Pranay Mule
06/30/2026, 12:21 PMkapa.ai
06/30/2026, 12:21 PMsource-posthog connector that can cause partial or missing data across multiple dates. Here's a breakdown:
Root Causes 1. PostHog's New 1000-Row Per-Request Cap (Primary Cause) On 2026-05-26, PostHog hard-capped the /api/projects/:project_id/events/ endpoint at 1000 events per request (previously 50,000). The old connector used OffsetIncrement pagination with page_size: 10000, which stops when a page returns fewer rows than the page size. Since every page now returns ā¤1000 rows, the connector stopped after the first page of each time slice, silently dropping the rest while reporting SUCCEEDED. [PR #78536]
2. Broken Pagination Following next URLs The events paginator was not correctly following PostHog's next URL for keyset (timestamp-based) pagination. Instead, it kept re-injecting the original slice's after and before parameters on paginated requests, causing:
⢠Misleading 403 errors (PostHog interpreted repeated requests as needing query:read scope)
⢠Pagination stopping prematurely or looping incorrectly [PR #78128]
3. Offset Pagination Failures The connector previously used offset-based pagination, which PostHog has deprecated. Offsets above 100,000 fail with HTTP 500 errors and Cloudflare 524 timeouts, causing syncs to fail or return incomplete data for high-volume date ranges. [GitHub Issue #60293]
4. Missing query:read Scope If your API key lacks the query:read scope, the events stream silently fails during sync (the connection check only validated projects, not events). [PR #79115]
Recommended Fixes Upgrade to the Latest Connector Version The fixes have been merged across several PRs:
Verify API Key Scopes Ensure your PostHog Personal API Key has both project:read and query:read scopes. [PR #79115]
Check Connector Version The fixes require airbyte-cdk >= 1.2.1 and connector version >= 1.1.26. If you're on 1.1.25 or earlier, upgrade the connector. [PR #78536]
Important Note PostHog considers the /events/ API deprecated and recommends migrating to their Query/HogQL API or Batch Exports for high-volume use cases. The current connector fixes work within the deprecated endpoint's constraints, but edge cases (e.g., >1000 events sharing the same microsecond timestamp at a page boundary) may still cause minor gaps. [GitHub Issue #60293] š 0 button š 0 button ā¹ļø Tip: Mention @kapa.ai in the thread for follow-ups or DM the app for 1-to-1 help., with interactive elements