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::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;
33mod 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    pub async fn get<R>(&self, id: &str, f: impl FnOnce(&Job) -> R) -> Option<R> {
112        self.sweep().await;
113        self.inner.map.lock().await.get(id).map(f)
114    }
115
116    /// 状态和从 `skip` 开始最多 `limit` 个文档。
117    pub(crate) async fn status(&self, id: &str, skip: usize, limit: usize) -> Option<StatusPage> {
118        self.sweep().await;
119        let mut page = {
120            let map = self.inner.map.lock().await;
121            let job = map.get(id)?;
122            let end = skip.saturating_add(limit).min(job.data.len());
123            StatusPage {
124                state: job.state,
125                total: job.total,
126                completed: job.completed,
127                expires_at: job.expires_at(),
128                docs: job.docs(),
129                data: job
130                    .data
131                    .get(skip.min(end)..end)
132                    .unwrap_or_default()
133                    .to_vec(),
134                summary: job.summary.clone(),
135            }
136        };
137        if let Some(store) = &self.inner.store
138            && page.data.is_empty()
139            && skip < page.docs
140        {
141            let key = list_key(id, "pages");
142            page.data = store
143                .blocking(move |s| s.tasks().list_range(key, skip as u64, limit))
144                .await
145                .unwrap_or_else(|e| {
146                    tracing::warn!(error = %e, "reading job pages failed");
147                    Vec::new()
148                })
149                .iter()
150                .filter_map(|v| serde_json::from_slice(v).ok())
151                .collect();
152        }
153        Some(page)
154    }
155
156    /// 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。
157    pub(crate) async fn subscribe(&self, id: &str) -> Option<tokio::sync::watch::Receiver<()>> {
158        self.sweep().await;
159        self.inner
160            .map
161            .lock()
162            .await
163            .get(id)
164            .map(|j| j.changed.subscribe())
165    }
166
167    /// 任务的页面错误和被 robots.txt 拦下的 URL。
168    pub(crate) async fn errors(&self, id: &str) -> Option<(Vec<JobError>, Vec<String>)> {
169        self.sweep().await;
170        let (mut errors, mut robots) = {
171            let map = self.inner.map.lock().await;
172            let job = map.get(id)?;
173            (job.errors.clone(), job.robots_blocked.clone())
174        };
175        if let Some(store) = &self.inner.store {
176            let key = list_key(id, "errors");
177            let stored = store
178                .blocking(move |s| s.tasks().list_items(key, usize::MAX))
179                .await
180                .unwrap_or_default();
181            for v in stored {
182                if let Ok(e) = serde_json::from_slice::<JobError>(&v) {
183                    if e.code == ErrorCode::RobotsDisallowed.as_str() {
184                        robots.push(e.url.clone());
185                    }
186                    errors.push(e);
187                }
188            }
189        }
190        Some((errors, robots))
191    }
192
193    pub async fn cancel(&self, id: &str) -> bool {
194        let found = {
195            let mut map = self.inner.map.lock().await;
196            match map.get_mut(id) {
197                Some(j) => {
198                    if j.state == JobState::Scraping {
199                        j.state = JobState::Cancelled;
200                        j.touch();
201                    }
202                    true
203                }
204                None => false,
205            }
206        };
207        if found {
208            self.save_state(id).await;
209        }
210        found
211    }
212
213    pub async fn active(&self) -> Vec<String> {
214        let map = self.inner.map.lock().await;
215        map.iter()
216            .filter(|(_, j)| j.state == JobState::Scraping)
217            .map(|(id, _)| id.clone())
218            .collect()
219    }
220
221    /// 按 (类型, 状态) 统计的任务数，给 `/metrics`。
222    pub(crate) async fn counts(&self) -> BTreeMap<(&'static str, &'static str), u64> {
223        let mut out = BTreeMap::new();
224        for j in self.inner.map.lock().await.values() {
225            *out.entry((j.kind.as_str(), j.state.as_str())).or_default() += 1;
226        }
227        out
228    }
229
230    /// 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。
231    async fn register(
232        &self,
233        kind: JobKind,
234        total: usize,
235        start: Start,
236        crawl: Option<CrawlRequest>,
237        batch: Option<BatchSpec>,
238    ) -> Result<(String, Option<Hook>), ApiError> {
239        let id = uuid::Uuid::new_v4().to_string();
240        let job = Job::new(kind, total, start.owner.clone());
241        let meta = Meta {
242            kind,
243            state: JobState::Scraping,
244            created_ms: to_ms(job.created),
245            total,
246            completed: 0,
247            owner: start.owner.as_ref().map(|c| c.account.to_string()),
248            webhook: start.webhook.clone(),
249            crawl,
250            batch,
251            summary: None,
252        };
253        {
254            let mut map = self.inner.map.lock().await;
255            if let Some(c) = &start.owner {
256                if c.limits.max_concurrent_jobs.is_some() {
257                    let active = map
258                        .values()
259                        .filter(|j| {
260                            j.state == JobState::Scraping
261                                && j.owner.as_ref().is_some_and(|o| o.account == c.account)
262                        })
263                        .count();
264                    self.inner.limiter.check_jobs(c, active)?;
265                }
