Log module question: does the buffer mess up the order of the log entries?
haifeng.813 at gmail.com
Mon Sep 23 01:59:29 UTC 2013
I am reading the log module source code, there is something difficult to make sure, so I ask for your help.
Access log module use a buffer to buffer log entries before writing to the file system, the buffer is initialised before the worker processes are forked, so I guess after the fork(), each worker has a copy, this also explains why there is no lock-unlock operations while using the buffer. To be sure about that, I did a simple test:
1, configure nginx to use 16k access log buffer, use the default keep-alive time(65), work in master-workers mode with a few worker processes;
2, open one browser, access nginx server, refresh a few times, no access log generated;
3, open another browser, do the same thing as 2, until the access log was flushed;
I think there is a chance that the two browser was served by different worker processes, and log entries may be buffered in different buffers, which buffer get full first, which will be flush first. According that, the order of the log entries could be messed up. Unfortunately, I didn't see that after testing for a few times.
My question is, Am I wrong about the log module behaviour, or I didn't get the right way to test it?
More information about the nginx