Fix false-positive search failures: real timeout + logged-in searches
The "has hits" searches produced false-positive failures in two ways:
- search() relied on Puppeteer's default 30s navigation timeout, which is
independent of KORAP_SEARCH_TIMEOUT, so complex queries on very large
corpora timed out at 30s even when the timeout was raised. The caller
timeout now flows into waitForNavigation and waitForFunction.
- "Logout works" ran before the searches: Mocha executes a suite's direct
it() tests before its nested describe() suites, so the top-level logout
test fired first and logged us out, leaving the searches to query the
(tiny) public corpus and report 0 hits. Logout is now wrapped in its own
suite declared after the searches, so it runs last; the searches' before
hook also asserts a required login actually succeeded.
Add an optional KORAP_VC env var to restrict the searches to a virtual
corpus (passed as cq), keeping complex queries fast enough to finish within
the timeout.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Change-Id: If72a0b9c36700ca1a546acfd4850ee447d586a38
diff --git a/Readme.md b/Readme.md
index 7af1131..20a427c 100644
--- a/Readme.md
+++ b/Readme.md
@@ -31,7 +31,8 @@
| `KORAP_PASSWORD` | `password2` | Password for KorAP login (also accepts legacy `KORAP_PWD`) |
| `KORAP_QUERIES` | `geht, [orth=geht & cmc/pos=VVFIN]` | Comma-separated list of queries to test |
| `KORAP_MIN_TOKENS_IN_CORPUS` | `100000` | Minimum expected number of tokens for corpus statistics test |
-| `KORAP_SEARCH_TIMEOUT` | `60000` | Per-query timeout in ms for the "has hits" search tests. Raise it for slow/complex queries on very large corpora to avoid false-positive timeout failures |
+| `KORAP_SEARCH_TIMEOUT` | `60000` | Per-query timeout in ms for the "has hits" search tests. Applies to the actual page navigation/result wait, so raising it genuinely helps slow/complex queries on very large corpora avoid false-positive timeout failures |
+| `KORAP_VC` | _(none)_ | Optional virtual corpus restriction applied to the "has hits" searches, passed as the corpus query (`cq`). E.g. `KORAP_VC="pubDate in 2020"`. Narrowing the corpus keeps complex queries fast enough to finish within the timeout (also accepts `VC`) |
| `KORAP_HEADLESS` | `true` | Set to `false` or `0` to run browser in UI mode (visible window) instead of headless |
| `SLACK_WEBHOOK_URL` | _(none)_ | Slack webhook URL for test failure notifications (text only) |
| `SLACK_TOKEN` | _(none)_ | Slack bot token for uploading failure screenshots |
diff --git a/lib/korap_rc.js b/lib/korap_rc.js
index 0bdf1af..e2d8a15 100644
--- a/lib/korap_rc.js
+++ b/lib/korap_rc.js
@@ -70,17 +70,38 @@
}
}
- async search(page, query) {
+ async search(page, query, options = {}) {
+ // Both the navigation and the post-navigation settle should honour the
+ // caller-supplied timeout. Without this, Puppeteer's default 30s
+ // navigation timeout silently caps the wait, so complex queries on very
+ // large corpora time out (false-positive failure) even when
+ // KORAP_SEARCH_TIMEOUT is raised. Default to 30s for backwards compat.
+ const timeout = options.timeout || 30000;
+ // Optional virtual corpus restriction (e.g. "pubDate in 2020"). When
+ // set, it is passed as the corpus query (cq) to narrow the search and
+ // keep it fast enough to finish within the timeout.
+ const vc = options.vc || "";
try {
- await page.waitForSelector("#q-field", { visible: true });
- const query_field = await page.$("#q-field");
- assert.notEqual(query_field, null, "Query field not found");
+ if (vc) {
+ // A VC can't be entered through the query field, so navigate to
+ // the search URL directly with q + cq. The session cookie is
+ // preserved, so the user stays logged in.
