NGINX and log rotation: is missing log entries possible?

Question

In it's simplest realization, one can implement NGINX log rotation as following:

mv access.log access.log.0
kill -USR1 `cat master.nginx.pid`

My question. Do we have a concurrency issue here? I mean, if NGINX writes a log entry between two actions (1) file was renamed and (2) it got a command to reload, could some of log entries be just lost?

if it helps to you, please mark the answer as complete – Anatoly Jul 13 '15 at 17:28 — Anatoly, Jul 13 '15 at 17:28
@mikhailov Sure. Tnx for helping out! – user1065145 Jul 14 '15 at 06:34 — user1065145, Jul 14 '15 at 06:34

score 3 · Accepted Answer · answered Jul 13 '15 at 10:40

3

Before send USR1 signal wait for 15 seconds and you will see that *access.log.0" is still being used. It happens because log is open for write access based on inode, not the file name. So it is concurrent-safe to use this technique: mv + USR1

answered Jul 13 '15 at 10:40

Anatoly

556
3
16

NGINX and log rotation: is missing log entries possible?

1 Answers1