 |
.await |
| 1592 |
.unwrap(); |
| 1593 |
assert_eq!(resp.status(), StatusCode::CONFLICT); |
| 1594 |
let body = body_string(resp).await; |
| 1595 |
assert!( |
| 1596 |
body.contains("boot_smoke"), |
| 1597 |
"expected boot_smoke to block; got: {body}" |
| 1598 |
); |
| 1599 |
} |
| 1600 |
|
| 1601 |
#[tokio::test] |
| 1602 |
async fn migration_bearing_promote_requires_fresh_confirm() { |
| 1603 |
|
| 1604 |
|
| 1605 |
|
| 1606 |
|
| 1607 |
let pool = fresh_pool().await; |
| 1608 |
for (i, name) in ["host", "a"].iter().enumerate() { |
| 1609 |
sqlx::query( |
| 1610 |
"INSERT INTO tiers (name, ord, provisioned, canary) VALUES (?, ?, 1, 'sequential')", |
| 1611 |
) |
| 1612 |
.bind(name) |
| 1613 |
.bind(i as i64) |
| 1614 |
.execute(&pool) |
| 1615 |
.await |
| 1616 |
.unwrap(); |
| 1617 |
sqlx::query("INSERT INTO tier_state (tier) VALUES (?)") |
| 1618 |
.bind(name) |
| 1619 |
.execute(&pool) |
| 1620 |
.await |
| 1621 |
.unwrap(); |
| 1622 |
} |
| 1623 |
let mut topo = test_topo(); |
| 1624 |
topo.tiers[0].gates = vec![]; |
| 1625 |
let executors = Arc::new(crate::state::build_executors(&topo)); |
| 1626 |
let state = AppState { |
| 1627 |
pool, |
| 1628 |
topo: Arc::new(topo), |
| 1629 |
cfg: Arc::new(test_cfg()), |
| 1630 |
active_build: Arc::new(tokio::sync::Mutex::new(None)), |
| 1631 |
deploy_lock: Arc::new(tokio::sync::Mutex::new(())), |
| 1632 |
events: crate::events::channel(), |
| 1633 |
executors, |
| 1634 |
api_token: None, |
| 1635 |
}; |
| 1636 |
sqlx::query("INSERT INTO versions (version, git_sha, built_at, artifact_path) VALUES ('0.8.12','sha',datetime('now'),'/tmp/x')") |
| 1637 |
.execute(&state.pool).await.unwrap(); |
| 1638 |
sqlx::query("UPDATE tier_state SET current_version = '0.8.12' WHERE tier = 'host'") |
| 1639 |
.execute(&state.pool) |
| 1640 |
.await |
| 1641 |
.unwrap(); |
| 1642 |
|
| 1643 |
let app = router(state); |
| 1644 |
let resp = app |
| 1645 |
.oneshot( |
| 1646 |
Request::builder() |
| 1647 |
.method("POST") |
| 1648 |
.uri("/promote/a") |
| 1649 |
.header("content-type", "application/json") |
| 1650 |
.body(Body::from(r#"{"bears_migration": true}"#)) |
| 1651 |
.unwrap(), |
| 1652 |
) |
| 1653 |
.await |
| 1654 |
.unwrap(); |
| 1655 |
assert_eq!(resp.status(), StatusCode::CONFLICT); |
| 1656 |
let body = body_string(resp).await; |
| 1657 |
assert!( |
| 1658 |
body.contains("manual_confirm"), |
| 1659 |
"expected the migration promote to block on manual_confirm; got: {body}" |
| 1660 |
); |
| 1661 |
} |
| 1662 |
|
| 1663 |
#[tokio::test] |
| 1664 |
async fn tier_state_advance_is_atomic_previous_from_old_current() { |
| 1665 |
|
| 1666 |
|
| 1667 |
|
| 1668 |
let pool = fresh_pool().await; |
| 1669 |
seed(&pool, "a", "1.0.0").await; |
| 1670 |
|
| 1671 |
sqlx::query("INSERT INTO versions (version, git_sha, built_at, artifact_path) VALUES ('2.0.0','sha',datetime('now'),'/tmp/x')") |
| 1672 |
.execute(&pool).await.unwrap(); |
| 1673 |
sqlx::query("UPDATE tier_state SET current_version = '1.0.0' WHERE tier = 'a'") |
| 1674 |
.execute(&pool) |
| 1675 |
.await |
| 1676 |
.unwrap(); |
| 1677 |
|
| 1678 |
|
| 1679 |
|
| 1680 |
let v = crate::domain::Version::parse("2.0.0").unwrap(); |
| 1681 |
crate::runs::advance_tier(&pool, "a", &v, None) |
| 1682 |
.await |
| 1683 |
.unwrap(); |
| 1684 |
|
| 1685 |
let (cur, prev): (Option<String>, Option<String>) = sqlx::query_as( |
| 1686 |
"SELECT current_version, previous_version FROM tier_state WHERE tier = 'a'", |
| 1687 |
) |
| 1688 |
.fetch_one(&pool) |
| 1689 |
.await |
| 1690 |
.unwrap(); |
| 1691 |
assert_eq!(cur.as_deref(), Some("2.0.0")); |
| 1692 |
assert_eq!( |
| 1693 |
prev.as_deref(), |
| 1694 |
Some("1.0.0"), |
| 1695 |
"previous = the pre-update current, atomically" |
| 1696 |
); |
| 1697 |
} |
| 1698 |
|
| 1699 |
#[tokio::test] |
| 1700 |
async fn canary_rollback_restores_deployed_nodes_to_previous_version() { |
| 1701 |
use crate::topology::{Node, default_actuate, default_observe}; |
| 1702 |
let tmp = tempfile::tempdir().unwrap(); |
| 1703 |
|
| 1704 |
|
| 1705 |
|
| 1706 |
let mut nodes = Vec::new(); |
| 1707 |
for name in ["n1", "n2"] { |
| 1708 |
let rr = tmp.path().join(name); |
| 1709 |
for v in ["1.0.0", "2.0.0"] { |
| 1710 |
tokio::fs::create_dir_all(rr.join("releases").join(v)) |
| 1711 |
.await |
| 1712 |
.unwrap(); |
| 1713 |
} |
| 1714 |
tokio::fs::symlink("releases/2.0.0", rr.join("current")) |
| 1715 |
.await |
| 1716 |
.unwrap(); |
| 1717 |
nodes.push(Node { |
| 1718 |
name: name.into(), |
| 1719 |
ssh_target: "local".into(), |
| 1720 |
release_root: rr.to_string_lossy().into_owned(), |
| 1721 |
service_name: "x.service".into(), |
| 1722 |
health_url: None, |
| 1723 |
config_check_env_file: None, |
| 1724 |
actuate: default_actuate(), |
| 1725 |
observe: default_observe(), |
| 1726 |
companions: Vec::new(), |
| 1727 |
}); |
| 1728 |
} |
| 1729 |
|
| 1730 |
let mut state = test_state().await; |
| 1731 |
|
| 1732 |
|
| 1733 |
sqlx::query("INSERT INTO versions (version, git_sha, built_at, artifact_path) VALUES ('1.0.0','sha',datetime('now'),'/tmp/staged/releases/1.0.0/makenotwork')") |
| 1734 |
.execute(&state.pool).await.unwrap(); |
| 1735 |
let execs: crate::state::ExecutorMap = nodes |
| 1736 |
.iter() |
| 1737 |
.map(|n| (n.name.clone(), crate::state::build_executor(n))) |
| 1738 |
.collect(); |
| 1739 |
state.executors = std::sync::Arc::new(execs); |
| 1740 |
|
| 1741 |
let refs: Vec<&Node> = nodes.iter().collect(); |
| 1742 |
let restored = rollback_deployed_nodes(&state, &tid("a"), &refs, "1.0.0").await; |
| 1743 |
