Files
Mangalord/backend/src/crawler/nav.rs
MechaCat02 46134c8760 fix: close Chromium tabs on crawler fetch error paths
new_page() opened a tab that was only closed on the happy path; a `?` on
wait_for_nav / content() leaked it (chromiumoxide doesn't close on drop),
so errored fetches accumulated tabs in the shared browser. Add a generic,
unit-tested `close_after(close, body)` helper and wrap the three fetch
sites (chapter content, session probe, source navigate) so the tab is
closed on every exit path — including navigate's content()-read branch,
which previously leaked.

Bump to 0.124.5.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-07 21:04:50 +02:00

298 lines
12 KiB
Rust

//! Page navigation helpers — wrap `chromiumoxide` `wait_for_navigation`
//! with a timeout so a hung TLS handshake or a page that never fires
//! `load` cannot wedge a worker (or the cron metadata pass) forever.
//!
//! [`NAV_TIMEOUT`] is the global budget. Callers in the crawler use
//! [`wait_for_nav`] to get back a typed error so transient timeouts can
//! be reported separately from underlying CDP errors.
use std::time::Duration;
use chromiumoxide::error::CdpError;
use chromiumoxide::Page;
use thiserror::Error;
/// Maximum wall-clock time we'll wait for a single page navigation. A
/// healthy Chromium reaches `load` in well under a second on the target
/// site; a 30-second cap is generous enough for slow TLS handshakes on
/// the first request after a fresh process while still catching real
/// hangs before they wedge the daemon.
pub const NAV_TIMEOUT: Duration = Duration::from_secs(30);
/// Outcome of a timed-out navigation. `Timeout` is the transient signal
/// callers translate into a retry-friendly error
/// ([`crate::crawler::detect::PageError::Transient`] in the source path,
/// a context'd anyhow elsewhere). `Cdp` carries the underlying
/// chromiumoxide error unchanged.
#[derive(Debug, Error)]
pub enum NavError {
#[error("navigation timed out after {0:?}")]
Timeout(Duration),
#[error(transparent)]
Cdp(#[from] CdpError),
}
/// Wait for the page's next navigation to complete, capped at
/// [`NAV_TIMEOUT`]. Replaces bare `page.wait_for_navigation().await`
/// throughout the crawler.
pub async fn wait_for_nav(page: &Page) -> Result<(), NavError> {
match tokio::time::timeout(NAV_TIMEOUT, page.wait_for_navigation()).await {
Err(_elapsed) => Err(NavError::Timeout(NAV_TIMEOUT)),
Ok(Err(e)) => Err(NavError::Cdp(e)),
Ok(Ok(_)) => Ok(()),
}
}
/// Poll interval for [`wait_for_selector`]. 100ms is fast enough that a
/// page rendering in 200ms isn't held back noticeably, and slow enough
/// not to spam CDP with `find_element` calls on a page that's actually
/// taking its time.
const SELECTOR_POLL_INTERVAL: Duration = Duration::from_millis(100);
/// Wait until `selector` matches at least one element on `page`, or
/// `timeout` elapses. Used after a navigation to confirm a page-type-
/// specific marker is in the DOM before parsing — replaces the fixed
/// post-nav sleep that previously masked partial-render races.
///
/// chromiumoxide 0.7.0 has no built-in `wait_for_selector`, so we poll
/// `find_element` at [`SELECTOR_POLL_INTERVAL`] until success or budget
/// exhaustion. A failed `find_element` is *not* an error here — it just
/// means "not yet" — we only surface an error once the overall
/// `timeout` is up.
pub async fn wait_for_selector(
page: &Page,
selector: &str,
timeout: Duration,
) -> Result<(), NavError> {
let deadline = tokio::time::Instant::now() + timeout;
loop {
if page.find_element(selector).await.is_ok() {
return Ok(());
}
if tokio::time::Instant::now() >= deadline {
return Err(NavError::Timeout(timeout));
}
let remaining = deadline.saturating_duration_since(tokio::time::Instant::now());
let sleep_for = SELECTOR_POLL_INTERVAL.min(remaining);
tokio::time::sleep(sleep_for).await;
}
}
/// Per-page-type budget for [`wait_for_selector`]. Shorter than
/// [`NAV_TIMEOUT`] because by the time we're waiting on a selector, the
/// page has already responded — we're only absorbing post-load JS
/// finishing its row injection, which on a healthy site takes well
/// under a second.
