Possible recurrence in syd1 of the edge issue reported in thread #28121.
Our Vercel syd1 service calls a Fly-hosted backend. Identical requests alternate between success and timeout: 566 of 1,076 failed with
ETIMEDOUT. Failures continued when each healthy Fly machine was tested separately.
Could Fly check whether a Sydney edge or certificate path is failing?
Do you have a source IP of your Vercel service in syd making those requests?
We could not capture the exact source IP of the failed request. We sent a follow-up request to debug.fly.dev from the same Vercel function execution after each timeout.
Of 94 diagnostic requests, 85 succeeded and nine also failed. The successful requests showed AWS Sydney egress and entered Fly through Tokyo 84 times and Singapore once. Example diagnostic source IPs were 3.26.75.191, 3.107.185.178, and 54.66.180.12.
Because each diagnostic used a new connection, these may not be the exact source IPs used by the failed requests.
The sustained failures have since stopped without a networking change from us, so we do not know why the connection recovered. Some intermittent timeouts remained in our latest sample, but every affected message succeeded after a retry.