Salam, I have two solutions:
1- In your application you check if the log file is available or not, if not then you queue your log messages in a buffer until the log file is available again, then you flush that buffer to the log file 2- If your process can wait a bit, then you SIGSTOP the process and once you finish you SIGCONT it Of course use mv instead of the cp since cp will take a hell of time while mv will finish in milliseconds which will help in either cases (in the first case your buffer will not grow that big and in the second case you will not stop your process that long time) Hope this helps, On Wed, May 13, 2009 at 11:53 AM, Mohammed Abu Rahmeh <[email protected]>wrote: > whats wrong with mv ? > > [code] > mv *old_file new_file* > touch *old_file* > [/code] > > not sure if this will cause data loss, but if you know the source of the > log (some script ?), then check the existence of old_file prior to the log > action > > [code] > if [ -a old_file ] > then > # log into old file > else > # log into temp file > [/code] > > if temp contained any lines (or existed at all), these are the lost data, > and since its the main concern here, this might help you to check whether a > method is appropriate. > if you dont know the source of the log, you may try > [code] > grep -r old_file / | grep -i log > [/code] > > On Tue, May 12, 2009 at 4:30 PM, Husam <[email protected]> wrote: > >> in case of move there wont be an old file >> >> On Tue, May 12, 2009 at 4:24 PM, Aram Yegenian <[email protected]>wrote: >> >>> >>> >>> Hi Yaman, >>> >>> Just out of the top of my head coz I just woke up... Why not use mv, coz >>> mv is atomic and any process that still has access to the file would still >>> be able to write to the old file until you restart that process. >>> >>> --- On Tue, 5/12/09, Yaman Saqqa <[email protected]> wrote: >>> >>> > From: Yaman Saqqa <[email protected]> >>> > Subject: [JoLUG-General] Question on cp >>> > To: [email protected] >>> > Date: Tuesday, May 12, 2009, 3:13 PM >>> > Hello Luggers, >>> > >>> > A quick question. >>> > >>> > Let's say I wanna rotate a log file upon reaching a >>> > certain size and the file name is time dependent. logrotate >>> > configuration wildcards will not help, so I'm writing a >>> > script to do it, my questions is, as per the pseudo below >>> > >>> > >>> > >>> > every x time >>> > check file size >>> > if size > y >>> > cp oldfile oldfile-1 && echo > oldfile # >>> > (copy-truncate) because processes have open fd's with >>> > oldfile >>> > fi >>> > >>> > If the file size is large, cp WILL take some time and >>> > CONCURRENTLY data is still being written to the oldfile. >>> > Will cp cater to every byte written to the oldfile up >>> > until the micro-second "&& echo kicks in"? >>> > Or will there be data loss? >>> > >>> > >>> > >>> > Thanks! >>> > >>> > -- >>> > abulyomon >>> > >>> > www.KiLLTHeUPLiNK.com >>> > >>> > >>> > >>> > >>> > >>> > >>> > > >>> > >>> > >>> > >>> >>> >>> >>> >>> >>> >> >> >> -- >> Husam Habannakeh >> +971 5055 164 89 Dubai >> +962 777 656 086 Amman >> +966 5611 54 798 Riyadh >> Sent from Riyadh, 01, Saudi Arabia >> >> >> > > > > -- Khamis Siksek http://saksoook.blogspot.com --~--~---------~--~----~------------~-------~--~----~ ### Jordan Linux Users Group ### http://Jolug.org/ http://groups.google.com/group/Jolug ### Ubuntu Jordan LoCo Team ### https://wiki.ubuntu.com/JordanTeam http://lists.ubuntu.com/ubuntu-jo ### Ojuba Linux ### http://ojuba.org/ ### Jordan PHP ### http://groups.google.com/group/JoPHP -~----------~----~----~----~------~----~------~--~---