pub const SELECTOR_TIMEOUT: Duration = Duration::from_secs(10);
/// Run `body` to completion, then run `close` — on *every* exit path,
/// including an early `?` / `return` inside `body`. chromiumoxide's `Page`
/// does **not** close its CDP target on drop, so a fetch helper that opens a
/// page with `browser.new_page(...)` and then bails via `?` on a nav /
/// content-read error would leak a browser tab for the process's lifetime.
/// Wrapping the fallible work in `close_after` — with `close` built from a
/// clone of the page — guarantees the tab is closed regardless of how `body`
/// returns.
///
/// Both arguments are pre-built futures, so this stays generic over the
/// body's return type (the anyhow and `PageError` fetch paths both use it)
/// and references no browser types — which also lets it be unit-tested
/// without standing up a real `Page`.
pub(crate) async fn close_after<R>(
close: impl std::future::Future<Output = ()>,
body: impl std::future::Future<Output = R>,
) -> R {
let result = body.await;
close.await;
result
}
impl NavError {
/// Does this navigation error indicate the underlying Chromium
/// process has died or its CDP connection has dropped? Used by the
/// dispatcher to decide whether to invalidate the
/// [`crate::crawler::browser_manager::BrowserManager`] handle so
/// the next acquire re-launches.
///
/// Both variants count: a `Timeout` past [`NAV_TIMEOUT`] is in
/// practice always either a hung CDP transport or a wedged page
/// the browser can't recover from on its own, and a `Cdp` error
/// surfacing at the navigation layer means the chromium-facing
/// channel is the failing layer.
pub fn is_likely_browser_dead(&self) -> bool {
match self {
Self::Timeout(_) => true,
Self::Cdp(_) => true,
}
}
}
/// Walk an `anyhow::Error` chain looking for typed evidence that the
/// chromium-facing layer is the failing one. Two markers count:
///
/// 1. A wrapped [`NavError`] flagged by [`NavError::is_likely_browser_dead`].
/// 2. A wrapped [`CdpError`] (via `anyhow::Error::from(CdpError)` at a
/// `Browser::new_page` call site, or any other direct CDP boundary).
///
/// Earlier versions also substring-matched the chain for "connection",
/// "closed", "channel", etc. as a fallback. That was too broad —
/// reqwest TCP-reset errors during CDN image downloads, sqlx
/// connection-pool errors, and similar non-browser failures contain
/// those words and triggered spurious chromium relaunches. The typed
/// downcasts cover every place we hand a chromium error to anyhow,
/// so the fallback is unnecessary.
pub fn anyhow_looks_browser_dead(err: &anyhow::Error) -> bool {
for cause in err.chain() {
if let Some(nav) = cause.downcast_ref::<NavError>() {
if nav.is_likely_browser_dead() {
return true;
}
}
if cause.downcast_ref::<CdpError>().is_some() {
return true;
}
}
false
}
#[cfg(test)]
mod tests {
use super::*;
use std::future::pending;
/// Sanity-check the timeout pattern used by [`wait_for_nav`]: a
/// future that never resolves must yield `Elapsed` within the
/// configured budget. We can't easily stand up a real `Page` in a
/// unit test, so we assert the underlying primitive behaves the way
/// the helper depends on.
#[tokio::test(flavor = "current_thread", start_paused = true)]
async fn timeout_elapses_on_a_future_that_never_resolves() {
let result =
tokio::time::timeout(Duration::from_millis(50), pending::<()>()).await;
assert!(result.is_err(), "expected Elapsed on a hung future");
}
#[tokio::test]
async fn close_after_runs_close_and_returns_value_on_ok() {
use std::sync::atomic::{AtomicBool, Ordering};
use std::sync::Arc;
let closed = Arc::new(AtomicBool::new(false));
let c = closed.clone();
let out: anyhow::Result<i32> = close_after(
async move { c.store(true, Ordering::SeqCst) },
async { Ok(7) },
)
.await;
assert_eq!(out.unwrap(), 7);
assert!(closed.load(Ordering::SeqCst), "close must run on the happy path");
}
#[tokio::test]
async fn close_after_runs_close_even_when_body_errs() {
use std::sync::atomic::{AtomicBool, Ordering};
use std::sync::Arc;
let closed = Arc::new(AtomicBool::new(false));
let c = closed.clone();
// This is the leak the fix targets: the body bails before it could
// close the page itself, and `close_after` must still close it.
