Previously—in the logs from the last session—there was a “hardlock” in effect: retryDelay: 48s. Now, the RetryInfo block is entirely absent. They have removed this timer. This implies that their load balancer no longer attempts to queue requests, nor does it instruct the client to “wait a minute and try again.” It has simply slammed the door shut.
- Rejection Speed (Instant Rejection)
Take a look at theServer-Timingheader line; it currently readsdur=302(302 milliseconds). In the previous log, this value was 780.
What this signifies at the hardware level is that the gateway no longer traverses the clusters in search of an available GPU. It doesn’t even make the attempt. Thecloudcode-paroute is now configured to instantly discard traffic (Fast Fail). This is the classic behavior of a system that has been switched into “Maintenance Mode” (a complete technical shutdown).