+ const url = new URL(this.korap_url);
+ url.searchParams.set('q', query);
+ url.searchParams.set('ql', 'poliqarp');
+ url.searchParams.set('cq', vc);
+ await page.goto(url.href, { waitUntil: 'domcontentloaded', timeout });
+ } else {
+ await page.waitForSelector("#q-field", { visible: true });
+ const query_field = await page.$("#q-field");
+ assert.notEqual(query_field, null, "Query field not found");
- await query_field.click({ clickCount: 3 });
- await page.keyboard.type(query);
- await page.keyboard.press("Enter");
+ await query_field.click({ clickCount: 3 });
+ await page.keyboard.type(query);
+ await page.keyboard.press("Enter");
- await page.waitForNavigation({ waitUntil: 'domcontentloaded' });
+ await page.waitForNavigation({ waitUntil: 'domcontentloaded', timeout });
+ }
// Wait until the results page has actually settled, then read the
// count immediately (no fixed sleep). Kalamar renders the precise
@@ -93,7 +114,7 @@
if (total && /\d/.test(total.textContent || '')) return true;
if (document.querySelectorAll('#search ol li').length > 0) return true;
return document.querySelector('#search .no-results, .no-results') !== null;
- }, { timeout: 15000, polling: 200 });
+ }, { timeout, polling: 200 });
const hits = await page.evaluate(() => {
const total = document.querySelector('#total-results');
diff --git a/test/korap-ui.js b/test/korap-ui.js
index bec14bd..aaf46c1 100644
--- a/test/korap-ui.js
+++ b/test/korap-ui.js
@@ -21,6 +21,7 @@
const KORAP_QUERIES = process.env.KORAP_QUERIES || 'geht, [orth=geht & cmc/pos=VVFIN]'
const KORAP_MIN_TOKENS_IN_CORPUS = parseInt(process.env.KORAP_MIN_TOKENS_IN_CORPUS || "100000", 10);
const KORAP_SEARCH_TIMEOUT = parseInt(process.env.KORAP_SEARCH_TIMEOUT || "60000", 10);
+const KORAP_VC = process.env.KORAP_VC || process.env.VC || "";
const NOTIFY_ON_SUCCESS = process.env.NOTIFY_ON_SUCCESS === 'true' || process.env.NOTIFY_ON_SUCCESS === '1';
const KORAP_HEADLESS = !(process.env.KORAP_HEADLESS === 'false' || process.env.KORAP_HEADLESS === '0');
const korap_rc = require('../lib/korap_rc.js').new(KORAP_URL)
@@ -332,23 +333,40 @@
describe('Running searches that should have hits', () => {
- before(async () => { await korap_rc.login(page, KORAP_LOGIN, KORAP_PWD) })
+ // The searches must run while logged in — most corpora are only
+ // available after authentication. Re-assert the session here and,
+ // when a login is required, fail loudly if it didn't take instead
+ // of silently testing the (tiny) public corpus logged out.
+ before(async () => {
+ const logged_in = await korap_rc.login(page, KORAP_LOGIN, KORAP_PWD)
+ if (KORAP_LOGIN != "") {
+ assert.isTrue(logged_in,
+ "Login is required for the has-hits searches but did not succeed; aborting so we don't test the public corpus while logged out")
+ }
+ })
KORAP_QUERIES.split(/[;,] */).forEach((query, i) => {
it('Search for "' + query + '" has hits',
(async () => {
await korap_rc.assure_glimpse_off(page)
- const hits = await korap_rc.search(page, query)
+ const hits = await korap_rc.search(page, query, { timeout: KORAP_SEARCH_TIMEOUT, vc: KORAP_VC })
hits.should.be.above(0)
- })).timeout(KORAP_SEARCH_TIMEOUT)
+ })).timeout(KORAP_SEARCH_TIMEOUT + 30000)
})
})
- ifConditionIt('Logout works',
- KORAP_LOGIN != "",
- (async () => {
- const logout_result = await korap_rc.logout(page)
- logout_result.should.be.true
- })).timeout(15000)
+ // Logout must be the LAST UI test. Mocha runs a suite's direct it()
+ // tests before its nested describe() suites, so a top-level "Logout"
+ // test would otherwise run *before* the searches above and log us out,
+ // leaving them to query the public corpus. Wrapping it in its own suite
+ // declared after the searches suite guarantees it runs last.
+ describe('Logging out', () => {
+ ifConditionIt('Logout works',
+ KORAP_LOGIN != "",
+ (async () => {
+ const logout_result = await korap_rc.logout(page)
+ logout_result.should.be.true
+ })).timeout(15000)
+ })
});
});
\ No newline at end of file