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;
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    /// 按 (类型, 状态) 统计的任务数，给 `/metrics`。
243    pub(crate) async fn counts(&self) -> BTreeMap<(&'static str, &'static str), u64> {
244        let mut out = BTreeMap::new();
245        for j in self.inner.map.lock().await.values() {
246            *out.entry((j.kind.as_str(), j.state.as_str())).or_default() += 1;
247        }
248        out
249    }
250
251    /// 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。
252    async fn register(
253        &self,
254        kind: JobKind,
255        total: usize,
256        start: Start,
257        crawl: Option<CrawlRequest>,
258        batch: Option<BatchSpec>,
259    ) -> Result<(String, Option<Hook>), ApiError> {
260        let id = uuid::Uuid::new_v4().to_string();
261        let job = Job::new(kind, total, start.owner.clone());
262        let meta = Meta {
263            kind,
264            state: JobState::Scraping,
265            created_ms: to_ms(job.created),
266            total,
267            completed: 0,
268            owner: start.owner.as_ref().map(|c| c.account.to_string()),
269            webhook: start.webhook.clone(),
270            crawl,
271            batch,
272            summary: None,
273        };
274        {
275            let mut map = self.inner.map.lock().await;
276            if let Some(c) = &start.owner {
277                if c.limits.max_concurrent_jobs.is_some() {
278                    let active = map
279                        .values()
280                        .filter(|j| {
281                            j.state == JobState::Scraping
282                                && j.owner.as_ref().is_some_and(|o| o.account == c.account)
283                        })
284                        .count();
285                    self.inner.limiter.check_jobs(c, active)?;
286                }
287                self.inner.limiter.check_quota(c)?;
288            }
289            map.insert(id.clone(), job);
290        }
291        if let Some(store) = &self.inner.store {
292            let (key, value) = (meta_key(&id), serde_json::to_vec(&meta).unwrap_or_default());
293            let m = self.inner.mirror.clone();
294            let put = move |s: &Store| mirror::write(s, m.as_deref(), |w| w.kv_put(&key, &value));
295            if let Err(e) = store.blocking(put).await {
296                tracing::warn!(job = %id, error = %e, "saving job failed; it will not survive a restart");
297            }
298        }
299        self.inner.metrics.job_started(kind.as_str());
300        tracing::info!(
301            job = %id,
302            kind = kind.as_str(),
303            account = start.owner.as_ref().map(|c| &*c.account),
304            "job started"
305        );
306        let hook = match (&self.inner.webhooks, start.webhook) {
307            (Some(w), Some(cfg)) => {
308                let hook = w.start(&id, kind.as_str(), cfg, self.inner.store.clone());
309                hook.send(Event::Started, Vec::new(), None).await;
310                Some(hook)
311            }
312            _ => None,
313        };
314        Ok((id, hook))
315    }
316
317    /// 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。
318    pub async fn start_batch(
319        &self,
320        engine: Arc<dyn Engine>,
321        urls: Vec<String>,
322        opts: ScrapeOptions,
323        concurrency: usize,
324    ) -> String {
325        self.start_batch_with(engine, urls, opts, concurrency, Start::default())
326            .await
327            .expect("no limits without a caller")
328    }
329
330    pub(crate) async fn start_batch_with(
331        &self,
332        engine: Arc<dyn Engine>,
333        urls: Vec<String>,
334        opts: ScrapeOptions,
335        concurrency: usize,
336        start: Start,
337    ) -> Result<String, ApiError> {
338        let spec = BatchSpec {
339            urls,
340            options: opts,
341            concurrency,
342        };
343        let total = spec.urls.len();
344        let (id, hook) = self
345            .register(
346                JobKind::BatchScrape,
347                total,
348                start,
349                None,
350                self.inner.store.is_some().then(|| spec.clone()),
351            )
352            .await?;
353        let pending = spec.urls.into_iter().enumerate().collect();
354        self.spawn_batch(id.clone(), engine, pending, spec.options, concurrency, hook);
355        Ok(id)
356    }
357
358    fn sink(
359        &self,
360        id: &str,
361        kind: JobKind,
362        hook: Option<Hook>,
363        resume: Option<CrawlResume>,
364    ) -> Arc<JobSink> {
365        Arc::new(JobSink {
366            jobs: self.clone(),
367            id: id.to_string(),
368            kind,
369            hook,
370            resume: StdMutex::new(resume),
371            quota_hit: AtomicBool::new(false),
372            keep_summary: true,
373            store_failed: StdMutex::new(None),
374        })
375    }
376
377    pub(super) fn spawn_batch(
378        &self,
379        id: String,
380        engine: Arc<dyn Engine>,
381        pending: Vec<(usize, String)>,
382        opts: ScrapeOptions,
383        concurrency: usize,
384        hook: Option<Hook>,
385    ) {
386        let sink = self.sink(&id, JobKind::BatchScrape, hook, None);
387        let opts = Arc::new(opts);
388        tokio::spawn(async move {
389            let sem = Arc::new(Semaphore::new(concurrency.max(1)));
390            let mut tasks = tokio::task::JoinSet::new();
391            for (i, url) in pending {
392                let (sink, engine, opts, sem) =
393                    (sink.clone(), engine.clone(), opts.clone(), sem.clone());
394                tasks.spawn(async move {
395                    let _permit = sem.acquire_owned().await.expect("semaphore open");
396                    if sink.stopped().await {
397                        return;
398                    }
399                    let done = i.to_string();
400                    match engine.scrape(&url, &opts).await {
401                        Ok(doc) => sink.record_page(done, doc).await,
402                        Err(e) => sink.record_error(done, url, e).await,
403                    }
404                });
405            }
406            while tasks.join_next().await.is_some() {}
407            sink.finish(Ok(())).await;
408        });
409    }
410
411    /// 启动整站爬取，立即返回任务 id。
412    pub async fn start_crawl(&self, engine: Arc<dyn Engine>, req: CrawlRequest) -> String {
413        self.start_crawl_with(engine, req, Start::default())
414            .await
415            .expect("no limits without a caller")
416    }
417
418    pub(crate) async fn start_crawl_with(
419        &self,
420        engine: Arc<dyn Engine>,
421        req: CrawlRequest,
422        start: Start,
423    ) -> Result<String, ApiError> {
424        let (id, hook) = self
425            .register(
426                JobKind::Crawl,
427                0,
428                start,
429                self.inner.store.is_some().then(|| req.clone()),
430                None,
431            )
432            .await?;
433        self.spawn_crawl(id.clone(), engine, req, hook, None);
434        Ok(id)
435    }
436
437    pub(super) fn spawn_crawl(
438        &self,
439        id: String,
440        engine: Arc<dyn Engine>,
441        req: CrawlRequest,
442        hook: Option<Hook>,
443        resume: Option<CrawlResume>,
444    ) {
445        let mut sink = self.sink(&id, JobKind::Crawl, hook, resume);
446        if let Some(s) = Arc::get_mut(&mut sink) {
447            s.keep_summary = req.query.is_some();
448        }
449        tokio::spawn(async move {
450            let result = engine.crawl(&req, sink.clone()).await;
451            sink.finish(result.map_err(|e| (req.url.clone(), e))).await;
452        });
453    }
454}
pagesift_api/jobs/mod.rs

pagesift_api/jobs/
mod.rs
