RIPE Database OIDC rollout hit socket exhaustion

RIPE Replaced the Database Cookie. Then the Web App Ran Out of Sockets.

Why it matters

The new login reached the RIPE Database website in September. A connection leak then stopped its web application for minutes while neighbouring database services kept running.

Theo March2 November 2026Reading Time 2 minLast update: 2 November 2026

RIPE NCC said its Database web app exhausted sockets after its OIDC rollout; other Database applications were unaffected.

Summary

At RIPE 93, RIPE NCC said its Database web application ran out of sockets after the OIDC login rollout because backend connections were not fully closed.
Users were affected for a few minutes; a restart was the workaround. Other Database applications were unaffected because the web app was containerised.

A new login solved one problem at RIPE NCC and exposed a more mundane one behind the door: the Database web application did not fully close its backend connections. The web application ran out of sockets. Users felt it for a few minutes; the applications next door carried on.

Edward Shryane gave that account at the RIPE 93 Database Working Group on 29 October. The website had moved from the old site-wide cookie to OIDC in mid-September. Shryane said a fix for the connection handling already existed, deployment was still needed, and restarting the web application served as the workaround. The public account does not date the incident more precisely or say when the fix reached production.

The October member update celebrates the completed authentication change: OIDC now replaces the Crowd cookie on the Database website. The release notes put OIDC software support in production release 1.124 on 27 August. Those are two milestones, not the outage clock. The code was ready before the website cutover; the exhausted sockets surfaced after it.

The boundary of the failure matters. Shryane said the web application was containerised, so other RIPE Database applications running alongside it were unaffected. A registry operator can change a login mechanism without changing the authority rules for records, and still knock a web workflow offline through connection housekeeping. This incident was a web-application availability failure, not a reported change to a registry object.

RIPE NCC has not published an incident timestamp, affected-request count or fix deployment receipt in these records. Those three items would show exactly how long the new login operated with the connection fault and when the workaround ceased to be necessary.