Hacker News Reader

Comment by 3am | Hacker News Reader

I can't watch the video right now, but the idea that "Everybody's the DBA" is risky in general. I'm hugely in favor of developers writing and optimizing their own SQL, being able to create normalized schema (and know when it's worth the tradeoff to denormalize), how to read a query plan, and generally be as competent as a DBA. But... it's good to have one person who has the global view of the database for things like tuning extent sizes, selecting the optimal types of storage for various partions, doing reviews on the schema, capacity planning, etc, etc.

The right setup (IMO) of having a DBA in an operational role with developers that are highly proficient/self-sufficient is hard to get right and expensive enough that it probably isn't right for an early stage company. And a bad DBA can be a nightmare. So there are tradeoffs on both sides.

View on HN

It's about moral authority. The only people touching production should be those whose pagers go off at 3am if it all goes hatstand. Everyone else is the peanut gallery.

Are you sure? Amazon is getting flak for that lately (in e.g. Yegge's rant), from having all of its developers on-call. You should be around to fix something you push if it breaks things, but that's not the same as being on-call at all hours for any given coworker's or manager's standard of "going hatstand".

>You should be around to fix something you push if it breaks things

Which has another advantage over Amazon's pager policy: if you are up at 3am, only semiconscious, you are likely to be more useful fixing code that you broke (because you just touched in the last day or two) than some other coworker's mistake.

Agree, with the caveat that those who carry the 3am possibility are allowed to delegate at their own risk.

i.e. the real rule is "nothing happens to prod without the agreement of those at the sharp end".

That allows good devops fun to happen with fewer ops bottlenecks, if some devs can provide enough ops-chops to keep the real ops people happy.

[Also note in passing that such devs are likely to be in the "second wave" of 3am people if things go sufficiently pearish.]

Agreed - that is a general rule too. E.g. those that do the actual work make the final call on which technology, language, database, whatever to use - not some "architect".

I love this explanation. Note my username :)

I think that with a modern database like SQL Server, the old-school "high priest" DBA is obsolete.

But... if you don't have someone dedicated to thinking about database issues, you need to treat database changes just like your code. It needs to be in a repository, it needs to be reviewed, and you need a change management regime.

From a anecdotal POV, I've noticed that many folks have a good process (or at least a consensus approach) to managing their code... but the database is often a red-headed stepchild that doesn't get the attention it deserves.

but the database is often a red-headed stepchild that doesn't get the attention it deserves

On my last three major projects, I have committed to devoting the proper level of attention to the database, with automatically building databases in some environments, scripted scheme changes and seed data loading as part of the mainline code base, etc. and I have found the difference to be immense in practical terms. I can move more quickly, more safely, and have a better quality of life as a developer.

One of the best returns on (effort) invested I have ever seen.

I definitely agree. We have been using fluent migrator (https://github.com/schambers/fluentmigrator) for our .net based projects at work for database migrations and versioning

Thanks. Fluent Migrator looks pretty cool. I've been rolling my own stuff as-needed. As a result, it has evolved into something pretty useful for the problems I've encountered so far, but does nothing outside of what I've already thought of. I can see using the migrator as a good jumping off point.

Sure, I wasn't suggesting a high priest. Maybe more a 'faculty advisor' dba? Just someone keeping an eye on the higher level problems (having the right Nagios alerts set up, keeping an eye on how long reindexing is taking and scheduling downtime according, disaster recovery planning, putting in the POs for new drives). That's even if you treat writing queries and schema changes like part of the code, it's a dynamic system. But I completely agree on treating DB changes like code changes, I think DevOps is a good approach.

Same as with a modern language, the old-school "high priest" developer is obsolete.

Thankfully, with Liquibase [1] and others like it, developers are starting to think about database change management.

I absolutely despise writing liquibase changesets, but I know how important they are.

[1] http://www.liquibase.org/

You'll want to watch the video, because I explain that in more detail. Of course I'm not against change control or security, but if you want to scale a database, everybody involved has to have DBA-level skills. StackExchange's sysadmins and developers are all keenly aware of the database impact of what they do, and they aggressively try to minimize that impact. That's what I talk about on the video when I say "Everybody's the DBA." It's like saying, "Only you can prevent forest fires."

What do you consider DBA level skills? Because you mention in one case people using SELECT * where they only needed a small subset of the fields. I'm all for your message that everyone should be aware of the database, but I don't consider that a DBA level mistake. I've just worked with some astonishingly good DBAs before, and I while I'm not a slouch it would be impossible to catch up with them without years of intensive study.

Anyway, I'm splitting hairs. Your presentation was great, and the people at SO are very impressive overall. Thanks for posting it.

I consider DBA-level skills as the ability to read an execution plan, figure out what indexes are being used, know the difference between a good plan and a bad one, and figure out how to rewrite your query to get the performance & plan you want without just adding indexes willy-nilly. Another good example would be knowing when it's right to normalize tables versus flattening out the data structures - you can't just knee-jerk and normalize everything because it's the best practice.

I wouldn't say that developers need to know about HA/DR, replication, memory tuning, SAN setups, etc, though. Just the how-queries-work and how-data-is-stored angles.

Thanks for the compliments! I had fun doing that one.

Reply on news.ycombinator.com