HP-UX

have you tried nzle's suggestion from #10 with the 'kill -1 PID' suggestion?
Maybe the app writing to the long will catch an interrupt and act accordingly...
Worth a try (not knowing the internals of the app).

What I dont understand is if the file is zeroed ( >file) how come it the next minute be greater than previously...
I just remebered that in 11.11 you have in SAM: "Routine tasks>Sytem log files " the possibilty by clicking on the menu Action to add to list and configure...
What about trying?

Is your log "per line"?

And Yes I would give a kill -1 a try... it doesnt always work but if it does...

You must update (restart, kill -HUP, ...) your application.
Read your application documentation.
Even logrotate would need such an action; a pre-rotate or post-rotate action.

Code:

man open

tells that the O_APPEND flag in open() jumps to the end of the (with the old file size) prior to every write.

That's incorrect. Before each write(), O_APPEND sets the descriptor's offset to the current end of file. If a file opened with O_APPEND is truncated to 0 size, the next write would see that and would place its bytes at the beginning of the file.

What we're seeing here is the absence of O_APPEND. Truncation does not affect the offsets of open file descriptions, so, without O_APPEND, writes continue where they had been located before truncation; file size is restored with null bytes replacing what had been truncated.

Regards,
Alister

Thanks to all.

Allister describes the exact problem. I cannot modify the application code and i try to find a solution with a script rotate (not possible if i don´t modify application) or splitting the output (not possible...i lost data when i stop the application).

Fail attempt!

Yes we've had good results with the copytruncate option as well, in logrotate on hp-ux 11.31.

We too have an application that is continuously writing output to the same log file, and it grows to many GB within a week. We configured logrotate to rotate it daily, keep 30 prior days worth of logs, compress the old logs, and use copytruncate to keep the application happy. It works very well.