pagesift_api/jobs/
mod.rs

1//! 异步任务（批量抓取和整站爬取）：进度、结果、错误，过期后清理。
2//!
3//! 给了存储（F-6.4）时任务落盘，服务重启后：已结束的任务照样能查（直到过期），
4//! 进行中的任务从断点继续。存储布局（都在 [`pagesift_store::Tasks`] 里）：
5//!
6//! - KV `job/{id}`：任务元数据（类型、状态、创建时间、原始请求、webhook、调用方）；
7//! - 列表 `job/{id}/pages`：完成的文档，按完成顺序，序号就是分页下标；
8//! - 列表 `job/{id}/errors`：页面错误；
9//! - 列表 `job/{id}/frontier`：爬取入过队的 URL 和深度，按入队顺序；
10//! - 集合 `job/{id}/done`：已完成的 URL（爬取）或下标（批量抓取）。
11//!
12//! 每个页面先写文档再记完成：崩溃时最多重复抓一个页面，不会丢页面。
13//! 不给存储时一切留在内存里，行为和以前一样。
14
15use std::collections::{BTreeMap, HashMap};
16use std::sync::atomic::AtomicBool;
17use std::sync::{Arc, Mutex as StdMutex};
18use std::time::SystemTime;
19
20use pagesift_store::Store;
21use tokio::sync::{Mutex, Semaphore};
22
23use crate::engine::{CrawlResume, Engine};
24use crate::error::{ApiError, ErrorCode};
25use crate::limits::{Caller, Limiter};
26use crate::metrics::Metrics;
27use crate::types::{CrawlRequest, JobError, ScrapeOptions};
28use crate::webhook::{Event, Hook, Webhooks};
29
30pub mod mirror;
31mod sink;
32pub mod state;
33pub(crate) mod store;
34pub mod time;
35
36pub use state::{JOB_TTL, Job, JobKind, JobState, Start, StatusPage};
37pub use time::rfc3339;
38
39use sink::JobSink;
40use store::{BatchSpec, Meta, delete_job, list_key, meta_key, to_ms};
41
42struct Inner {
43    map: Mutex<HashMap<String, Job>>,
44    store: Option<Store>,
45    limiter: Arc<Limiter>,
46    webhooks: Option<Webhooks>,
47    metrics: Arc<Metrics>,
48    /// 多服务器网关：任务存储的写入同时复制到温备（见 [`mirror`]）。
49    mirror: Option<Arc<mirror::JobMirror>>,
50}
51
52#[derive(Clone)]
53pub struct Jobs {
54    inner: Arc<Inner>,
55}
56
57impl Default for Jobs {
58    /// 只在内存里的任务表：不落盘、不限额、不发 webhook。
59    fn default() -> Self {
60        Self::new(None, Arc::default(), None, Arc::default(), None)
61    }
62}
63
64impl Jobs {
65    pub(crate) fn new(
66        store: Option<Store>,
67        limiter: Arc<Limiter>,
68        webhooks: Option<Webhooks>,
69        metrics: Arc<Metrics>,
70        mirror: Option<Arc<mirror::JobMirror>>,
71    ) -> Self {
72        Self {
73            inner: Arc::new(Inner {
74                map: Mutex::new(HashMap::new()),
75                store,
76                limiter,
77                webhooks,
78                metrics,
79                mirror,
80            }),
81        }
82    }
83
84    /// 清掉过期任务（连同存储里的数据）。
85    async fn sweep(&self) {
86        let now = SystemTime::now();
87        let expired: Vec<String> = {
88            let mut map = self.inner.map.lock().await;
89            let ids: Vec<String> = map
90                .iter()
91                .filter(|(_, j)| j.expires_at() <= now)
92                .map(|(id, _)| id.clone())
93                .collect();
94            for id in &ids {
95                map.remove(id);
96            }
97            ids
98        };
99        if let (Some(store), false) = (&self.inner.store, expired.is_empty()) {
100            let m = self.inner.mirror.clone();
101            store
102                .blocking(move |s| {
103                    for id in expired {
104                        delete_job(s, m.as_deref(), &id);
105                    }
106                })
107                .await;
108        }
109    }
110
111    /// 任务存储（监控也落在这里）。
112    pub(crate) fn store(&self) -> Option<&Store> {
113        self.inner.store.as_ref()
114    }
115
116    /// 附件写进的地方：任务存储和镜像（见 `attachments`）。
117    pub(crate) fn disk(&self) -> Option<crate::attachments::Disk> {
118        self.inner
119            .store
120            .clone()
121            .map(|store| crate::attachments::Disk {
122                store,
123                mirror: self.inner.mirror.clone(),
124            })
125    }
126
127    pub async fn get<R>(&self, id: &str, f: impl FnOnce(&Job) -> R) -> Option<R> {
128        self.sweep().await;
129        self.inner.map.lock().await.get(id).map(f)
130    }
131
132    /// 状态和从 `skip` 开始最多 `limit` 个文档。
133    pub(crate) async fn status(&self, id: &str, skip: usize, limit: usize) -> Option<StatusPage> {
134        self.sweep().await;
135        let mut page = {
136            let map = self.inner.map.lock().await;
137            let job = map.get(id)?;
138            let end = skip.saturating_add(limit).min(job.data.len());