assert_eq!(restored, 2, "both deployed nodes should be restored"); |
| 1744 |
|
| 1745 |
for n in &nodes { |
| 1746 |
let cur = tokio::fs::read_link(std::path::Path::new(&n.release_root).join("current")) |
| 1747 |
.await |
| 1748 |
.unwrap(); |
| 1749 |
assert_eq!( |
| 1750 |
cur.to_string_lossy(), |
| 1751 |
"releases/1.0.0", |
| 1752 |
"node {} rolled back", |
| 1753 |
n.name |
| 1754 |
); |
| 1755 |
} |
| 1756 |
} |
| 1757 |
|
| 1758 |
#[tokio::test] |
| 1759 |
async fn canary_rollback_is_noop_without_a_previous_artifact() { |
| 1760 |
use crate::topology::{Node, default_actuate, default_observe}; |
| 1761 |
let tmp = tempfile::tempdir().unwrap(); |
| 1762 |
let rr = tmp.path().join("n1"); |
| 1763 |
tokio::fs::create_dir_all(&rr).await.unwrap(); |
| 1764 |
let node = Node { |
| 1765 |
name: "n1".into(), |
| 1766 |
ssh_target: "local".into(), |
| 1767 |
release_root: rr.to_string_lossy().into_owned(), |
| 1768 |
service_name: "x.service".into(), |
| 1769 |
health_url: None, |
| 1770 |
config_check_env_file: None, |
| 1771 |
actuate: default_actuate(), |
| 1772 |
observe: default_observe(), |
| 1773 |
companions: Vec::new(), |
| 1774 |
}; |
| 1775 |
let state = test_state().await; |
| 1776 |
let restored = rollback_deployed_nodes(&state, &tid("a"), &[&node], "9.9.9").await; |
| 1777 |
assert_eq!( |
| 1778 |
restored, 0, |
| 1779 |
"no artifact to roll back to -> nothing restored, no panic" |
| 1780 |
); |
| 1781 |
} |
| 1782 |
|
| 1783 |
|
| 1784 |
|
| 1785 |
|
| 1786 |
|
| 1787 |
|
| 1788 |
|
| 1789 |
|
| 1790 |
|
| 1791 |
|
| 1792 |
|
| 1793 |
|
| 1794 |
struct FleetFake { |
| 1795 |
tag: String, |
| 1796 |
caps: CapabilitySet, |
| 1797 |
log: Arc<StdMutex<Vec<String>>>, |
| 1798 |
fail_if_contains: Option<String>, |
| 1799 |
} |
| 1800 |
|
| 1801 |
impl FleetFake { |
| 1802 |
fn fails(&self, text: &str) -> bool { |
| 1803 |
self.fail_if_contains |
| 1804 |
.as_deref() |
| 1805 |
.is_some_and(|m| text.contains(m)) |
| 1806 |
} |
| 1807 |
} |
| 1808 |
|
| 1809 |
#[async_trait] |
| 1810 |
impl Executor for FleetFake { |
| 1811 |
async fn run_streaming( |
| 1812 |
&self, |
| 1813 |
step: &Step, |
| 1814 |
_sink: &mut dyn LogSink, |
| 1815 |
) -> anyhow::Result<RunOutput> { |
| 1816 |
let script = step.argv.last().cloned().unwrap_or_default(); |
| 1817 |
self.log |
| 1818 |
.lock() |
| 1819 |
.unwrap() |
| 1820 |
.push(format!("{}:run:{script}", self.tag)); |
| 1821 |
Ok(RunOutput { |
| 1822 |
status: std::process::ExitStatus::from_raw(if self.fails(&script) { |
| 1823 |
1 << 8 |
| 1824 |
} else { |
| 1825 |
0 |
| 1826 |
}), |
| 1827 |
stdout: Vec::new(), |
| 1828 |
stderr: Vec::new(), |
| 1829 |
}) |
| 1830 |
} |
| 1831 |
async fn pull_file( |
| 1832 |
&self, |
| 1833 |
_r: &std::path::Path, |
| 1834 |
_l: &std::path::Path, |
| 1835 |
_o: &SyncOpts, |
| 1836 |
) -> anyhow::Result<()> { |
| 1837 |
Ok(()) |
| 1838 |
} |
| 1839 |
async fn pull_dir( |
| 1840 |
&self, |
| 1841 |
_r: &std::path::Path, |
| 1842 |
_l: &std::path::Path, |
| 1843 |
_o: &SyncOpts, |
| 1844 |
) -> anyhow::Result<()> { |
| 1845 |
Ok(()) |
| 1846 |
} |
| 1847 |
async fn pull_glob( |
| 1848 |
&self, |
| 1849 |
_g: &str, |
| 1850 |
_l: &std::path::Path, |
| 1851 |
_o: &SyncOpts, |
| 1852 |
) -> anyhow::Result<()> { |
| 1853 |
Ok(()) |
| 1854 |
} |
| 1855 |
async fn push_dir( |
| 1856 |
&self, |
| 1857 |
_local: &std::path::Path, |
| 1858 |
remote: &std::path::Path, |
| 1859 |
_o: &SyncOpts, |
| 1860 |
) -> anyhow::Result<()> { |
| 1861 |
let dst = remote.display().to_string(); |
| 1862 |
self.log |
| 1863 |
.lock() |
| 1864 |
.unwrap() |
| 1865 |
.push(format!("{}:push:{dst}", self.tag)); |
| 1866 |
if self.fails(&dst) { |
| 1867 |
anyhow::bail!("fake rsync failure on {}", self.tag); |
| 1868 |
} |
| 1869 |
Ok(()) |
| 1870 |
} |
| 1871 |
fn capabilities(&self) -> &CapabilitySet { |
| 1872 |
&self.caps |
| 1873 |
} |
| 1874 |
} |
| 1875 |
|
| 1876 |
|
| 1877 |
|
| 1878 |
|
| 1879 |
|
| 1880 |
|
| 1881 |
async fn fleet_fixture( |
| 1882 |
names: &[&str], |
| 1883 |
fail_node: Option<&str>, |
| 1884 |
fail_marker: &str, |
| 1885 |
) -> (AppState, Arc<StdMutex<Vec<String>>>) { |
| 1886 |
use crate::topology::{default_actuate, default_observe}; |
| 1887 |
let mut state = test_state().await; |
| 1888 |
let log = Arc::new(StdMutex::new(Vec::<String>::new())); |
| 1889 |
|
| 1890 |
let nodes: Vec<Node> = names |
| 1891 |
.iter() |
| 1892 |
.map(|name| Node { |
| 1893 |
name: (*name).into(), |
| 1894 |
ssh_target: format!("deploy@{name}"), |
| 1895 |
release_root: format!("/tmp/fleet/{name}"), |
| 1896 |
service_name: "makenotwork.service".into(), |
| 1897 |
health_url: None, |
| 1898 |
config_check_env_file: None, |
| 1899 |
actuate: default_actuate(), |
| 1900 |
observe: default_observe(), |
| 1901 |
companions: Vec::new(), |
| 1902 |
}) |
| 1903 |
.collect(); |
| 1904 |
|
| 1905 |
let mut topo = (*state.topo).clone(); |
| 1906 |
topo.tiers[1].nodes = nodes.clone(); |
| 1907 |
|
| 1908 |
|
| 1909 |
topo.tiers[1].gates = vec![]; |
| 1910 |
state.topo = Arc::new(topo); |
| 1911 |
|
| 1912 |
let execs: crate::state::ExecutorMap = nodes |
| 1913 |
.iter() |
| 1914 |
.map(|n| { |
| 1915 |
let nm = n.name.to_string(); |
| 1916 |
let fail = fail_node == Some(nm.as_str()); |
| 1917 |
let exec: Arc<dyn Executor> = Arc::new(FleetFake { |
| 1918 |
tag: nm, |
| 1919 |
caps: CapabilitySet::from_tokens(["deploy", "restart"], ["health"]), |
| 1920 |
log: log.clone(), |
| 1921 |
fail_if_contains: fail.then(|| fail_marker.to_string()), |
| 1922 |
}); |
| 1923 |
(n.name.clone(), exec) |
| 1924 |
}) |
| 1925 |
