Sorry I wrote too soon, there were errors in the log.
Still, I believe nginx -t should test for valid (usable) directories
before starting to avoid running badly.
If HttpProxyModule can’t write to the directories inside proxy_temp
(eg they are owned by root not server uid) then requests for URLs of
large size fail after 102400 bytes, they hang in the client, then
timeout and no errors are printed.
In fact access_log shows code 200 for each request which was quite
puzzling to debug.
maybe nginx -t should alert if one of the _temp directories has
unwritable files or directories in it?