Between 14:35 and 16:30 SAST on Thursday 10 May, SAFIRE’s federation hub experienced an unexpected failure that would have prevented any users from logging into federated services.
The cause of the failure was some invalid metadata imported from another federation via eduGAIN. The faulty metadata was imported by our aggregator at about 10AM on 9 May and shortly afterwards, SAFIRE’s hub to refuse to load it. Transient failures like this are not uncommon in a system that depends on metadata from may other places, and the hub automatically relied on cached copies for continued operations.
Shortly after 12:30 on 10 May, our monitoring system detected that the hub’s metadata had still not been refreshed and that the cache was depleting. This prompted an investigation into the root cause, and it was discovered that a SAML profile error was preventing the hub from loading metadata.
Unfortunately attempting to determine which of the approximately 3850 entities contained the error caused the cached metadata to expire prematurely at 14:35. It then took several hours to uncover the cause and restore functional metadata, during which time the federation hub had no operational metadata and thus users would have been unable to log into any federated service.
At 16:30 the problem was discovered and cached metadata was restored. However, the problematic entity remained in the source metadata and it was another hour or so before it could be filtered out.
Contact was made with the home federation of the problematic entity to alert them to the issue. In addition, the filter mechanism described above should prevent the same error from recurring (it filters on the metadata problem rather than the specific entity).