.collect(); |
| 1926 |
state.executors = Arc::new(execs); |
| 1927 |
|
| 1928 |
for n in &nodes { |
| 1929 |
|
| 1930 |
sqlx::query( |
| 1931 |
"INSERT INTO nodes (name, tier, ssh_target, release_root) VALUES (?, 'a', ?, ?)", |
| 1932 |
) |
| 1933 |
.bind(&n.name) |
| 1934 |
.bind(&n.ssh_target) |
| 1935 |
.bind(&n.release_root) |
| 1936 |
.execute(&state.pool) |
| 1937 |
.await |
| 1938 |
.unwrap(); |
| 1939 |
} |
| 1940 |
for v in ["1.0.0", "2.0.0", "3.0.0"] { |
| 1941 |
|
| 1942 |
|
| 1943 |
|
| 1944 |
sqlx::query("INSERT INTO versions (version, git_sha, built_at, artifact_path) VALUES (?, 'sha', datetime('now'), ?)") |
| 1945 |
.bind(v).bind(format!("/tmp/staged/releases/{v}/makenotwork")).execute(&state.pool).await.unwrap(); |
| 1946 |
} |
| 1947 |
sqlx::query("UPDATE tier_state SET current_version = '3.0.0' WHERE tier = 'host'") |
| 1948 |
.execute(&state.pool) |
| 1949 |
.await |
| 1950 |
.unwrap(); |
| 1951 |
sqlx::query( |
| 1952 |
"UPDATE tier_state SET current_version = '2.0.0', previous_version = '1.0.0' WHERE tier = 'a'", |
| 1953 |
) |
| 1954 |
.execute(&state.pool) |
| 1955 |
.await |
| 1956 |
.unwrap(); |
| 1957 |
(state, log) |
| 1958 |
} |
| 1959 |
|
| 1960 |
|
| 1961 |
|
| 1962 |
fn first_touch(log: &[String], node: &str) -> usize { |
| 1963 |
let prefix = format!("{node}:"); |
| 1964 |
log.iter() |
| 1965 |
.position(|e| e.starts_with(&prefix)) |
| 1966 |
.unwrap_or_else(|| panic!("node {node:?} was never deployed to; log: {log:#?}")) |
| 1967 |
} |
| 1968 |
|
| 1969 |
#[tokio::test] |
| 1970 |
async fn promote_deploys_every_node_in_tier_order_and_advances() { |
| 1971 |
let (state, log) = fleet_fixture(&["a1", "a2", "a3"], None, "").await; |
| 1972 |
let pool = state.pool.clone(); |
| 1973 |
|
| 1974 |
let Json(body) = promote_inner(state, "a".into(), PromoteBody::default()) |
| 1975 |
.await |
| 1976 |
.expect("all nodes deploy, so the promote succeeds"); |
| 1977 |
assert_eq!( |
| 1978 |
body["nodes_deployed"], |
| 1979 |
serde_json::json!(["a1", "a2", "a3"]), |
| 1980 |
"the response names every node the promote reached", |
| 1981 |
); |
| 1982 |
|
| 1983 |
let (cur, prev) = tier_versions(&pool, "a").await; |
| 1984 |
assert_eq!(cur.as_deref(), Some("3.0.0")); |
| 1985 |
assert_eq!(prev.as_deref(), Some("2.0.0")); |
| 1986 |
|
| 1987 |
|
| 1988 |
let log = log.lock().unwrap().clone(); |
| 1989 |
assert!( |
| 1990 |
first_touch(&log, "a1") < first_touch(&log, "a2") |
| 1991 |
&& first_touch(&log, "a2") < first_touch(&log, "a3"), |
| 1992 |
"nodes must deploy in tier order: {log:#?}", |
| 1993 |
); |
| 1994 |
|
| 1995 |
|
| 1996 |
let ok: i64 = sqlx::query_scalar( |
| 1997 |
"SELECT COUNT(*) FROM deploys WHERE version = '3.0.0' AND outcome = 'ok'", |
| 1998 |
) |
| 1999 |
.fetch_one(&pool) |
| 2000 |
.await |
| 2001 |
.unwrap(); |
| 2002 |
assert_eq!(ok, 3, "one ok deploy row per node"); |
| 2003 |
} |
| 2004 |
|
| 2005 |
#[tokio::test] |
| 2006 |
async fn a_mid_canary_deploy_failure_rolls_touched_nodes_back_and_does_not_advance() { |
| 2007 |
|
| 2008 |
|
| 2009 |
|
| 2010 |
|
| 2011 |
let (state, log) = fleet_fixture(&["a1", "a2", "a3"], Some("a2"), "releases/3.0.0").await; |
| 2012 |
let pool = state.pool.clone(); |
| 2013 |
|
| 2014 |
let err = promote_inner(state, "a".into(), PromoteBody::default()) |
| 2015 |
.await |
| 2016 |
.expect_err("a mid-canary deploy failure must fail the promote"); |
| 2017 |
assert!( |
| 2018 |
matches!(err, crate::error::Error::Other(_)), |
| 2019 |
"a deploy failure propagates as Other, got: {err:?}", |
| 2020 |
); |
| 2021 |
|
| 2022 |
|
| 2023 |
let (cur, prev) = tier_versions(&pool, "a").await; |
| 2024 |
assert_eq!( |
| 2025 |
cur.as_deref(), |
| 2026 |
Some("2.0.0"), |
| 2027 |
"a failed rollout must not advance" |
| 2028 |
); |
| 2029 |
assert_eq!(prev.as_deref(), Some("1.0.0")); |
| 2030 |
|
| 2031 |
let log = log.lock().unwrap().clone(); |
| 2032 |
|
| 2033 |
assert!( |
| 2034 |
!log.iter().any(|e| e.starts_with("a3:")), |
| 2035 |
"nodes after the failure must not be deployed to: {log:#?}", |
| 2036 |
); |
| 2037 |
|
| 2038 |
for n in ["a1", "a2"] { |
| 2039 |
assert!( |
| 2040 |
log.iter() |
| 2041 |
.any(|e| e.starts_with(&format!("{n}:")) && e.contains("releases/2.0.0")), |
| 2042 |
"touched node {n} must be restored to 2.0.0: {log:#?}", |
| 2043 |
); |
| 2044 |
} |
| 2045 |
|
| 2046 |
|
| 2047 |
let a1: String = sqlx::query_scalar( |
| 2048 |
"SELECT outcome FROM deploys WHERE version = '3.0.0' AND node = 'a1'", |
| 2049 |
) |
| 2050 |
.fetch_one(&pool) |
| 2051 |
.await |
| 2052 |
.unwrap(); |
| 2053 |
assert_eq!(a1, "ok"); |
| 2054 |
let a2: String = sqlx::query_scalar( |
| 2055 |
"SELECT outcome FROM deploys WHERE version = '3.0.0' AND node = 'a2'", |
| 2056 |
) |
| 2057 |
.fetch_one(&pool) |
| 2058 |
.await |
| 2059 |
.unwrap(); |
| 2060 |
assert_eq!(a2, "failed"); |
| 2061 |
|
| 2062 |
|
| 2063 |
|
| 2064 |
let reason: Option<String> = |
| 2065 |
sqlx::query_scalar("SELECT partial_reason FROM tier_state WHERE tier = 'a'") |
| 2066 |
.fetch_one(&pool) |
| 2067 |
.await |
| 2068 |
.unwrap(); |
| 2069 |
assert_eq!( |
| 2070 |
reason, None, |
| 2071 |
"a fully-restored canary leaves the tier consistent, not partial", |
| 2072 |
); |
| 2073 |
} |
| 2074 |
|
| 2075 |
|
| 2076 |
|
| 2077 |
|
| 2078 |
|
| 2079 |
async fn rollback_fixture(prev: &str, current: &str) -> (AppState, tempfile::TempDir) { |
| 2080 |
use crate::topology::{Node, default_actuate, default_observe}; |
| 2081 |
let tmp = tempfile::tempdir().unwrap(); |
| 2082 |
let rr = tmp.path().join("a-local"); |
| 2083 |
