Hi,
While testing MogileFS, I bumped into an issue that may not be unique to
my environment.
If a process that contains a MogileFS::Client object closes stdin, as is
often true if the process is an Apache mod_perl child, there is
potential for the client to lose readability to its trackers and die in
the midst of its request.
This event occurs because my systems assign fd(0) to stdin. If stdin is
shut, fd(0) goes back into the pool of available descriptors. If the
system then assigns fd(0) to a tracker socket, &_wait_for_readability of
Backend.pm always returns false.
sub _wait_for_readability {
my ($fileno, $timeout) = @_;
return 0 unless $fileno && $timeout;
...
}
:: The patch is simple::
sub _wait_for_readability {
my ($fileno, $timeout) = @_;
return 0 unless defined($fileno) && $timeout;
...
}
I hope this information helps somebody.
There are also a couple fileno checks in MogileFS::Util that may need
review.
Backend.pm-patch.txt
Description: Backend.pm-patch.txt