139            StatusPage {
140                state: job.state,
141                total: job.total,
142                completed: job.completed,
143                expires_at: job.expires_at(),
144                docs: job.docs(),
145                data: job
146                    .data
147                    .get(skip.min(end)..end)
148                    .unwrap_or_default()
149                    .to_vec(),
150                summary: job.summary.clone(),
151            }
152        };
153        if let Some(store) = &self.inner.store
154            && page.data.is_empty()
155            && skip < page.docs
156        {
157            let key = list_key(id, "pages");
158            page.data = store
159                .blocking(move |s| s.tasks().list_range(key, skip as u64, limit))
160                .await
161                .unwrap_or_else(|e| {
162                    tracing::warn!(error = %e, "reading job pages failed");
163                    Vec::new()
164                })
165                .iter()
166                .filter_map(|v| serde_json::from_slice(v).ok())
167                .collect();
168        }
169        Some(page)
170    }
171
172    /// 订阅任务的变化（新文档、状态变化）。任务不存在时返回 `None`；任务过期删掉后接收端随之关闭。
173    pub(crate) async fn subscribe(&self, id: &str) -> Option<tokio::sync::watch::Receiver<()>> {
174        self.sweep().await;
175        self.inner
176            .map
177            .lock()
178            .await
179            .get(id)
180            .map(|j| j.changed.subscribe())
181    }
182
183    /// 任务的页面错误和被 robots.txt 拦下的 URL。
184    pub(crate) async fn errors(&self, id: &str) -> Option<(Vec<JobError>, Vec<String>)> {
185        self.sweep().await;
186        let (mut errors, mut robots) = {
187            let map = self.inner.map.lock().await;
188            let job = map.get(id)?;
189            (job.errors.clone(), job.robots_blocked.clone())
190        };
191        if let Some(store) = &self.inner.store {
192            // 页面错误，然后是任务级的错误（配额、失败原因、投不出去的 webhook 事件）
193            let (key, notes) = (list_key(id, "errors"), list_key(id, "notes"));
194            let stored = store
195                .blocking(move |s| {
196                    let t = s.tasks();
197                    let mut v = t.list_items(key, usize::MAX).unwrap_or_default();
198                    v.extend(t.list_items(notes, usize::MAX).unwrap_or_default());
199                    v
200                })
201                .await;
202            for v in stored {
203                if let Ok(e) = serde_json::from_slice::<JobError>(&v) {
204                    if e.code == ErrorCode::RobotsDisallowed.as_str() {
205                        robots.push(e.url.clone());
206                    }
207                    errors.push(e);
208                }
209            }
210        }
211        Some((errors, robots))
212    }
213
214    pub async fn cancel(&self, id: &str) -> bool {
215        let found = {
216            let mut map = self.inner.map.lock().await;
217            match map.get_mut(id) {
218                Some(j) => {
219                    if j.state == JobState::Scraping {
220                        j.state = JobState::Cancelled;
221                        j.touch();
222                    }
223                    true
224                }
225                None => false,
226            }
227        };
228        if found {
229            self.save_state(id).await;
230        }
231        found
232    }
233
234    pub async fn active(&self) -> Vec<String> {
235        let map = self.inner.map.lock().await;
236        map.iter()
237            .filter(|(_, j)| j.state == JobState::Scraping)
238            .map(|(id, _)| id.clone())
239            .collect()
240    }
241
242    pub(crate) async fn owned_by(&self, id: &str, caller: Option<&Caller>) -> bool {
243        self.sweep().await;
244        let map = self.inner.map.lock().await;
245        map.get(id).is_some_and(|job| match (&job.owner, caller) {
246            (Some(owner), Some(caller)) => owner.account == caller.account,
247            (None, Some(caller)) => caller.account.as_ref() == "anonymous",
248            (None, None) => true,
249            _ => false,
250        })
251    }
252
