How about a simple timestamp test?
Have pakfire touch something when it has created its lockfile, then the calling script could check its timestamp before executing system_background() and afterwards check if it has changed.
That should be easy enough to implement.
Tapani
On Fri, Oct 15, 2021 at 11:14:09AM +0100, Michael Tremer (michael.tremer@ipfire.org) wrote:
Hello,
This is a classic race condition, but I am not sure if there is a more elegant way that can be implemented very simply.
The problem presumably is that system_background() is too fast. It forks first and then launches pakfire. The behaviour was different previously. We started a shell which then launched pakfire and we waited until that shell has returned.
If anything, sleeping longer will increase your chances to only continue after pakfire has been launched.
Anything that checks whether pakfire’s lock file shows up is probably not a good solution because it might run indefinitely when pakfire finishes its job very quickly.
With the right checks around it, this might be our only option unless we want to create a more complicated inter-process communication system for this.
-Michael
On 15 Oct 2021, at 10:33, Stefan Schantl stefan.schantl@ipfire.org wrote:
Hello List,
after some more testing I had to decide to reject this patch from beeing merged.
It simply does not work all the time, because may the sleep time is too short if the system is busy, or pakfire needs to long to startup and place the lockfile.
So I have to go back to the drawing board and do a better solution for this issue.
Sorry for the noise on the list,
-Stefan
In case the package list should be grabbed or the system should be upgraded, pakfire got called which writes a lock file to prevent from beeing launched multiple times and to lock the pakfire.cgi with the nice log output.
In case update or upgrade has been performed via WUI, pakfire has been called and written the file in the background but the WUI script has been executed further and because of a race condition it did not recognize the lockfile at this moment because it was not present.
So a simple sleep should to the trick and give pakfire the required time to write out it's lockfile.
Fixes #12696.
Signed-off-by: Stefan Schantl stefan.schantl@ipfire.org
html/cgi-bin/pakfire.cgi | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/html/cgi-bin/pakfire.cgi b/html/cgi-bin/pakfire.cgi index 0cf522ba1..aaf63d469 100644 --- a/html/cgi-bin/pakfire.cgi +++ b/html/cgi-bin/pakfire.cgi @@ -133,8 +133,10 @@ END
} elsif (($cgiparams{'ACTION'} eq 'update') && (! -e $Pakfire::lockfile)) { &General::system_background("/usr/local/bin/pakfire", "update", "--force", "--no-colors");
sleep(1);} elsif (($cgiparams{'ACTION'} eq 'upgrade') && (!-e $Pakfire::lockfile)) { &General::system_background("/usr/local/bin/pakfire", "upgrade", "-y", "--no-colors");
sleep(1);} elsif ($cgiparams{'ACTION'} eq "$Lang::tr{'save'}") { $pakfiresettings{"TREE"} = $cgiparams{"TREE"};