On March 25th, 2026, our Frankfurt region experienced an outage caused by a gateway issue. We want to take a moment to first apologize to our Frankfurt customers. You trust Meilisearch to provide a service that is critical to your business and we understand that any incident can have a major impact. We are deeply sorry about that, and we want to provide more information about what happened and share the next steps.
On March 25th, 2026, between 10:22 and 13:00 UTC, customers in the Frankfurt region experienced complete service unavailability. All requests to Meilisearch instances returned SSL connection errors for approximately 2 hours and 30 minutes.
To give some context, we’re using the Kong open source version to serve requests to your Meilisearch instances and provide various features (including analytics collection). To ensure scalability, we have two components running on Kubernetes:
On March 25th, 2026 at 10:22 UTC, the gateway started returning SSL errors due to an invalid configuration used to serve requests. At the same time, the KIC was getting rejected when trying to push any configuration to the gateway for unspecified reasons.
This was actually caused by an upgrade from the KIC that was shipped in a patch version 3.4.12 on March 24th, 2026, with a breaking change on the configuration generation that was incompatible for users of the latest OSS gateway version. The upgrade from 3.4.11 to 3.4.12 went unnoticed because we were using the official Helm chart, which specifies the default tag value 3.4. As a result, when Kong scaled up on a different set of nodes, it pulled the latest 3.4.12 version.
The issue was first addressed by narrowing down the problem to a specific set of unrelated Kong consumers that were causing the configuration rejection. After removing the set of consumers, we were able to re-create all customer routes and resume service.
The issue was then permanently fixed by enforcing a specific version of the KIC to revert back to 3.4.11 and prevent any future silent upgrade.
While we are disappointed with this breaking change published as a patch version by the Kong community, which ultimately caused the downtime, we take full responsibility and want to again apologize. You trust us to have your Meilisearch instances up and running, and it is our duty to address that with architectural decisions that align with that goal.