for v in [prev, current] { |
| 2084 |
tokio::fs::create_dir_all(rr.join("releases").join(v)) |
| 2085 |
.await |
| 2086 |
.unwrap(); |
| 2087 |
} |
| 2088 |
tokio::fs::symlink(format!("releases/{current}"), rr.join("current")) |
| 2089 |
.await |
| 2090 |
.unwrap(); |
| 2091 |
let node = Node { |
| 2092 |
name: "a-local".into(), |
| 2093 |
ssh_target: "local".into(), |
| 2094 |
release_root: rr.to_string_lossy().into_owned(), |
| 2095 |
service_name: "x.service".into(), |
| 2096 |
health_url: None, |
| 2097 |
config_check_env_file: None, |
| 2098 |
actuate: default_actuate(), |
| 2099 |
observe: default_observe(), |
| 2100 |
companions: Vec::new(), |
| 2101 |
}; |
| 2102 |
|
| 2103 |
let mut state = test_state().await; |
| 2104 |
let mut topo = (*state.topo).clone(); |
| 2105 |
topo.tiers[1].nodes = vec![node]; |
| 2106 |
state.executors = Arc::new(crate::state::build_executors(&topo)); |
| 2107 |
state.topo = Arc::new(topo); |
| 2108 |
|
| 2109 |
|
| 2110 |
sqlx::query("INSERT INTO nodes (name, tier, ssh_target, release_root) VALUES ('a-local', 'a', 'local', ?)") |
| 2111 |
.bind(rr.to_string_lossy().into_owned()) |
| 2112 |
.execute(&state.pool).await.unwrap(); |
| 2113 |
for v in [prev, current] { |
| 2114 |
sqlx::query("INSERT INTO versions (version, git_sha, built_at, artifact_path) VALUES (?, 'sha', datetime('now'), ?)") |
| 2115 |
.bind(v).bind(format!("/tmp/staged/releases/{v}/makenotwork")).execute(&state.pool).await.unwrap(); |
| 2116 |
} |
| 2117 |
sqlx::query( |
| 2118 |
"UPDATE tier_state SET current_version = ?, previous_version = ? WHERE tier = 'a'", |
| 2119 |
) |
| 2120 |
.bind(current) |
| 2121 |
.bind(prev) |
| 2122 |
.execute(&state.pool) |
| 2123 |
.await |
| 2124 |
.unwrap(); |
| 2125 |
(state, tmp) |
| 2126 |
} |
| 2127 |
|
| 2128 |
async fn tier_versions(pool: &SqlitePool, tier: &str) -> (Option<String>, Option<String>) { |
| 2129 |
sqlx::query_as("SELECT current_version, previous_version FROM tier_state WHERE tier = ?") |
| 2130 |
.bind(tier) |
| 2131 |
.fetch_one(pool) |
| 2132 |
.await |
| 2133 |
.unwrap() |
| 2134 |
} |
| 2135 |
|
| 2136 |
#[tokio::test] |
| 2137 |
async fn rollback_clears_previous_version_rather_than_swapping_it() { |
| 2138 |
|
| 2139 |
|
| 2140 |
|
| 2141 |
let (state, _tmp) = rollback_fixture("1.0.0", "2.0.0").await; |
| 2142 |
let pool = state.pool.clone(); |
| 2143 |
|
| 2144 |
let _ = rollback(State(state), Path("a".to_string())).await.unwrap(); |
| 2145 |
|
| 2146 |
let (cur, prev) = tier_versions(&pool, "a").await; |
| 2147 |
assert_eq!( |
| 2148 |
cur.as_deref(), |
| 2149 |
Some("1.0.0"), |
| 2150 |
"rolled back to the previous version" |
| 2151 |
); |
| 2152 |
assert_eq!( |
| 2153 |
prev, None, |
| 2154 |
"the version we rolled off must NOT become the rollback target" |
| 2155 |
); |
| 2156 |
} |
| 2157 |
|
| 2158 |
#[tokio::test] |
| 2159 |
async fn second_rollback_refuses_instead_of_rolling_forward() { |
| 2160 |
let (state, _tmp) = rollback_fixture("1.0.0", "2.0.0").await; |
| 2161 |
let pool = state.pool.clone(); |
| 2162 |
let release_root = state.topo.tiers[1].nodes[0].release_root.clone(); |
| 2163 |
|
| 2164 |
let _ = rollback(State(state.clone()), Path("a".to_string())) |
| 2165 |
.await |
| 2166 |
.unwrap(); |
| 2167 |
let err = rollback(State(state), Path("a".to_string())) |
| 2168 |
.await |
| 2169 |
.expect_err("only one step of history is tracked; a second rollback must refuse"); |
| 2170 |
assert!( |
| 2171 |
matches!(&err, crate::error::Error::GateBlocked(m) if m.contains("no previous_version")), |
| 2172 |
"expected a loud refusal, got: {err:?}", |
| 2173 |
); |
| 2174 |
|
| 2175 |
|
| 2176 |
let (cur, _) = tier_versions(&pool, "a").await; |
| 2177 |
assert_eq!(cur.as_deref(), Some("1.0.0")); |
| 2178 |
let link = tokio::fs::read_link(std::path::Path::new(&release_root).join("current")) |
| 2179 |
.await |
| 2180 |
.unwrap(); |
| 2181 |
assert_eq!( |
| 2182 |
link.to_string_lossy(), |
| 2183 |
"releases/1.0.0", |
| 2184 |
"node stays on the rolled-back version" |
| 2185 |
); |
| 2186 |
} |
| 2187 |
|
| 2188 |
#[tokio::test] |
| 2189 |
async fn promote_refuses_an_unprovisioned_tier() { |
| 2190 |
|
| 2191 |
|
| 2192 |
|
| 2193 |
let (mut state, _tmp) = rollback_fixture("1.0.0", "2.0.0").await; |
| 2194 |
let mut topo = (*state.topo).clone(); |
| 2195 |
topo.tiers[1].provisioned = false; |
| 2196 |
topo.tiers[1].nodes.clear(); |
| 2197 |
state.topo = Arc::new(topo); |
| 2198 |
let pool = state.pool.clone(); |
| 2199 |
sqlx::query("UPDATE tier_state SET current_version = '2.0.0' WHERE tier = 'host'") |
| 2200 |
.execute(&pool) |
| 2201 |
.await |
| 2202 |
.unwrap(); |
| 2203 |
|
| 2204 |
let err = promote_inner(state, "a".into(), PromoteBody::default()) |
| 2205 |
.await |
| 2206 |
.expect_err("promoting to an unprovisioned tier must be refused"); |
| 2207 |
assert!( |
| 2208 |
matches!(&err, crate::error::Error::GateBlocked(m) if m.contains("not provisioned")), |
| 2209 |
"got: {err:?}", |
| 2210 |
); |
| 2211 |
|
| 2212 |
|
| 2213 |
let (cur, _) = tier_versions(&pool, "a").await; |
| 2214 |
assert_eq!( |
| 2215 |
cur.as_deref(), |
| 2216 |
Some("2.0.0"), |
| 2217 |
"a refused promote must not advance tier_state", |
| 2218 |
); |
| 2219 |
} |
| 2220 |
|
| 2221 |
#[tokio::test] |
| 2222 |
async fn promote_advances_the_tier_and_flips_the_symlink_when_gates_are_green() { |
| 2223 |
|
| 2224 |
|
| 2225 |
|
| 2226 |
|
| 2227 |
|
| 2228 |
use crate::topology::Gate; |
| 2229 |
let (mut state, _tmp) = rollback_fixture("1.0.0", "2.0.0").await; |
| 2230 |
|
| 2231 |
|
| 2232 |
|
