Increased Subscribe error rate and latency observed globally

Update timeline

1. resolved Oct 09, 2026, 06:45 PM UTC

   Starting at 17:32 UTC on October 9, 2026 we observed elevated 5XX error rates and latency on the Subscribe endpoint globally. The issue was resolved by 18:18 UTC and services have returned to normal levels. A limited number of customers may have been affected. A root cause analysis will be published in the coming days. We sincerely apologize for any impact this may have caused. If you feel you were affected and wish to discuss, please email [email protected].

2. postmortem Oct 14, 2026, 05:19 PM UTC

   ### **Problem Description, Impact, and Resolution** On October 9, 2026, between 17:30 and 18:20 UTC, we observed elevated HTTP 5xx errors affecting the Subscribe endpoint for a subset of customers. Some subscribe requests may have failed, and a subset of messages may have been delayed or not delivered in real time. No messages were lost, and customers with Message Persistence enabled could retrieve affected messages from History. The incident was caused by an unexpected increase in traffic from a high-volume load test against a development keyset that did not have the expected limits set. The resulting load temporarily exhausted memory in some shared infrastructure, which caused the issue. We stabilized the affected infrastructure by applying a rate limit to the keyset generating the excess load. Service returned to normal by 18:20 UTC. ‌ ### **Mitigation Steps and Recommended Future Preventative Measures** We applied a temporary publish rate limit to contain the unexpected traffic and restore service while preserving the affected keyset’s availability. To help prevent similar incidents, PubNub is adding additional checks to ensure that safeguards are correctly in place to detect and automatically limit unusually heavy load before it can affect other traffic sharing the same infrastructure. We are also updating operational procedures for subscriber memory and autoscaling changes to ensure that capacity adjustments remain effective during incident mitigation.