253    pub(crate) async fn active_for(&self, caller: Option<&Caller>) -> Vec<String> {
254        let map = self.inner.map.lock().await;
255        map.iter()
256            .filter(|(_, job)| {
257                job.state == JobState::Scraping
258                    && match (&job.owner, caller) {
259                        (Some(owner), Some(caller)) => owner.account == caller.account,
260                        (None, Some(caller)) => caller.account.as_ref() == "anonymous",
261                        (None, None) => true,
262                        _ => false,
263                    }
264            })
265            .map(|(id, _)| id.clone())
266            .collect()
267    }
268
269    /// 按 (类型, 状态) 统计的任务数，给 `/metrics`。
270    pub(crate) async fn counts(&self) -> BTreeMap<(&'static str, &'static str), u64> {
271        let mut out = BTreeMap::new();
272        for j in self.inner.map.lock().await.values() {
273            *out.entry((j.kind.as_str(), j.state.as_str())).or_default() += 1;
274        }
275        out
276    }
277
278    /// 登记新任务：检查调用方的并发任务数，落盘元数据，发 `started` 事件。
279    async fn register(
280        &self,
281        kind: JobKind,
282        total: usize,
283        start: Start,
284        crawl: Option<CrawlRequest>,
285        batch: Option<BatchSpec>,
286    ) -> Result<(String, Option<Hook>), ApiError> {
287        let id = uuid::Uuid::new_v4().to_string();
288        let job = Job::new(kind, total, start.owner.clone());
289        let meta = Meta {
290            kind,
291            state: JobState::Scraping,
292            created_ms: to_ms(job.created),
293            total,
294            completed: 0,
295            owner: start.owner.as_ref().map(|c| c.account.to_string()),
296            webhook: start.webhook.clone(),
297            crawl,
298            batch,
299            summary: None,
300        };
301        {
302            let mut map = self.inner.map.lock().await;
303            if let Some(c) = &start.owner {
304                if c.limits.max_concurrent_jobs.is_some() {
305                    let active = map
306                        .values()
307                        .filter(|j| {
308                            j.state == JobState::Scraping
309                                && j.owner.as_ref().is_some_and(|o| o.account == c.account)
310                        })
311                        .count();
312                    self.inner.limiter.check_jobs(c, active)?;
313                }
314                self.inner.limiter.check_quota(c)?;
315            }
316            map.insert(id.clone(), job);
317        }
318        if let Some(store) = &self.inner.store {
319            let (key, value) = (meta_key(&id), serde_json::to_vec(&meta).unwrap_or_default());
320            let m = self.inner.mirror.clone();
321            let put = move |s: &Store| mirror::write(s, m.as_deref(), |w| w.kv_put(&key, &value));
322            if let Err(e) = store.blocking(put).await {
323                tracing::warn!(job = %id, error = %e, "saving job failed; it will not survive a restart");
324            }
325        }
326        self.inner.metrics.job_started(kind.as_str());
327        tracing::info!(
328            job = %id,
329            kind = kind.as_str(),
330            account = start.owner.as_ref().map(|c| &*c.account),
331            "job started"
332        );
333        let hook = match (&self.inner.webhooks, start.webhook) {
334            (Some(w), Some(cfg)) => {
335                let hook = w.start(&id, kind.as_str(), cfg, self.inner.store.clone());
336                hook.send(Event::Started, Vec::new(), None).await;
337                Some(hook)
338            }
339            _ => None,
340        };
341        Ok((id, hook))
342    }
343
344    /// 启动批量抓取，立即返回任务 id；每个 URL 的结果和错误陆续写入任务。
345    pub async fn start_batch(
346        &self,
347        engine: Arc<dyn Engine>,
348        urls: Vec<String>,
349        opts: ScrapeOptions,
350        concurrency: usize,
351    ) -> String {
352        self.start_batch_with(engine, urls, opts, concurrency, Start::default())
353            .await
354            .expect("no limits without a caller")
355    }
356
357    pub(crate) async fn start_batch_with(
358        &self,
359        engine: Arc<dyn Engine>,