266                self.inner.limiter.check_quota(c)?;
267            }
268            map.insert(id.clone(), job);
269        }
270        if let Some(store) = &self.inner.store {
271            let (key, value) = (meta_key(&id), serde_json::to_vec(&meta).unwrap_or_default());
272            let m = self.inner.mirror.clone();
273            let put = move |s: &Store| mirror::write(s, m.as_deref(), |w| w.kv_put(&key, &value));
274            if let Err(e) = store.blocking(put).await {
275                tracing::warn!(job = %id, error = %e, "saving job failed; it will not survive a restart");
276            }
277        }
278        self.inner.metrics.job_started(kind.as_str());
279        tracing::info!(
280            job = %id,
281            kind = kind.as_str(),
282            account = start.owner.as_ref().map(|c| &*c.account),
283            "job started"
284        );
285        let hook = match (&self.inner.webhooks, start.webhook) {
286            (Some(w), Some(cfg)) => {
287                let hook = w.start(&id, kind.as_str(), cfg, self.inner.store.clone());
288                hook.send(Event::Started, Vec::new(), None).await;
289                Some(hook)
290            }
291            _ => None,
292        };
293        Ok((id, hook))
294    }
295
296    /// 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。
297    pub async fn start_batch(
298        &self,
299        engine: Arc<dyn Engine>,
300        urls: Vec<String>,
301        opts: ScrapeOptions,
302        concurrency: usize,
303    ) -> String {
304        self.start_batch_with(engine, urls, opts, concurrency, Start::default())
305            .await
306            .expect("no limits without a caller")
307    }
308
309    pub(crate) async fn start_batch_with(
310        &self,
311        engine: Arc<dyn Engine>,
312        urls: Vec<String>,
313        opts: ScrapeOptions,
314        concurrency: usize,
315        start: Start,
316    ) -> Result<String, ApiError> {
317        let spec = BatchSpec {
318            urls,
319            options: opts,
320            concurrency,
321        };
322        let total = spec.urls.len();
323        let (id, hook) = self
324            .register(
325                JobKind::BatchScrape,
326                total,
327                start,
328                None,
329                self.inner.store.is_some().then(|| spec.clone()),
330            )
331            .await?;
332        let pending = spec.urls.into_iter().enumerate().collect();
333        self.spawn_batch(id.clone(), engine, pending, spec.options, concurrency, hook);
334        Ok(id)
335    }
336
337    fn sink(
338        &self,
339        id: &str,
340        kind: JobKind,
341        hook: Option<Hook>,
342        resume: Option<CrawlResume>,
343    ) -> Arc<JobSink> {
344        Arc::new(JobSink {
345            jobs: self.clone(),
346            id: id.to_string(),
347            kind,
348            hook,
349            resume: StdMutex::new(resume),
350            quota_hit: AtomicBool::new(false),
351            keep_summary: true,
352        })
353    }
354
355    pub(super) fn spawn_batch(
356        &self,
357        id: String,
358        engine: Arc<dyn Engine>,
359        pending: Vec<(usize, String)>,
360        opts: ScrapeOptions,
361        concurrency: usize,
362        hook: Option<Hook>,
363    ) {
364        let sink = self.sink(&id, JobKind::BatchScrape, hook, None);
365        let opts = Arc::new(opts);
366        tokio::spawn(async move {
367            let sem = Arc::new(Semaphore::new(concurrency.max(1)));
368            let mut tasks = tokio::task::JoinSet::new();
369            for (i, url) in pending {
370                let (sink, engine, opts, sem) =
371                    (sink.clone(), engine.clone(), opts.clone(), sem.clone());
372                tasks.spawn(async move {
373                    let _permit = sem.acquire_owned().await.expect("semaphore open");
374                    if sink.stopped().await {
375                        return;
376                    }
377                    let done = i.to_string();
378                    match engine.scrape(&url, &opts).await {
379                        Ok(doc) => sink.record_page(done, doc).await,
380                        Err(e) => sink.record_error(done, url, e).await,
381                    }
382                });
383            }
384            while tasks.join_next().await.is_some() {}
385            sink.finish(Ok(())).await;
386        });
387    }
388
389    /// 启动整站爬取，立即返回任务 id。
390    pub async fn start_crawl(&self, engine: Arc<dyn Engine>, req: CrawlRequest) -> String {
391        self.start_crawl_with(engine, req, Start::default())
392            .await
393            .expect("no limits without a caller")
394    }
395
396    pub(crate) async fn start_crawl_with(
397        &self,
398        engine: Arc<dyn Engine>,
399        req: CrawlRequest,
400        start: Start,
401    ) -> Result<String, ApiError> {
402        let (id, hook) = self
403            .register(
404                JobKind::Crawl,
405                0,
406                start,
407                self.inner.store.is_some().then(|| req.clone()),
408                None,
409            )
410            .await?;
411        self.spawn_crawl(id.clone(), engine, req, hook, None);
412        Ok(id)
413    }
414
415    pub(super) fn spawn_crawl(
416        &self,
417        id: String,
418        engine: Arc<dyn Engine>,
419        req: CrawlRequest,
420        hook: Option<Hook>,
421        resume: Option<CrawlResume>,
422    ) {
423        let mut sink = self.sink(&id, JobKind::Crawl, hook, resume);
424        if let Some(s) = Arc::get_mut(&mut sink) {
425            s.keep_summary = req.query.is_some();
426        }
427        tokio::spawn(async move {
428            let result = engine.crawl(&req, sink.clone()).await;
429            sink.finish(result.map_err(|e| (req.url.clone(), e))).await;
430        });
431    }
432}
pagesift_api/jobs/mod.rs

pagesift_api/jobs/
mod.rs
