Date: April 9 – April 14, 2026 Duration: ~87 hours (peak impact ~24 hours) Severity: Major
Between April 9 and April 14, 2026, LinkedIn PDF carousel posts published through Buffer began failing. About 3,500 posts failed before we identified the cause and deployed a fix. Other LinkedIn post types and other platforms were unaffected.
For most of April 9 and the morning of April 10, most LinkedIn PDF carousel posts failed to publish. Failure rates dropped after the peak but stayed above normal through the weekend. Because LinkedIn marked the failed uploads on their side, these posts couldn't be retried automatically — customers had to recreate them from scratch. Around 40 support messages came in during the first 12 hours. Only PDF carousel posts on LinkedIn were affected; everything else continued to work normally.
The cause was a combination of an unannounced change on LinkedIn's side and a change we had made months earlier. In January, we changed how Buffer uploads documents to LinkedIn. It worked fine for three months. Around April 8, LinkedIn quietly changed how their service processed those uploads, and the documents we sent started being marked as failed on their end. The three-month gap between the two changes made the connection non-obvious — when failures started, we checked recent code and found nothing, which initially led us to assume the issue was entirely on LinkedIn's end.
We posted a status page update, activated our support workflows, and began investigating. After ruling out recent code changes, we contacted LinkedIn developer support. The turning point came when we posted the same PDF through another tool and it succeeded — that told us the issue was specific to how we were uploading, not a LinkedIn-wide outage. From there we connected the failures to our earlier change, switched the upload method back to the older approach, deployed the fix, and verified recovery. Affected customers needed to recreate their failed posts.
End-to-end testing would have caught this. Our existing tests confirmed uploads were succeeding, but didn't verify the full lifecycle through to a post being available. A test that checks all the way to a successful post would catch this kind of breakage.
Comparing against another tool earlier would have saved time. We spent the first two days assuming this was entirely a LinkedIn issue. Testing the same content through another tool was what redirected us.
We need earlier automated detection. Only PDF carousels were failing while everything else worked. Alerting on success rates broken down by content type would have flagged this within minutes, rather than the 27+ hours it took.
Weekend incidents need a clearer path to subject matter experts so the right engineering knowledge can be brought in faster.
What worked. Customer communications stayed continuous across the multi-day incident through a Comms Lead rotation, status page wording was refined for accuracy, and we waited for engineering verification before announcing the fix.