| 2233 |
|
| 2234 |
let mut topo = (*state.topo).clone(); |
| 2235 |
topo.tiers[0].gates = vec![Gate::CargoTest]; |
| 2236 |
state.topo = Arc::new(topo); |
| 2237 |
let pool = state.pool.clone(); |
| 2238 |
let release_root = state.topo.tiers[1].nodes[0].release_root.clone(); |
| 2239 |
|
| 2240 |
|
| 2241 |
tokio::fs::create_dir_all( |
| 2242 |
std::path::Path::new(&release_root) |
| 2243 |
.join("releases") |
| 2244 |
.join("3.0.0"), |
| 2245 |
) |
| 2246 |
.await |
| 2247 |
.unwrap(); |
| 2248 |
sqlx::query("INSERT INTO versions (version, git_sha, built_at, artifact_path) VALUES ('3.0.0','sha',datetime('now'),'/tmp/staged/releases/3.0.0/makenotwork')") |
| 2249 |
.execute(&pool).await.unwrap(); |
| 2250 |
sqlx::query("UPDATE tier_state SET current_version = '3.0.0' WHERE tier = 'host'") |
| 2251 |
.execute(&pool) |
| 2252 |
.await |
| 2253 |
.unwrap(); |
| 2254 |
insert_gate(&pool, "host", "3.0.0", "cargo_test", 1).await; |
| 2255 |
|
| 2256 |
|
| 2257 |
set_partial(&state, &tid("a"), "left over from a previous canary").await; |
| 2258 |
|
| 2259 |
let Json(body) = promote_inner(state, "a".into(), PromoteBody::default()) |
| 2260 |
.await |
| 2261 |
.expect("gates are green and the node deploys, so the promote succeeds"); |
| 2262 |
|
| 2263 |
assert_eq!(body["tier"], "a"); |
| 2264 |
assert_eq!(body["version"], "3.0.0"); |
| 2265 |
assert_eq!( |
| 2266 |
body["nodes_deployed"], |
| 2267 |
serde_json::json!(["a-local"]), |
| 2268 |
"the response names every node the promote reached", |
| 2269 |
); |
| 2270 |
|
| 2271 |
let (cur, prev) = tier_versions(&pool, "a").await; |
| 2272 |
assert_eq!(cur.as_deref(), Some("3.0.0")); |
| 2273 |
assert_eq!( |
| 2274 |
prev.as_deref(), |
| 2275 |
Some("2.0.0"), |
| 2276 |
"previous_version is the version we came off, so a rollback aims at it", |
| 2277 |
); |
| 2278 |
|
| 2279 |
|
| 2280 |
let link = tokio::fs::read_link(std::path::Path::new(&release_root).join("current")) |
| 2281 |
.await |
| 2282 |
.unwrap(); |
| 2283 |
assert_eq!(link.to_string_lossy(), "releases/3.0.0"); |
| 2284 |
|
| 2285 |
let reason: Option<String> = |
| 2286 |
sqlx::query_scalar("SELECT partial_reason FROM tier_state WHERE tier = 'a'") |
| 2287 |
.fetch_one(&pool) |
| 2288 |
.await |
| 2289 |
.unwrap(); |
| 2290 |
assert_eq!( |
| 2291 |
reason, None, |
| 2292 |
"a clean full rollout clears a stale partial flag", |
| 2293 |
); |
| 2294 |
|
| 2295 |
|
| 2296 |
|
| 2297 |
let (node, outcome): (String, String) = |
| 2298 |
sqlx::query_as("SELECT node, outcome FROM deploys WHERE version = '3.0.0'") |
| 2299 |
.fetch_one(&pool) |
| 2300 |
.await |
| 2301 |
.unwrap(); |
| 2302 |
assert_eq!(node, "a-local"); |
| 2303 |
assert_eq!(outcome, "ok"); |
| 2304 |
} |
| 2305 |
|
| 2306 |
#[tokio::test] |
| 2307 |
async fn promote_fails_and_flags_the_tier_when_post_deploy_gates_are_red() { |
| 2308 |
|
| 2309 |
|
| 2310 |
|
| 2311 |
|
| 2312 |
use crate::topology::Gate; |
| 2313 |
let (mut state, _tmp) = rollback_fixture("1.0.0", "2.0.0").await; |
| 2314 |
|
| 2315 |
|
| 2316 |
|
| 2317 |
let mut topo = (*state.topo).clone(); |
| 2318 |
topo.tiers[1].gates = vec![Gate::NodeHealth]; |
| 2319 |
state.topo = Arc::new(topo); |
| 2320 |
state.executors = Arc::new(crate::state::ExecutorMap::new()); |
| 2321 |
let pool = state.pool.clone(); |
| 2322 |
|
| 2323 |
sqlx::query("INSERT INTO versions (version, git_sha, built_at, artifact_path) VALUES ('3.0.0','sha',datetime('now'),'/tmp/staged/releases/3.0.0/makenotwork')") |
| 2324 |
.execute(&pool).await.unwrap(); |
| 2325 |
sqlx::query("UPDATE tier_state SET current_version = '3.0.0' WHERE tier = 'host'") |
| 2326 |
.execute(&pool) |
| 2327 |
.await |
| 2328 |
.unwrap(); |
| 2329 |
|
| 2330 |
let err = promote_inner(state, "a".into(), PromoteBody::default()) |
| 2331 |
.await |
| 2332 |
.expect_err("red post-deploy gates must fail the promote"); |
| 2333 |
assert!( |
| 2334 |
matches!(&err, crate::error::Error::GateBlocked(m) if m.contains("node_health")), |
| 2335 |
"the failure must name the red gate, got: {err:?}", |
| 2336 |
); |
| 2337 |
|
| 2338 |
let (cur, _) = tier_versions(&pool, "a").await; |
| 2339 |
assert_eq!( |
| 2340 |
cur.as_deref(), |
| 2341 |
Some("3.0.0"), |
| 2342 |
"tier_state tracks what the nodes actually run" |
| 2343 |
); |
| 2344 |
let reason: Option<String> = |
| 2345 |
sqlx::query_scalar("SELECT partial_reason FROM tier_state WHERE tier = 'a'") |
| 2346 |
.fetch_one(&pool) |
| 2347 |
.await |
| 2348 |
.unwrap(); |
| 2349 |
assert!( |
| 2350 |
reason.as_deref().is_some_and(|r| r.contains("node_health")), |
| 2351 |
"the tier must be flagged for /state and the TUI, got: {reason:?}", |
| 2352 |
); |
| 2353 |
} |
| 2354 |
|
| 2355 |
#[tokio::test] |
| 2356 |
async fn set_partial_then_clear_roundtrips() { |
| 2357 |
let state = test_state().await; |
| 2358 |
let read = || async { |
| 2359 |
sqlx::query_scalar::<_, Option<String>>( |
| 2360 |
"SELECT partial_reason FROM tier_state WHERE tier = 'a'", |
| 2361 |
) |
| 2362 |
.fetch_one(&state.pool) |
| 2363 |
.await |
| 2364 |
.unwrap() |
| 2365 |
}; |
| 2366 |
assert_eq!(read().await, None, "consistent tier starts clean"); |
| 2367 |
set_partial(&state, &tid("a"), "canary rollback incomplete: 1/2").await; |
| 2368 |
assert_eq!( |
| 2369 |
read().await.as_deref(), |
| 2370 |
Some("canary rollback incomplete: 1/2") |
| 2371 |
); |
| 2372 |
clear_partial(&state, &tid("a")).await; |
| 2373 |
assert_eq!(read().await, None, "clear nulls it back out"); |
| 2374 |
} |
| 2375 |
|
| 2376 |
#[tokio::test] |
| 2377 |
async fn state_surfaces_partial_reason() { |
