Hi.
The problem is to write logs in a shared file from several processes.
The cooperate call of fprintf() leads to messing the content even in one call
example:
Code:
fprintf(f,"a1 \n a2");
fprintf(f,"a3 \n a4");
out:
Code:
a1
a3
a2
a4
Ofcourse this is possible to implement the file as a critical section with semaphores. But there is a significant chanse to crash for one of the processes, so there is not any garanty that the crashed prosess would not be in the critical section at the moment of crash. So it can cause locking of all processes around the file.
Is there another solution for this problem?
Last edited by Don Cragun; 09-28-2013 at 01:53 AM..
Reason: Use CODE tags for sample input, output, and code segments.
The two fprintf() statements you listed will never produce the output you listed for the file named out no matter what order the fprintf() statements were executed in and no matter whether both fprintf()s were in the same process or different processes. Depending on the mode argument(s) supplied to your call(s) to fopen() to create the STDIO stream(s) f and any calls to setbuf() or setvbuf() on the stream(s) f to set the output buffering mode, there are several possible outcomes, but none of them will result in "a2" or "a4" appearing at the start of a line and none of them will have a <newline> immediately following "a2" or "a4".
However, if all of your calls to fopen() use the mode "a+" and you use setbuf() or setvbuf() to unbuffered (or line buffered if you have each fprintf() format string end output fields result in the last character written being a <newline> character) you could guarantee that the output from each fprintf() would appear in the output as a contiguos string and that the output from one call to fprintf() would not overwrite the output written by another call to fprintf() whether or not those calls occurred in the same thread, same process, or multiple processes as long as no single call to fprintf() produces more than BUFSIZ bytes of output. As long as you use append mode in all cases where you open the stream, you don't need critical sections surrounding the fprintf() calls.
Thanks for detailed answer.
Indeed I've got the similar situation in my log after two vfprintf calls. I'll check my situation and try to formulate my question more accurate.
A larger buffer won't help. Setting the buffer size to zero won't help. The problem is that each vfprintf() call can generate multiple calls to write() - the calls that actually write the data to the file. Each individual write() call is guaranteed to be atomic, but a series of multiple write() calls from different processes and/or threads will wind up as you're seeing - all interleaved together.
You need to make certain each call to your writeLog() function results in one and only one underlying write() call.
write(). Not fwrite() or fprintf().
Code:
#define BUF_LEN 4096
void writeLog( const char *format, ... )
{
char buffer[ BUF_LEN ];
char *ptr;
int length;
va_list args;
va_start( args, format );
// expand the format and args into a single string
length = vsnprintf( buffer, BUF_LEN, format, args );
// if the buffer isn't big enough, allocate a temporary
// one on the stack (if strings can be really long and
// blow up the stack, use malloc() here - but then you
// have to add complexity to the code to make sure
// any malloc()'d buffer is freed.)
if ( length >= BUF_LEN )
{
// I don't recall offhand if the "+ 1" is needed, but
// it doesn't hurt to have it
ptr = ( char * ) alloca( length + 1 );
length = vsnprintf( ptr, length + 1, format, args );
}
else
{
ptr = buffer;
}
write( logFd, ptr, length );
va_end( args );
}
In theory, you could pass a zero length to the first call to vsnprintf() and get back the known length of the final string. In practice there are still vsnprintf() implementations that don't do that per current standards. (The old SUSv2 standard specified different return values.)
Thanks. I did not know about the fact what 'write' is an atomic function. I'll use it. I am using pretty old compiller version on the target system. And with high possibility there are no implementation of vsnprintf.
---------- Post updated at 10:49 PM ---------- Previous update was at 10:39 PM ----------
I've checked this and I found the implementation. The system was updated recently. So the problem with determination of buffer length was solved.
Hi,
While running tcpdump command on my Fedora 16 machine I am get shared library issue.
# tcpdump
tcpdump: error while loading shared libraries: libcrypto.so.6: cannot open shared object file: No such file or directory
# which tcpdump
/usr/software/sbin/tcpdump
I have tried... (3 Replies)
When unlocking a Linux server's console there's no event indicating successful logging
Is there a way I can fix this ?
I have the following in my rsyslog.conf
auth.info /var/log/secure
authpriv.info /var/log/secure (1 Reply)
Hi,
I've the following code to log the errors any after the command is executed.
# Ksh 88 Version
log_path=/home/etc/fls/fls_log.log
del_path=/home/etc/fls/to_day
rm $del_path/* >> $log_path 2>&1
But I even want to log if the rm command is success without any error along with... (1 Reply)
Hello.
I am new to this forum and I would like to ask for advice about low level POSIX programming.
I have to implement a POSIX compliant C shared library.
A file will have some variables and the shared library will have some functions which need those variables.
There is one special... (5 Replies)
We are trying to install third party software on this unix server...
Here is the error message we are getting...
error while loading shared libraries: libodbc.so.1: cannot open shared object file: No such file or directory
It seems like odbc driver is not installed...
>rpm -q unixODBC... (1 Reply)
I am trying to figure a way to have a log file and still keep the output in the terminal in a script.
The example below logs to a file nicely but i still want the output in the terminal as well
#!/bin/bash
#Create a log
exec >> /path/to/my/logfile
echo "hello world"
Any help would be... (3 Replies)
I am writing a shared library in Linux (but compatible with other UNIXes) and I want to allow multiple instances to share a piece of memory -- 1 byte is enough. What's the "best" way to do this? I want to optimize for speed and portability.
Obviously, I'll have to worry about mutual exclusion. (0 Replies)
Hi all,
Well I've had a bit more experience with Unix-like environments since my last post, now that I have started working on my website in earnest and am doing much of the file manipulation via the command line through SSH.
The thing is, I want to be able to log all console activity,... (4 Replies)
I need to create a shared library to access an in memory DB. The DB is not huge, but big enough to make it cumbersome to carry around in every single process using the shared library. Luckily, it is pretty static information, so I don't need to worry much about synchronizing the data between... (12 Replies)
09-28-2013
