Forcing a 'process' to wait on an object specific lock

Rails
1

I have a synchronization problem that I am not sure how to solve in
Ruby/Rails. I have a mental model of what I need to do and how it should
work, but can’t find the way to do it in Ruby/Rails. (I’ve learned all
about Mutexes and transactions and lock! but despite my research, am
still
stuck). I come from the kernel programming world, so I may be framing
this
incorrectly.

Here’s the deal -

  1. I have a database object, call it a Meter. Meters have all sorts of
    associated objects.
  2. My users can do things to meters and their associated objects via
    standard web requests.
  3. I have a cron taks that runs on the server (Heroku) periodically and
    also does things to meters and their associated objects.
  4. If a user is mucking about with a specific meter or it’s associated
    objects at the same time as the cron task is trying to muck with the
    same
    meter or associated objects, then there’s potential for a data
    corruption
    problem.

So - I need to block the cron task from mucking about with a Meter and
all
its associated objects while a user is doing so (and vice versa).

In my old kernel world, I would have defined a lock on each meter.
Wherever
I want to assure that a section of code that mucks with a meter is not
running concurrently with another section of code that mucks with the
same
meter, I would wrap each code section that mucks with meters in a
Lock/Unlock pair as follows:

section1:
m = get_my_meter
OSLock(m.lock)

do section 1 stuff to meter and associated objects

OSUnlock(m.lock)

section2:
m = get_my_meter
OSLock(m.lock)

do section 2 stuff to meter and associated objects

OSUnlock(m.lock)

The resulting behavior is that (assuming that two code sections tried to
operate on the same meter concurrently), one would get there first and
the
second would be blocked until the first called the OSUnlock. At that
time,
the second would run and I would be assured that the code sections would
always run in sequence for any given meter (as opposed to concurrently).

In Ruby/Rails, I see that there are mutexes, but they seem to be
applicable
to multiple threads running on the same in memory objects. In my case,
the
user request is going to load its copy of the meter from the database
and
the cron task is going to load its copy, so - mutexes don’t help.
Transactions and pessimistic locking seemed promising, but from my
experimentation with them they will only prevent concurrent operations
on
the specific object that is locked. Operations on associated objects,
even
if they are inside transaction blocks, can still run concurrently. So -
it
seems that the executing code is not waiting, rather, the database
access,
for the locked object only, is waiting.

I need to block the cron task from executing any further than where the
lock is taken until the user task releases the lock (and vice versa).
How
do I do that?

Thanks,
Yoram

2

On Thursday, January 31, 2013 11:55:40 PM UTC, [email protected] wrote:

case, the user request is going to load its copy of the meter from the
do I do that?

You can still use a database lock on the meter as the gatekeeper to
modifying either the meter or its associated objects (although obviously
you need to do this in the webapp side too). In fact you don’t
necessarily
need a long lived database lock - it is sometimes easier to have a
locked /
locked_by attribute on the model, and only use an actual database lock
(or
optimistic locking) when writing to it.

Actually blocking for a long time on something is not something you
really
want to do in a web app (assuming this background processing is lengthy)

you’re better off failing fast and telling the use to try later

Fred

3

Thanks Fred.

You did drive right to the heart of the issue. The lock itself is indeed
easy enough. It’s the waiting on the lock that i’m not sure how to
implement (other than a loop that tests the dbase lock, which doesn’t
seem
right). I agree that in general, we don’t want to block threads in a web
app. But - in this case, the code section that needs to execute without
interruption is very brief and the likelihood of collision between the
two
processes on the same object is very small.

I may yet adopt your suggestion of just telling the user to try again
later. It’s much more complicated if the cron task gets blocked (it
would
have to maintain a list of objects that it couldn’t process, then go
back
and process them later). But - is there a way to block a process on some
sort of lock/semaphore, other than loop/test/loop?

4

On Feb 1, 2013, at 8:28 AM, [email protected] wrote:

You did drive right to the heart of the issue. The lock itself is indeed easy
enough. It’s the waiting on the lock that i’m not sure how to implement (other
than a loop that tests the dbase lock, which doesn’t seem right).

What you’re looking for is called a “condition variable”, and ruby has a
class for that.

–
Scott R.
[email protected]
http://www.elevated-dev.com/
(303) 722-0567 voice

5

On Friday, February 1, 2013 3:28:57 PM UTC, [email protected] wrote:

I may yet adopt your suggestion of just telling the user to try again
later. It’s much more complicated if the cron task gets blocked (it would
have to maintain a list of objects that it couldn’t process, then go back
and process them later). But - is there a way to block a process on some
sort of lock/semaphore, other than loop/test/loop?

you should be able to block on a pessimistic lock - select … for update
will either acquire the lock or timeout/deadlock. (The .lock! method
does
that for you). Don’t forget that the lock is released at the end of the
current transaction (so if there is no transaction the lock is released
immediately.

Fred