Is there any thing in ruby that acts as a global mutex?
For example, across multiple mongrels each running the same application,
is there anyway for them to communicate between them and do something
like
global_semaphore.synchronize do #block
end
So far Im thinking this would involve a PID file that stores which ruby
process currently has focus, and then you would constantly poll this
file to see when it changes to another ruby process can take the focus.
Is there any thing in ruby that acts as a global mutex?
Between multiple processes on the same machine: I’d suggest using
File#flock on a local file (this probably won’t work over NFS) with
File::LOCK_EX. You can include File::LOCK_NB if you want to test the
lock but not block if someone else has it.
Between multiple processes on multiple machines: probably run a DRb
server one one machine which has a local mutex and yields back to the
caller.
I actually ran into that global bottleneck issue when ruby processes
would get KILLED midprocess leaving this lock file. I added a “cleanup”
action which makes sure the PID file is no being occupied by a process
that doesn’t exist.
Thanks for the pointers guys. I think my solution is 98% robust.
I actually ran into that global bottleneck issue when ruby processes
would get KILLED midprocess leaving this lock file. I added a “cleanup”
action which makes sure the PID file is no being occupied by a process
that doesn’t exist.
Actually that was not what I meant. I referred to the fact that if you
have a single global lock that all concurrent processes must acquire,
then your overall performance will suffer.
Thanks for the pointers guys. I think my solution is 98% robust.
Let’s see… That makes 365 days * 2% = 7.3 days downtime per year.
I used something like this using “ln -s” system command. Used to work
great,
so if you want a simple approach and implementation you can use it. One
catch: if the process which has the lock dies in between, you will have
to
ensure locks are properly released.
Both versions I proposed don’t suffer from this problem.
if you flock a file, the lock is dropped when the process owning it
dies
if you synchronize using ‘yield’ within a DRb server, then if the
client dies the TCP connection will be dropped and the block will
terminate.
I used something like this using “ln -s” system command. Used to work
great,
so if you want a simple approach and implementation you can use it. One
catch: if the process which has the lock dies in between, you will have
to
ensure locks are properly released.
BTW, “ln -s” approached worked with processes running on multiple
machines
as well. But yes, all the machines had access to a small common working
directory.