| 2378 |
use axum::extract::State; |
| 2379 |
let state = test_state().await; |
| 2380 |
set_partial( |
| 2381 |
&state, |
| 2382 |
&tid("a"), |
| 2383 |
"first-deploy canary failed: 1 node(s) on 2.0.0", |
| 2384 |
) |
| 2385 |
.await; |
| 2386 |
let Json(view) = get_state(State(state)).await.unwrap(); |
| 2387 |
let a = view.tiers.iter().find(|t| t.name == "a").unwrap(); |
| 2388 |
assert_eq!( |
| 2389 |
a.partial_reason.as_deref(), |
| 2390 |
Some("first-deploy canary failed: 1 node(s) on 2.0.0"), |
| 2391 |
); |
| 2392 |
let host = view.tiers.iter().find(|t| t.name == "host").unwrap(); |
| 2393 |
assert_eq!( |
| 2394 |
host.partial_reason, None, |
| 2395 |
"untouched tier stays clean in /state" |
| 2396 |
); |
| 2397 |
} |
| 2398 |
|
| 2399 |
#[tokio::test] |
| 2400 |
async fn state_build_is_null_until_first_rebuild_then_surfaces_latest() { |
| 2401 |
use axum::extract::State; |
| 2402 |
let state = test_state().await; |
| 2403 |
|
| 2404 |
|
| 2405 |
let Json(view) = get_state(State(state.clone())).await.unwrap(); |
| 2406 |
assert!(view.build.is_none()); |
| 2407 |
|
| 2408 |
|
| 2409 |
let run_id = crate::runs::create(&state.pool, "deadbeef").await.unwrap(); |
| 2410 |
crate::runs::mark_failed(&state.pool, run_id, "cargo_test: 3 test(s) failed") |
| 2411 |
.await |
| 2412 |
.unwrap(); |
| 2413 |
let Json(view) = get_state(State(state)).await.unwrap(); |
| 2414 |
let b = view.build.expect("build surfaced"); |
| 2415 |
assert_eq!(b.run_id, run_id.0); |
| 2416 |
assert_eq!(b.result, "failed"); |
| 2417 |
assert_eq!( |
| 2418 |
b.failure_summary.as_deref(), |
| 2419 |
Some("cargo_test: 3 test(s) failed") |
| 2420 |
); |
| 2421 |
} |
| 2422 |
|
| 2423 |
#[tokio::test] |
| 2424 |
async fn status_json_serves_the_shared_payload_over_the_real_router() { |
| 2425 |
|
| 2426 |
|
| 2427 |
|
| 2428 |
|
| 2429 |
let state = test_state().await; |
| 2430 |
set_partial(&state, &tid("a"), "canary rollback incomplete: 1/2").await; |
| 2431 |
|
| 2432 |
let resp = router(state) |
| 2433 |
.oneshot( |
| 2434 |
Request::builder() |
| 2435 |
.uri("/status.json") |
| 2436 |
.body(Body::empty()) |
| 2437 |
.unwrap(), |
| 2438 |
) |
| 2439 |
.await |
| 2440 |
.unwrap(); |
| 2441 |
assert_eq!(resp.status(), StatusCode::OK); |
| 2442 |
|
| 2443 |
let body = http_body_util::BodyExt::collect(resp.into_body()) |
| 2444 |
.await |
| 2445 |
.unwrap() |
| 2446 |
.to_bytes(); |
| 2447 |
let payload: ops_status::Payload = serde_json::from_slice(&body).unwrap(); |
| 2448 |
|
| 2449 |
assert_eq!(payload.source, crate::status::SOURCE); |
| 2450 |
assert_eq!(payload.schema, ops_status::SCHEMA_VERSION); |
| 2451 |
assert_eq!(payload.validate(), Ok(())); |
| 2452 |
assert_eq!( |
| 2453 |
payload.node("tier:a").unwrap().status, |
| 2454 |
ops_status::Status::Failed, |
| 2455 |
"a partial tier must surface as failed over the wire" |
| 2456 |
); |
| 2457 |
assert_eq!(payload.worst_status(), ops_status::Status::Failed); |
| 2458 |
} |
| 2459 |
|
| 2460 |
#[tokio::test] |
| 2461 |
async fn promote_with_explicit_version_but_missing_artifact_404s() { |
| 2462 |
|
| 2463 |
|
| 2464 |
let state = test_state().await; |
| 2465 |
let app = router(state); |
| 2466 |
let resp = app |
| 2467 |
.oneshot( |
| 2468 |
Request::builder() |
| 2469 |
.method("POST") |
| 2470 |
.uri("/promote/a") |
| 2471 |
.header("content-type", "application/json") |
| 2472 |
.body(Body::from(r#"{"version":"9.9.9"}"#)) |
| 2473 |
.unwrap(), |
| 2474 |
) |
| 2475 |
.await |
| 2476 |
.unwrap(); |
| 2477 |
assert_eq!(resp.status(), StatusCode::NOT_FOUND); |
| 2478 |
} |
| 2479 |
|
| 2480 |
|
| 2481 |
|
| 2482 |
async fn state_with_logs_root(logs_root: PathBuf) -> AppState { |
| 2483 |
let mut s = test_state().await; |
| 2484 |
let mut cfg = (*s.cfg).clone(); |
| 2485 |
cfg.logs_root = logs_root; |
| 2486 |
s.cfg = Arc::new(cfg); |
| 2487 |
s |
| 2488 |
} |
| 2489 |
|
| 2490 |
#[tokio::test] |
| 2491 |
async fn get_gate_log_returns_file_contents() { |
| 2492 |
let tmp = tempfile::tempdir().unwrap(); |
| 2493 |
let dir = tmp.path().join("0.9.5"); |
| 2494 |
tokio::fs::create_dir_all(&dir).await.unwrap(); |
| 2495 |
tokio::fs::write(dir.join("cargo_test.log"), b"hello sandod\n") |
| 2496 |
.await |
| 2497 |
.unwrap(); |
| 2498 |
|
| 2499 |
let state = state_with_logs_root(tmp.path().to_path_buf()).await; |
| 2500 |
let app = router(state); |
| 2501 |
let resp = app |
| 2502 |
.oneshot( |
| 2503 |
Request::builder() |
| 2504 |
.uri("/logs/0.9.5/cargo_test") |
| 2505 |
.body(Body::empty()) |
| 2506 |
.unwrap(), |
| 2507 |
) |
| 2508 |
.await |
| 2509 |
.unwrap(); |
| 2510 |
assert_eq!(resp.status(), StatusCode::OK); |
| 2511 |
assert_eq!(body_string(resp).await, "hello sandod\n"); |
| 2512 |
} |
| 2513 |
|
| 2514 |
#[tokio::test] |
| 2515 |
async fn get_gate_log_404s_when_missing() { |
| 2516 |
let tmp = tempfile::tempdir().unwrap(); |
| 2517 |
let state = state_with_logs_root(tmp.path().to_path_buf()).await; |
| 2518 |
let app = router(state); |
| 2519 |
let resp = app |
| 2520 |
.oneshot( |
| 2521 |
Request::builder() |
| 2522 |
.uri("/logs/0.9.5/cargo_test") |
| 2523 |
.body(Body::empty()) |
| 2524 |
.unwrap(), |
| 2525 |
) |
| 2526 |
.await |
| 2527 |
.unwrap(); |
| 2528 |
assert_eq!(resp.status(), StatusCode::NOT_FOUND); |
| 2529 |
} |
| 2530 |
|
| 2531 |
|
| 2532 |
|
| 2533 |
#[tokio::test] |
| 2534 |
async fn mutating_route_requires_bearer_when_token_set() { |
| 2535 |
let mut state = test_state().await; |
| 2536 |
state.api_token = Some(std::sync::Arc::from("s3cr3t")); |
| 2537 |
let app = router(state); |
