The database is not the world's most efficient way to do this. However if it isn't a bottleneck, who cares? It is one less dependency to maintain, understand, etc.
We're looking to create working solutions, not works of art. Seeking the perfect component for every piece is often simply not worthwhile.
I find it ironic that he writes this as a rant, which is an anti-pattern for influencing people. One of his commenters hints at how it might have been better presented: introduce messaging systems, explain how they solve the problem, and describe their advantages over the database. For further discussion on why a rant is probably an anti-pattern, see Dale Carnegie's "How to Win Friends and Influence People."
Using databases as a queue is not an anti-pattern (delayed_job is an example).
A full blown message queue system is often overkill; it adds another system that you have to learn. (And frankly, queues are often backed by databases anyway)
That said, it's best to avoid having application state in the queue, it should be generic. This makes it more scalable and flexible.
Also, if your using the DB as a queue index on the status element. Inserts are still low overhead and the DB can handle poling that table 10,000 time a second without issue.
Edit: Just don't pole the 'finished' status as that can have a lot of elements in it.
I use MySql+cron as a simple message queue every now and again--it's great for jobs like sending confirmation emails or registering new users for a mailing list. ie. Tasks that make synchronous calls over the network and can degrade the speed of important processes (registration, etc). You just have to make sure that these tasks are going to be infrequent and not prone to spikes that might overwhelm the server.
A lot of that post is specific to SQL Server. I used a MySQL table as a queue as part of Signal Spam (https://www.signal-spam.fr/) and it worked flawlessly. Polling was at 1 minute intervals and queue entries that were completed were simply deleted.
That particular use-case is such an anti-pattern in SQL Server ( because SQL Server isn't or hasn't until recently been MVCC ) that Microsoft added Service Broker to the product.
It isn't nearly the anti-pattern, at least for many loads, in an MVCC database engine.
I would be happy to use something else but every time I start to look at ZeroMQ or celery, I lose interest and just go back to the DB which surprisingly, seems a lot easier to use. Is there a delayed_job for Python?
btilly|14 years ago
We're looking to create working solutions, not works of art. Seeking the perfect component for every piece is often simply not worthwhile.
bunderbunder|14 years ago
redcircle|14 years ago
tarr11|14 years ago
A full blown message queue system is often overkill; it adds another system that you have to learn. (And frankly, queues are often backed by databases anyway)
That said, it's best to avoid having application state in the queue, it should be generic. This makes it more scalable and flexible.
Retric|14 years ago
Edit: Just don't pole the 'finished' status as that can have a lot of elements in it.
gaius|14 years ago
dclaysmith|14 years ago
jgrahamc|14 years ago
jpitz|14 years ago
It isn't nearly the anti-pattern, at least for many loads, in an MVCC database engine.
pbreit|14 years ago
tszming|14 years ago
But you need to understand not all jobs are created equal..