Twitter releases Gizzard, a framework for creating distributed datastores

http://engineering.twitter.com/2010/04/introducing-gizzard-framework-for.html

by aviv • 16 years ago

41 11 16 years ago

engineering.twitter.com

benologist4• 16 years ago

With all the downtime and seemingly permanent "Older tweets are temporary unavailable" messages why would anyone want to build on platforms they've created?

Maybe they have good ideas but I think part of the reason people are looking at Cassandra etc is because the high-profile sites using/developing these platforms actually work.

jamie• 16 years ago

My guess is that a lot of this infrastructure is to support them moving to new technologies while the old ones are in place. Changing the engine while the car is in motion requires a lot of piecemeal engineering work, and I think you're seeing that effort here.

gfodor• 16 years ago

This is such a cynical and short-sighted comment. Did twitter have scalability problems? Yes. Does this mean that any and all technology that comes out of the company is junk? On the contrary. If anything, getting battle tested software that was built in response to solve massive scalability issues is going to be better than not.

Clearly they've worked through a lot of their problems at this point; I'd imagine this is some of the code that let them do so.

Nevermind the fact that a reply like this is disrespectful towards people who are giving you the fruits of their labor for free. Shame on you.

benologist4• 16 years ago

Maybe they nailed everything with Gizzard, but it'd be more convincing if their scalability and other problems were actually solved. I'm not sure why you alluded to those problems in the past tense - http://status.twitter.com/.

If I wanted to be disrespectful I'd have said their problems are still very real and very often and for that you'd have to be retarded to assume they've gotten anything but marketing right so far.

Would you build on gizzard right now? Their status blog is full of very recent reasons why it's premature for them to release anything and even more so why it's premature to trust or be grateful to them for doing so.

jluxenberg• 16 years ago

If a shard goes down, writes to it are queued on a Gizzard middleware host. What happens if that Gizzard host is then lost?

eaceaser• 16 years ago

The queues are local to each host so you'd lose those writes, but since writes are written to a journal on disk, you can apply some redundancy strategy to that as necessary. Depends on the SLA really.

Twitter releases Gizzard, a framework for creating distributed datastores

Related Stories