the buffer for access log is too large

[spacewander]

Currently we configure the buffer of the access log like this:

    access_log {* http.access_log *} main buffer=32768 flush=3;

According to Nginx's doc:

The buffer size must not exceed the size of an atomic write to a disk file

http://nginx.org/en/docs/http/ngx_http_log_module.html

And for Linux, the default atomic write size is PIPE_BUF (4096), see man 7 pipe.

Therefore, the default size is too large to be safe.

[Miss-you]

+1
A log buffer that is too small is not good, 4096 is more appropriate.

[Miss-you]

On the other hand, is flush=1 better?
Since the log buffer is lowered, the flush time should also be lowered.

[Miss-you]

#1570