Take the 2-minute tour ×
Database Administrators Stack Exchange is a question and answer site for database professionals who wish to improve their database skills and learn from others in the community. It's 100% free, no registration required.

This is a very complex scenario, but I figured a state-of-the-art challenge might interest some of the many high-end users of dba.se.

Problem

I'm working on an intercontinental data replication solution for a document production system, somewhat similar to a wiki, utilizing Oracle GoldenGate. The primary goals are to increase application performance and availability around the globe.

The solution must allow for simultaneous read/write access to the same pool of data from multiple locations, which means that we need some clever way of preventing or resolving conflicting updates without user interaction.

Focusing on collision prevention, we must allow an object (a document, an illustration, a set of metadata etc) to be locked globally, thus preventing multiple users from simultaneously editing the same object from different locations - ultimately causing a conflict.

Similarly an object must remain locked until any user's connected database have received the updated data for that object, less a user may start editing an old object without the latest updates.

Background

The application is somewhat latency sensitive, making access to a central data center slow from remote locations. Like many content focused systems, the read/write ratio is in the line of 4 to 1, making it a good candidate for a distributed architecture. If well-managed, the latter wil also work towards ensuring availability during site or network outages.

I have used a somewhat unconventional multi-loop bi-directional replication topology. This keeps the complexity at a manageable level {2(n-1) ways}, adds resilience for site outages and allows for fairly simple addition or removal of sites. The slight drawback is that it may take up to 30 seconds for a transaction to be replicated between the most remote sites via the central master database.

A more conventional design with direct replication between all sites would cut that time in half, but would also significantly increase the complexity of the configuration {n(n-1) ways}.

With five locations that would mean a 20-way replication as opposed to the 8-way replication in my design.

This illustration shows my current test environment across data centers in Europe, Asia and North America. The production environment is expected to have additional locations.

Illustration of replication topology

All the databases are Oracle 11.2.0.3 with Oracle GoldenGate 11.2.1.

My thoughts so far

I've been thinking along the lines of doing locking by inserting a row into a "locking" table over a database link to the central database, while letting the unlock (update or delete of the previously mentioned row) be replicated along with the updated data.

On behalf of the user we must then check the availability of a lock in both the central and local database before acquiring the lock and opening the object for editing. When editing is completed, we must release the lock in the local database which will then replicate the changes and the release of the lock to all other locations via the central database.

However, queries over a high latency database link can sometimes be very slow (tests show anywhere from 1.5 seconds to 7 seconds for a single insert), and I'm not sure if we can guarantee that the update or delete statement that removes a lock is the last statement to be replicated.

Calling a remote PL/SQL procedure to do the checking and locking will at least limit the operation to a single remote query, but seven seconds is still a very long time. Something like two seconds would be more acceptable. I'm hoping the database links can be optimized somehow.

There may also be an additional issues like trying to delete or update a row in the local locking table before that row have been successfully replicated from the central database.

On the bright side, with this kind of solution, it should be relatively simple to let the application enter a read-only state if communications to the central database is distrupted, or to redirect clients if a data center should become unavailable.

Are there anyone who have done anything similar? What might be the best way to approach this?

Like I said initially, this is a rather complex solution, feel free to ask about anything left unclear or left out.

share|improve this question

1 Answer 1

Keep in mind most of the work I do is with PostgreSQL so this may or may not be 100% on the money but I think this should be close enough to be helpful.

The basic issue is that in an environment like this you are going to have a great deal of trouble managing the locks. It seems you can either do some sort of conflict resolution or conflict prevention on the lock level. Conflict prevention seems like despite the performance difficulties, it would reduce the level of user frustration significantly.

My approach here would be indeed to do the locking on the central server in a pl/sql stored procedure, which would insert into the locking table if possible, returning a value indicating success, or if that is not possible, either returning a value indicating failure, or raising an exception (what I have done in the past, for example, is to return something identifying who has the lock if it is already locked).

What I would omit is actually checking with the local server. If you have a high read to write ratio and the chance of collisions is otherwise relatively small, you are going to have to check the central server most times anyway, so there isn't much to be gained by checking the local server also. Certainly you wouldn't want to write to both the local and remote servers for locking. Keep locking simple. It is likely to be a source of pain no matter what you do.

The second thing I would suggest here is that I highly recommend expiring locks like this, perhaps after 2 hours or something. There are two major reasons for doing so. The first is that bugs in the application-layer of the code can cause locks not to be released, and secondly if this is over a web interface, HTTP is stateless and therefore you have no real way of knowing that state has dropped. In this way you can give a lock which is valid for a certain period of time, can be renewed pre-emptively (in the background if needed), and times out if the individual closes the browser window and goes home for the day. Some sort of administrative utility to free locks is also something I would recommend.

I share your sense that 7 sec to acquire a lock is a significant performance cost, but in the end I don't see any better way to do this. Your options are limited significantly by the CAP theorem and a single central locking system is probably what's needed.

I suppose another option is that one could have the central server merely lock to branch location, and have the branch location release the lock as soon as no valid locks have been held for a certain period of time. This might have the advantage of allowing faster collaboration by a team, meaning that only the first editor on a team would have to incur that cost.

share|improve this answer
    
Thanks for your insights! Brewer's theorem does not fully apply here, as we're implementing eventual consistency without distributed availability (If central site is down, remote sites will effectively be read-only). How do you suggest handling the delay between a lock being released in the central database and the data actually being updated on the remote sites? –  Roy Mar 7 '13 at 11:12
    
But it does affect your lock management which is the point of the question, right? As far as dealing with the delay, what I usually do is "set user expectatins" and possibly display a message saying "Making sure nobody else is editing this page" or something like that. that way the users know what's going on and they are comfortable with it. –  Chris Travers Mar 7 '13 at 11:54
    
I also want to say that lock tables IME always add performance headaches. They are often a necessary evil but they never function as well as good old fashion locks of other types (advisory locks etc) because there is additional disk I/O in management of the locks, their timeouts, etc. Such is the nature of the beast. The other option as I mentioned is to do two-stage locking, once to the branch (time out after a few days of no activity) and the other is within the branch, meaning that only the lock by branch operation has the delay. –  Chris Travers Mar 7 '13 at 11:57

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.