let out: anyhow::Result<()> =
close_after(async move { c.store(true, Ordering::SeqCst) }, async {
anyhow::bail!("nav failed")
})
.await;
assert!(out.is_err());
assert!(
closed.load(Ordering::SeqCst),
"the page must be closed even when the body returns Err"
);
}
#[test]
fn nav_error_timeout_message_includes_duration() {
let e = NavError::Timeout(Duration::from_secs(30));
assert_eq!(e.to_string(), "navigation timed out after 30s");
}
#[test]
fn timeout_is_treated_as_likely_browser_dead() {
let e = NavError::Timeout(NAV_TIMEOUT);
assert!(e.is_likely_browser_dead());
}
#[test]
fn anyhow_with_nav_timeout_in_chain_is_flagged() {
let outer = NavError::Timeout(NAV_TIMEOUT);
let wrapped: anyhow::Error =
anyhow::Error::new(outer).context("wait for chapter nav");
assert!(anyhow_looks_browser_dead(&wrapped));
}
#[test]
fn anyhow_with_cdp_error_in_chain_is_flagged() {
// `Browser::new_page` errors get wrapped via
// `anyhow::Error::from(CdpError)` at the navigate / dispatch
// call sites. Walking the chain and downcasting to CdpError is
// what catches that path. Any CdpError variant counts; the
// Serde variant is the easiest to construct in a unit test.
let serde_err: serde_json::Error =
serde_json::from_str::<i32>("not a number").unwrap_err();
let cdp = CdpError::Serde(serde_err);
let wrapped: anyhow::Error =
anyhow::Error::from(cdp).context("open chapter page");
assert!(anyhow_looks_browser_dead(&wrapped));
}
#[test]
fn anyhow_with_innocuous_parse_error_is_not_flagged() {
let e: anyhow::Error =
anyhow::anyhow!("parse manga detail: chapter row regex did not match");
assert!(!anyhow_looks_browser_dead(&e));
}
#[test]
fn anyhow_with_reqwest_style_connection_message_is_not_flagged() {
// Regression: the earlier substring fallback flagged any error
// whose message contained "connection" or "closed" as browser-
// dead. A TCP reset from a CDN during image download, or a
// sqlx pool-connection error, would burn a chromium relaunch
// even though the browser is fine. Typed downcasts only —
// these untyped strings must pass through.
for msg in [
"error sending request: connection reset by peer",
"PoolTimedOut: timed out waiting for a connection",
"request to https://cdn/x.jpg: connection closed before message completed",
"transport error during image fetch",
] {
let e: anyhow::Error = anyhow::anyhow!("{msg}");
assert!(
!anyhow_looks_browser_dead(&e),
"must not flag non-browser error: {msg}"
);
}
}
/// Same sanity check as [`timeout_elapses_on_a_future_that_never_resolves`],
/// but for the [`wait_for_selector`] polling pattern: the loop must
/// surrender on `Elapsed` rather than spinning past the deadline.
#[tokio::test(flavor = "current_thread", start_paused = true)]
async fn selector_polling_pattern_surrenders_at_deadline() {
let timeout = Duration::from_millis(300);
let start = tokio::time::Instant::now();
let deadline = start + timeout;
// Simulate find_element forever returning "not found".
let mut polls = 0u32;
let result: Result<(), NavError> = loop {
polls += 1;
if tokio::time::Instant::now() >= deadline {
break Err(NavError::Timeout(timeout));
}
tokio::time::sleep(SELECTOR_POLL_INTERVAL).await;
};
assert!(matches!(result, Err(NavError::Timeout(_))));
// 300ms / 100ms poll interval ≈ 3 iterations plus the final check
// that breaks out. Allow some slack since the first poll happens
// before any sleep.
assert!(polls >= 3, "expected at least 3 poll iterations, got {polls}");
}
}