asked on

DBM File Locking (FLOCK) Problem...

The following script is part of a login procedure. When a user logs in, the script does the following...

If it is a new user, a counter field in a dbm is incremented and an entry is added to the same dbm containing the path to the user's record file. The name of the record file is "rec#" (where # is the value of the counter).

The problem is that we are encountering circumstances where if two (or more) users log in at the same time (less than a second difference between each one pressing the "submit" button), the counter doesn't get incremented for each of the users. When this happens, an entry is created for only one of the users, but the script returns the same record number (rec and counter) for all three. We first implemented the flock control using O'Reilly's example with no success (we later found out that it causes a racing situation). Our current script uses a "dummy" file for locking. This improved the situation to its current status (<1 second causes error). Please help! The script follows:

#!/usr/bin/perl
use CGI qw(:standard);
use POSIX(O_CREAT,O_RDWR,O_RDONLY);
BEGIN { @AnyDBM_File::ISA = qw(DB_File GDBM_File) }
use AnyDBM_File;

$myLock_SH = 1;
$myLock_EX = 2;
$myLock_NB = 4;
$myLock_UN = 8;

#path to name/record list file
my $caseRef = param("caseRef");
my $tmp = '';
until ($tmp eq '/') {
 $tmp = chop($caseRef);
}
substr($caseRef,0,1) = "";
$caseRef =~ s/DxR\///g;

my ($dir,$regName,$longCaseRef,$usrPW,$lockFileFlag,%usrArray,$usrArray,%recArray,$recArray,$nextFile);

 $longCaseRef = $caseRef;
 $regName = $caseRef.'/caseRec/dxr_stor';
 $lockFileFlag = $caseRef.'/caseRec/intList.txt';

my $usrName = param("dxrUser");
my $enteredPW = param("dxrPassword");
chomp $enteredPW;

my $tError = 'OK';
my $hypCount = 0;

####### open pat database file

###local *myLock;
 # Wait until database is available
 unless ( open(myLock, ">$lockFileFlag") ) {
 $tError = "Open lock file error $! ";
 }
 unless ( flock(myLock, 2) ) {
 $tError = "lock error $!";
 }


unless (tie(%usrArray,AnyDBM_File,$regName,O_RDWR,0666)) {
$tError = "Name/Record List file not found. $!";
}

#is file initialized
if ($tError eq 'OK') {


######check password
 if ($usrPW = $usrArray{$usrName . 'pw'}) {
 chomp $usrPW;
 if ("$usrPW" ne "$enteredPW") {
 $tError = "Password not correct." ;
 }
 } else {
 $tError = "Name not found." ;
 }
}
if ($tError eq 'OK') {
 if ($usrFile = $usrArray{$usrName}) {

 untie(%usrArray);
 close(myLock);

 #record of file exists open and get variables?
 if(tie(%recArray,AnyDBM_File,$usrFile,O_RDWR,0666)) {
 $hypCount = $recArray{'hypCount'};
 untie(%recArray);
 } else {
 $tError = "Record file not opened. $!";
 }
 } else {
 #start file

 $nextFile = $usrArray{'nextFile'};
 $usrFile = $longCaseRef."/caseRec/rec" . $nextFile;
 $usrArray{$usrName} = $usrFile;
 $usrArray{'nextFile'} += 1;

 untie(%usrArray);
 close(myLock);


 if(tie(%recArray,AnyDBM_File,$usrFile,O_RDWR|O_CREAT,0666)) {
 $recArray{'hypCount'} = $hypCount;
 untie(%recArray);
 } else {
 $tError = "Record file not created. $!";
 }
 }
}
#output hyp html page

print <<END_Hyp;
<html>
<head>
<meta http-equiv="content-type" content="text/html;charset=iso-8859-1">
<title>Log Result</title>

</head>
<body bgcolor="#E6D6BD" >
User: $usrName 
Status: $tError 
User Storage: $regName 
Case Path: $caseRef 
User Record: $usrFile $$ $txError

</body>
</html>

END_Hyp

exit;

ozo

The lock is released when you close the lock file, and you're updating %usrArray and %recArray after the close.

rhej

ASKER

I agree that the lock is closed before the modifications to recArray are complete (we don't lock this array because it is user specific and should never encounter a multiuser scenario), but the usrArray was untied before the lock was closed and was not updated afterward. Please let me know if I am incorrect in this assumption.

Please note that both of the untie and close statements for usrArray are within an if/else statement (I apologize for the lack of indentation. Didn't paste properly...).

ASKER CERTIFIED SOLUTION

ozo

membership

This solution is only available to members.

To access this solution, you must be a member of Experts Exchange.

Start Free Trial

rhej

ASKER

The script posted is not the complete script. For space reasons I only posted the parts that used usrArray and flock. $hypCount refers to some of the missing portions and is actually implemented in both branches. caseRef is also error trapped in the complete script.
The only times we have ever seen error messages from $tError have been permissions problems (script didn't have write access).
AnyDBM_File appears to be using DB_File on our testing box. I do not know whether they need to flush. Finding definitive information on them has been somewhat arduous.
I believe a sleep (60) has been implemented to test this, but I need to double check with the author (it's not my script). I know that I recommended such a test, I'm just not sure he did it. I will check and let you know.

rhej

ASKER

Thanks for the assistance! It appears to have been a flushing (sync) problem. Implementing a sync command for the tied dbm before unlocking the lock file appears to have fixed the problem.

Thanks!