| 2538 |
|
| 2539 |
|
| 2540 |
let resp = app |
| 2541 |
.clone() |
| 2542 |
.oneshot( |
| 2543 |
Request::builder() |
| 2544 |
.method("POST") |
| 2545 |
.uri("/promote/a") |
| 2546 |
.body(Body::empty()) |
| 2547 |
.unwrap(), |
| 2548 |
) |
| 2549 |
.await |
| 2550 |
.unwrap(); |
| 2551 |
assert_eq!(resp.status(), StatusCode::UNAUTHORIZED); |
| 2552 |
|
| 2553 |
|
| 2554 |
let resp = app |
| 2555 |
.clone() |
| 2556 |
.oneshot( |
| 2557 |
Request::builder() |
| 2558 |
.method("POST") |
| 2559 |
.uri("/promote/a") |
| 2560 |
.header("authorization", "Bearer nope") |
| 2561 |
.body(Body::empty()) |
| 2562 |
.unwrap(), |
| 2563 |
) |
| 2564 |
.await |
| 2565 |
.unwrap(); |
| 2566 |
assert_eq!(resp.status(), StatusCode::UNAUTHORIZED); |
| 2567 |
|
| 2568 |
|
| 2569 |
|
| 2570 |
let resp = app |
| 2571 |
.clone() |
| 2572 |
.oneshot( |
| 2573 |
Request::builder() |
| 2574 |
.method("POST") |
| 2575 |
.uri("/promote/a") |
| 2576 |
.header("authorization", "Bearer s3cr3t") |
| 2577 |
.body(Body::empty()) |
| 2578 |
.unwrap(), |
| 2579 |
) |
| 2580 |
.await |
| 2581 |
.unwrap(); |
| 2582 |
assert_ne!(resp.status(), StatusCode::UNAUTHORIZED); |
| 2583 |
} |
| 2584 |
|
| 2585 |
#[tokio::test] |
| 2586 |
async fn read_routes_require_token_when_set() { |
| 2587 |
|
| 2588 |
|
| 2589 |
|
| 2590 |
let mut state = test_state().await; |
| 2591 |
state.api_token = Some(std::sync::Arc::from("s3cr3t")); |
| 2592 |
let app = router(state); |
| 2593 |
|
| 2594 |
|
| 2595 |
let resp = app |
| 2596 |
.clone() |
| 2597 |
.oneshot( |
| 2598 |
Request::builder() |
| 2599 |
.uri("/state") |
| 2600 |
.body(Body::empty()) |
| 2601 |
.unwrap(), |
| 2602 |
) |
| 2603 |
.await |
| 2604 |
.unwrap(); |
| 2605 |
assert_eq!(resp.status(), StatusCode::UNAUTHORIZED); |
| 2606 |
|
| 2607 |
|
| 2608 |
let resp = app |
| 2609 |
.oneshot( |
| 2610 |
Request::builder() |
| 2611 |
.uri("/state") |
| 2612 |
.header("authorization", "Bearer s3cr3t") |
| 2613 |
.body(Body::empty()) |
| 2614 |
.unwrap(), |
| 2615 |
) |
| 2616 |
.await |
| 2617 |
.unwrap(); |
| 2618 |
assert_eq!(resp.status(), StatusCode::OK); |
| 2619 |
} |
| 2620 |
|
| 2621 |
#[tokio::test] |
| 2622 |
async fn read_routes_open_without_a_token() { |
| 2623 |
|
| 2624 |
|
| 2625 |
let state = test_state().await; |
| 2626 |
let app = router(state); |
| 2627 |
let resp = app |
| 2628 |
.oneshot( |
| 2629 |
Request::builder() |
| 2630 |
.uri("/state") |
| 2631 |
.body(Body::empty()) |
| 2632 |
.unwrap(), |
| 2633 |
) |
| 2634 |
.await |
| 2635 |
.unwrap(); |
| 2636 |
assert_eq!(resp.status(), StatusCode::OK); |
| 2637 |
} |
| 2638 |
|
| 2639 |
#[tokio::test] |
| 2640 |
async fn self_update_malformed_body_is_400_not_422() { |
| 2641 |
|
| 2642 |
|
| 2643 |
let state = test_state().await; |
| 2644 |
let app = router(state); |
| 2645 |
let resp = app |
| 2646 |
.oneshot( |
| 2647 |
Request::builder() |
| 2648 |
.method("POST") |
| 2649 |
.uri("/self-update") |
| 2650 |
.header("content-type", "application/json") |
| 2651 |
.body(Body::from("{ not valid json")) |
| 2652 |
.unwrap(), |
| 2653 |
) |
| 2654 |
.await |
| 2655 |
.unwrap(); |
| 2656 |
assert_eq!(resp.status(), StatusCode::BAD_REQUEST); |
| 2657 |
} |
| 2658 |
|
| 2659 |
#[tokio::test] |
| 2660 |
async fn gate_log_rejects_unknown_gate_kind() { |
| 2661 |
|
| 2662 |
|
| 2663 |
let state = test_state().await; |
| 2664 |
let app = router(state); |
| 2665 |
let resp = app |
| 2666 |
.oneshot( |
| 2667 |
Request::builder() |
| 2668 |
.uri("/logs/0.9.6/passwd") |
| 2669 |
.body(Body::empty()) |
| 2670 |
.unwrap(), |
| 2671 |
) |
| 2672 |
.await |
| 2673 |
.unwrap(); |
| 2674 |
assert_eq!(resp.status(), StatusCode::NOT_FOUND); |
| 2675 |
} |
| 2676 |
|
| 2677 |
|
| 2678 |
|
| 2679 |
|
| 2680 |
|
| 2681 |
#[tokio::test] |
| 2682 |
async fn get_gate_log_rejects_dotdot_segments() { |
| 2683 |
let tmp = tempfile::tempdir().unwrap(); |
| 2684 |
let state = state_with_logs_root(tmp.path().to_path_buf()).await; |
| 2685 |
let app = router(state); |
| 2686 |
let resp = app |
| 2687 |
.oneshot( |
| 2688 |
Request::builder() |
| 2689 |
.uri("/logs/../etc/passwd") |
| 2690 |
.body(Body::empty()) |
| 2691 |
.unwrap(), |
| 2692 |
) |
| 2693 |
.await |
| 2694 |
.unwrap(); |
| 2695 |
assert_eq!(resp.status(), StatusCode::NOT_FOUND); |
| 2696 |
} |
| 2697 |
|
| 2698 |
|
| 2699 |
|
| 2700 |
|
| 2701 |
|
| 2702 |
|
| 2703 |
async fn seed_build(pool: &SqlitePool, sha: &str, version: &str, staged_path: &str) -> i64 { |
| 2704 |
sqlx::query_scalar( |
| 2705 |
"INSERT INTO build_runs (sha, version, phase, result, started_at, bundle_digest, staged_path) |
| 2706 |
VALUES (?, ?, 'done', 'passed', datetime('now'), ?, ?) RETURNING id", |
| 2707 |
) |
| 2708 |
.bind(sha) |
| 2709 |
.bind(version) |
| 2710 |
.bind(format!("{sha}-digest")) |
| 2711 |
.bind(staged_path) |
| 2712 |
.fetch_one(pool) |
| 2713 |
.await |
| 2714 |
.unwrap() |
| 2715 |
} |
| 2716 |
|
| 2717 |
async fn insert_gate_build( |
| 2718 |
pool: &SqlitePool, |
| 2719 |
tier: &str, |
| 2720 |
version: &str, |
| 2721 |
kind: &str, |
| 2722 |
passed: i64, |
| 2723 |
build_id: i64, |
| 2724 |
) { |
| 2725 |
let status = if passed == 1 { "passed" } else { "failed" }; |
| 2726 |
sqlx::query( |
| 2727 |
"INSERT INTO gate_runs (version, tier, gate_kind, started_at, finished_at, status, build_id) \ |
| 2728 |
VALUES (?, ?, ?, datetime('now'), datetime('now'), ?, ?)", |
| 2729 |
) |
| 2730 |
.bind(version) |
| 2731 |
.bind(tier) |
| 2732 |
.bind(kind) |
| 2733 |
.bind(status) |
| 2734 |
.bind(build_id) |
| 2735 |
.execute(pool) |
