27/08/2026 13:15 GMT: Canary.CDN upgrade reverted

The cluster is reverted back to working version 1.0.469. 100% of frontends is operational.

27/08/2026 10:20 GMT: Canary.CDN upgrade - reverting

Our operations team decided to revert the upgrade, as it's impossible to continue it safely. We will provide more information as soon as the situation will stabilize.

27/08/2026 08:40 GMT: Canary.CDN upgrade - disruption on some frontends - update

We learned that the cluster is not able to stabilize by itself. Our operators are now fixing the situation live on the cluster in order to unblock processing.

27/08/2026 08:25 GMT: Canary.CDN upgrade - disruption on some frontends - update

The amount of affected frontends is 3.78% (3.87% it was miscalculated), random from the cluster.

The system is processing in order to put back all in proper state.

27/08/2026 06:50 GMT: Canary.CDN upgrade - disruption on some frontends

Note: The time was updated from 07:05 GMT to 06:50 GMT as new data regarding unavailabilty arrived.

We've detected that some shared frontends have been turned off during the upgrade. We stopped part of the processing in order to not escalate the situation. As soon as master node will be upgraded, we will revert back the situation. We are aware of the problem for the Premium.CDN upgrade and also we are going to work on fully fixing it.

27/08/2026 06:45 GMT: Canary.CDN upgrade starts

The upgrade starts now.

27/08/2026: Canary.CDN upgrade

The cluster will be upgraded to rapid-cdn 1.0.506, delivering every change since 1.0.469. Please consult the changelog for details.

• Custom per-site 502/503/504 error pages
• Accept-Encoding normalisation, and gzip compression by default for clients that send none
• HTTP/2 and HTTP/3 handling fixes
• Slave parameters serialised as XML (custom-personal, default)
• Cluster error-page manager and passive slave-validation audit database
• New parameters: instance-retention-delay, dns-nameserver, monitor-interface-url
• Per-site error-page editor hardening: an empty save is rejected instead of blanking a served page, and edits typed into the other codes' fields are preserved
• Per-site 502/503/504 pages are also served on failover; redirect sites are no longer offered an error-page upload URL
• Error Page Manager rewritten for resilience and scale: threaded request handling, and per-site publication cost follows the number of overrides
• Node-level schemas accept unknown parameters, allowing mixed-release clusters during upgrades with fast rollback
• HAProxy upgraded to the 3.4 LTS branch (3.4.2, PCRE2); backend SSL verification for authenticate-to-backend kept chain-only
• Failed site publication to the master retries before alarming; a self-healing ticket is opened only after five consecutive failures
• Builds are CPU-portable again, and the release refuses to build a CPU-specific openssl

The cluster is going to be upgraded.