360        urls: Vec<String>,
361        opts: ScrapeOptions,
362        concurrency: usize,
363        start: Start,
364    ) -> Result<String, ApiError> {
365        let spec = BatchSpec {
366            urls,
367            options: opts,
368            concurrency,
369        };
370        let total = spec.urls.len();
371        let (id, hook) = self
372            .register(
373                JobKind::BatchScrape,
374                total,
375                start,
376                None,
377                self.inner.store.is_some().then(|| spec.clone()),
378            )
379            .await?;
380        let pending = spec.urls.into_iter().enumerate().collect();
381        self.spawn_batch(id.clone(), engine, pending, spec.options, concurrency, hook);
382        Ok(id)
383    }
384
385    fn sink(
386        &self,
387        id: &str,
388        kind: JobKind,
389        hook: Option<Hook>,
390        resume: Option<CrawlResume>,
391    ) -> Arc<JobSink> {
392        Arc::new(JobSink {
393            jobs: self.clone(),
394            id: id.to_string(),
395            kind,
396            hook,
397            resume: StdMutex::new(resume),
398            quota_hit: AtomicBool::new(false),
399            keep_summary: true,
400            store_failed: StdMutex::new(None),
401        })
402    }
403
404    pub(super) fn spawn_batch(
405        &self,
406        id: String,
407        engine: Arc<dyn Engine>,
408        pending: Vec<(usize, String)>,
409        opts: ScrapeOptions,
410        concurrency: usize,
411        hook: Option<Hook>,
412    ) {
413        let sink = self.sink(&id, JobKind::BatchScrape, hook, None);
414        let opts = Arc::new(opts);
415        tokio::spawn(async move {
416            let sem = Arc::new(Semaphore::new(concurrency.max(1)));
417            let mut tasks = tokio::task::JoinSet::new();
418            for (i, url) in pending {
419                let (sink, engine, opts, sem) =
420                    (sink.clone(), engine.clone(), opts.clone(), sem.clone());
421                tasks.spawn(async move {
422                    let _permit = sem.acquire_owned().await.expect("semaphore open");
423                    if sink.stopped().await {
424                        return;
425                    }
426                    let done = i.to_string();
427                    match engine.scrape(&url, &opts).await {
428                        Ok(doc) => sink.record_page(done, doc).await,
429                        Err(e) => sink.record_error(done, url, e).await,
430                    }
431                });
432            }
433            while tasks.join_next().await.is_some() {}
434            sink.finish(Ok(())).await;
435        });
436    }
437
438    /// 启动整站爬取，立即返回任务 id。
439    pub async fn start_crawl(&self, engine: Arc<dyn Engine>, req: CrawlRequest) -> String {
440        self.start_crawl_with(engine, req, Start::default())
441            .await
442            .expect("no limits without a caller")
443    }
444
445    pub(crate) async fn start_crawl_with(
446        &self,
447        engine: Arc<dyn Engine>,
448        req: CrawlRequest,
449        start: Start,
450    ) -> Result<String, ApiError> {
451        let (id, hook) = self
452            .register(
453                JobKind::Crawl,
454                0,
455                start,
456                self.inner.store.is_some().then(|| req.clone()),
457                None,
458            )
459            .await?;
460        self.spawn_crawl(id.clone(), engine, req, hook, None);
461        Ok(id)
462    }
463
464    pub(super) fn spawn_crawl(
465        &self,
466        id: String,
467        engine: Arc<dyn Engine>,
468        req: CrawlRequest,
469        hook: Option<Hook>,
470        resume: Option<CrawlResume>,
471    ) {
472        let mut sink = self.sink(&id, JobKind::Crawl, hook, resume);
473        if let Some(s) = Arc::get_mut(&mut sink) {
474            s.keep_summary = req.query.is_some();
475        }
476        tokio::spawn(async move {
477            let result = engine.crawl(&req, sink.clone()).await;
478            sink.finish(result.map_err(|e| (req.url.clone(), e))).await;
479        });
480    }
481}
pagesift_api/jobs/mod.rs

pagesift_api/jobs/
mod.rs
