tail on deleted/re-recreated file

[Zigor Buruaga]

Hi,

First, my OS's are HP-UX 10.20 & 11.0
When I use "tail" command to monitor a log file, if that file is deleted and recreated with the same name ( while tail is running ), then "tail" still runs but never show new lines added to the file. As far as I know, that is the normal behaviour.
Anybody knows if exists some sort of utility/command that can "monitor" a file even if it's deleted and recreated with the same name?

Thanks in advance.
Kind regards,
Zigor

[Mark Grant]

As names don't mean too much to unix, I don't think you are going to succeed here. When you do a "tail -f" on a file and the file is deleted, the file doesn't really go away until you stop doing your "tail". Therefore, if you create a new file, even with the same name, it will be a different file.The only way this is going to work is if you script tail to set an alarm every now and then to stop "tailing" and start again.

[David_246]

Hi Zigor,

As explained a file is not a file, but an inode. If you want to empty the file to be able to start all over, it's better to null it then to remove and re-create it. Else there is no "tail" solution possible. Of course Perl or other scripts can be written to work around it.

>/path/to/my/file

is an easy way to null a file, the inode remains then so that also tail -f will start working.

Regs David

B.T.W. "cp" will create a new inode where a "mv" on the same FS will keep the inode.

[john korterman]

Hi Zigor,

I have just tried a workaround, which unfortunately did not work! First, I had a continous tail running on a log file, e.g.:
# tail -fn1 your_log_file
I then appended to that file, e.g.:
# echo "helli" >> your_log_file
# echo "hello" >> your_log_file
to which the continous tail reacted.
I then tried to set the file pointer of your_log_file into zero position:
# cat /dev/null > your_log_file
which truncates the file to zero bytes. Unfortunately the tail command is not aware of this; it still looks at the position it was on before the /dev/null action. Therefore, it is not until you write past the position before the "dev/null" action that the tail command reacts again.
I cannot see how to overcome this, except by creating the log file as a fifo, e.g.:
# mkfifo your_log_file
# tail -1fn your_log_file
# echo "halli hallo" > your_log_file

which of course is not what you want, as it raises a number of other problems, but I am looking forward to the coming suggestions!


regards,
John K.

[Zigor Buruaga]

Hi,

Thanks for the replies.
Forgot to mention that I have not control over the log files, so I can not create, copy or move them.
I think I need some kind of new "tail" binary, that can do some kind of monitoring over the original inode and reopen the original file name in case it has been re-created ... or something like that ;-)

Thanks again for all your replies.
Kind regards,
Zigor

[Mark Grant]

Zigor,You will not be able to do that either.When you the log file gets re-created, it will have a new inode even if it has the same name. There is no way around this.As far as I can see, your only possibility is to "tail -f" and then interupt it and start again. You might miss some lines though. You can interupt it if you use a script which starts a sleep process in the background which also send the main script a signal say SIGUSR1 when the sleep has finished. The main script traps this signal and starts the tail again.

[Zigor Buruaga]

Hi,

Now it's clear, I'm looking for something impossible, ... or not developed yet ( it could be a good and very useful improvement ).
Anyway, thanks to all.

Kind regards,
Zigor

[Martin Robinson_4 · Accepted solution]

Although this entry is old, this reply mnay still be of some use. I had the same problem and used Google.

There is a module for Python call TailFollow which addresses this exact problem. Unfortunately HPUX ships with Python 1.something; TailFollow expects 2.0 or later. I was able to downgrade the core component of TailFollow to 1.something capabilities.

[Kent Ostby]

Zigor --

Another option would be to periodically (once a minute or whatever), make a copy of the log file and diff it to your previous copy and then have your script react that diff instead.

Best regards,

Kent M. Ostby