| 2736 |
.await |
| 2737 |
.unwrap(); |
| 2738 |
} |
| 2739 |
|
| 2740 |
#[tokio::test] |
| 2741 |
async fn unsatisfied_gates_keys_build_evidence_on_build_id() { |
| 2742 |
let pool = fresh_pool().await; |
| 2743 |
seed(&pool, "a", "3.0.0").await; |
| 2744 |
|
| 2745 |
let b1 = seed_build(&pool, "sha1", "3.0.0", "/rel/1111111111111111").await; |
| 2746 |
let b2 = seed_build(&pool, "sha2", "3.0.0", "/rel/2222222222222222").await; |
| 2747 |
insert_gate_build(&pool, "a", "3.0.0", "cargo_test", 1, b1).await; |
| 2748 |
|
| 2749 |
|
| 2750 |
let ok = unsatisfied_gates( |
| 2751 |
&pool, |
| 2752 |
&tid("a"), |
| 2753 |
&[Gate::CargoTest], |
| 2754 |
"3.0.0", |
| 2755 |
Some(b1), |
| 2756 |
false, |
| 2757 |
) |
| 2758 |
.await |
| 2759 |
.unwrap(); |
| 2760 |
assert!(ok.is_empty(), "the build that passed its gate is satisfied"); |
| 2761 |
|
| 2762 |
|
| 2763 |
|
| 2764 |
let bad = unsatisfied_gates( |
| 2765 |
&pool, |
| 2766 |
&tid("a"), |
| 2767 |
&[Gate::CargoTest], |
| 2768 |
"3.0.0", |
| 2769 |
Some(b2), |
| 2770 |
false, |
| 2771 |
) |
| 2772 |
.await |
| 2773 |
.unwrap(); |
| 2774 |
assert_eq!( |
| 2775 |
bad, |
| 2776 |
vec!["cargo_test".to_string()], |
| 2777 |
"a different build of the same version does not inherit the evidence" |
| 2778 |
); |
| 2779 |
|
| 2780 |
|
| 2781 |
let legacy = unsatisfied_gates(&pool, &tid("a"), &[Gate::CargoTest], "3.0.0", None, false) |
| 2782 |
.await |
| 2783 |
.unwrap(); |
| 2784 |
assert!(legacy.is_empty(), "version-keyed legacy path unchanged"); |
| 2785 |
} |
| 2786 |
|
| 2787 |
#[tokio::test] |
| 2788 |
async fn promote_rejects_an_explicit_version_that_is_not_the_source_build() { |
| 2789 |
|
| 2790 |
|
| 2791 |
|
| 2792 |
|
| 2793 |
let state = test_state().await; |
| 2794 |
seed_version(&state.pool, "3.0.0").await; |
| 2795 |
let b = seed_build(&state.pool, "shaB", "3.0.0", "/rel/deadbeefdeadbeef").await; |
| 2796 |
sqlx::query( |
| 2797 |
"UPDATE tier_state SET current_version='3.0.0', current_build_id=? WHERE tier='host'", |
| 2798 |
) |
| 2799 |
.bind(b) |
| 2800 |
.execute(&state.pool) |
| 2801 |
.await |
| 2802 |
.unwrap(); |
| 2803 |
|
| 2804 |
let err = promote_inner( |
| 2805 |
state, |
| 2806 |
"a".into(), |
| 2807 |
PromoteBody { |
| 2808 |
version: Some("2.0.0".into()), |
| 2809 |
..Default::default() |
| 2810 |
}, |
| 2811 |
) |
| 2812 |
.await |
| 2813 |
.expect_err("promoting a version other than the source build must be refused"); |
| 2814 |
assert!( |
| 2815 |
matches!(&err, crate::error::Error::GateBlocked(m) if m.contains("vouched for")), |
| 2816 |
"the refusal must name the mismatch, got: {err:?}", |
| 2817 |
); |
| 2818 |
} |
| 2819 |
|
| 2820 |
#[tokio::test] |
| 2821 |
async fn promote_identity_path_deploys_the_source_build_and_advances_build_id() { |
| 2822 |
|
| 2823 |
|
| 2824 |
|
| 2825 |
let mut state = test_state().await; |
| 2826 |
let node_root = tempfile::tempdir().unwrap(); |
| 2827 |
|
| 2828 |
let mut topo = (*state.topo).clone(); |
| 2829 |
topo.tiers[1].nodes[0].release_root = node_root.path().to_string_lossy().into_owned(); |
| 2830 |
topo.tiers[1].gates = vec![]; |
| 2831 |
state.topo = Arc::new(topo); |
| 2832 |
state.executors = Arc::new(crate::state::build_executors(&state.topo)); |
| 2833 |
|
| 2834 |
|
| 2835 |
sqlx::query("INSERT INTO nodes (name, tier, ssh_target, release_root) VALUES ('a-local', 'a', 'local', ?)") |
| 2836 |
.bind(node_root.path().to_string_lossy().into_owned()) |
| 2837 |
.execute(&state.pool) |
| 2838 |
.await |
| 2839 |
.unwrap(); |
| 2840 |
|
| 2841 |
seed_version(&state.pool, "3.0.0").await; |
| 2842 |
let staged = format!( |
| 2843 |
"{}/releases/abc123abc123abc1", |
| 2844 |
node_root.path().to_string_lossy() |
| 2845 |
); |
| 2846 |
let b = seed_build(&state.pool, "shaB", "3.0.0", &staged).await; |
| 2847 |
sqlx::query( |
| 2848 |
"UPDATE tier_state SET current_version='3.0.0', current_build_id=? WHERE tier='host'", |
| 2849 |
) |
| 2850 |
.bind(b) |
| 2851 |
.execute(&state.pool) |
| 2852 |
.await |
| 2853 |
.unwrap(); |
| 2854 |
|
| 2855 |
let pool = state.pool.clone(); |
| 2856 |
let Json(body) = promote_inner(state, "a".into(), PromoteBody::default()) |
| 2857 |
.await |
| 2858 |
.expect("identity-path promote succeeds"); |
| 2859 |
assert_eq!(body["version"], "3.0.0"); |
| 2860 |
|
| 2861 |
|
| 2862 |
let (cur_v, cur_b): (Option<String>, Option<i64>) = sqlx::query_as( |
| 2863 |
"SELECT current_version, current_build_id FROM tier_state WHERE tier = 'a'", |
| 2864 |
) |
| 2865 |
.fetch_one(&pool) |
| 2866 |
.await |
| 2867 |
.unwrap(); |
| 2868 |
assert_eq!(cur_v.as_deref(), Some("3.0.0")); |
| 2869 |
assert_eq!(cur_b, Some(b), "target records the promoted build id"); |
| 2870 |
|
| 2871 |
|
| 2872 |
let deploy_b: Option<i64> = sqlx::query_scalar( |
| 2873 |
"SELECT build_id FROM deploys WHERE tier = 'a' ORDER BY id DESC LIMIT 1", |
| 2874 |
) |
| 2875 |
.fetch_one(&pool) |
| 2876 |
.await |
| 2877 |
.unwrap(); |
| 2878 |
assert_eq!(deploy_b, Some(b)); |
| 2879 |
|
| 2880 |
|
| 2881 |
|
| 2882 |
let link = tokio::fs::read_link(node_root.path().join("current")) |
| 2883 |
.await |
| 2884 |
.unwrap(); |
| 2885 |
assert_eq!(link.to_string_lossy(), "releases/abc123abc123abc1"); |
| 2886 |
} |
| 2887 |
|
| 2888 |
|
| 2889 |
async fn seed_version(pool: &SqlitePool, version: &str) { |
| 2890 |
sqlx::query("INSERT INTO versions (version, git_sha, built_at, artifact_path) VALUES (?, 'sha', datetime('now'), '/tmp/x') ON CONFLICT DO NOTHING") |
| 2891 |
.bind(version).execute(pool).await.unwrap(); |
| 2892 |
} |
| 2893 |
} |
| 2894 |
|