From March 27th 7 am UTC, a cutomer's APIcast keep logging upstream prematurely closed connection while reading response header from upstream error frequently.
From March 27th to April 1st, they got 60-120 times per hour.

2019/04/01 10:23:40 [error] 16977#16977: *7 upstream prematurely closed connection while reading response header from upstream, client: <client ip address>, server: _, request: "GET /example HTTP/1.1", subrequest: "/transactions/authrep.xml", upstream: "https://<ip address of 3scale management api>:443/transactions/authrep.xml?service_token=<SERVICE_TOKEN>&service_id=<SERVICE_ID>&usage%5Bhits%5D=1&user_key=<USER_KEY>", host: "<host ip address>"

However, the error seemed be recovered by following request; it was logged as debug level logs.
Getting the error 60-120 times per hour is too many; it makes difficult to monitor the log.

Regarding auth endpoint, this issue is fixed temporarily:

https://github.com/3scale/APIcast/blob/cfa12cb0b66ac2635ba1c56f1985a374b5382f51/gateway/src/apicast/backend_client.lua#L214

Please consider whether the error log is appropriate.