fix(selectors): redact email path segments from provider failure logs; correct table timeout guidance

Author: waleedlatif1

apps/sim/lib/selectors/server/providers/http.test.ts

@@ -1,3 +1,4 @@
+import { getMockLogger } from '@sim/testing/mocks/logger.mock'
 import { afterAll, beforeEach, describe, expect, it, vi } from 'vitest'
 import {
   SelectorConnectionUnavailableError,
@@ -96,4 +97,23 @@ describe('provider HTTP selector boundary', () => {
       expect(cancel).toHaveBeenCalledOnce()
     }
   )
+
+  it('logs a failed provider call without its query string, search text, or email path segments', async () => {
+    mockFetch.mockResolvedValueOnce(new Response('{}', { status: 404 }))
+
+    await expect(
+      fetchProviderJson(
+        "https://www.googleapis.com/calendar/v3/calendars/jane%40example.com/root/search(q='payroll')?access_token=secret"
+      )
+    ).rejects.toBeInstanceOf(SelectorOptionsUnavailableError)
+
+    expect(getMockLogger('SelectorProviderHttp').warn).toHaveBeenCalledWith(
+      'Selector provider request failed',
+      {
+        endpoint: 'www.googleapis.com/calendar/v3/calendars/{redacted}/root/search()',
+        reason: 'status',
+        status: 404,
+      }
+    )
+  })
 })

apps/sim/lib/selectors/server/providers/http.ts

@@ -48,13 +48,17 @@ async function readBoundedBody(response: Response): Promise<string> {
 }
 
 /**
- * Host and path of a provider request. The query string is dropped and OData-style path
- * arguments (`search(q='...')`) are blanked, since both can carry tokens or a user's search text.
+ * Host and path of a provider request, with the parts that can carry user data removed: the query
+ * string (tokens, filters), OData-style path arguments (`search(q='...')` holds search text), and
+ * any segment holding an email address (a Google calendar id is usually its owner's address).
  */
 function providerEndpoint(input: RequestInfo | URL): string | undefined {
   try {
     const url = new URL(input instanceof Request ? input.url : input)
-    return `${url.host}${url.pathname.replace(/\([^/]*\)/g, '()')}`
+    const path = url.pathname
+      .replace(/\([^/]*\)/g, '()')
+      .replace(/[^/]*(?:@|%40)[^/]*/gi, '{redacted}')
+    return `${url.host}${path}`
   } catch {
     return undefined
   }

apps/sim/lib/table/application/rows.ts

@@ -506,17 +506,18 @@ async function throwValidationResponse(
 }
 
 /**
- * A structured row query cancelled by its own `statement_timeout`: the query shape costs more than
- * the table's size allows. Sorting or range-filtering on a stored column reads every row's data, so it is the
- * caller's to narrow, and retrying the same query only times out again.
+ * A structured row query cancelled by its own `statement_timeout`. Sorting, range-filtering, or
+ * counting on a stored column reads every row's full data, so on a large table the query only
+ * finishes when that data happens to be cached. The fix is the caller's (narrow the query), so this
+ * is a 400 with an actionable message rather than an opaque 500.
  */
 function rethrowReadQueryError(error: unknown, tableId: string): never {
   if (getPostgresCancellationReason(error) === 'statement_timeout') {
     logger.warn('Table read hit its statement timeout', { tableId })
     throw new TableRowsValidationError(
       `The query took longer than ${READ_STATEMENT_TIMEOUT_MS / 1000} seconds and was stopped. ` +
-        'Sorting or range-filtering on a column reads every row in the table, which is too slow ' +
-        'at this table size. Sort by createdAt, add an equality filter, or keep ' +
+        'Sorting, range-filtering, or counting on a column reads every row in the table, which ' +
+        'is too slow at this table size. Sort by createdAt, add an equality filter, or keep ' +
         'fewer rows (an Expiration column deletes old rows automatically).',
       { code: 'TABLE_QUERY_TIMEOUT' }
     )