Brad Fitzpatrick (bradfitz) wrote in lj_maintenance,
Brad Fitzpatrick
bradfitz
lj_maintenance

Yo yo

Site'll be down for 2.5 minutes sometime tonight after traffic goes down. We're altering a table which'll make the site faster in the future, after clustering conversion is done.

It's been awhile since I posted here. Over 9 days since our last real downtime. Well, a little problem Saturday night, but hardly. And we're breaking new bandwidth usage numbers in the past few days also. (over 20 Mbps at one point, usually just under)

I'm lovin' this clustering setup, btw. Everything's so much more sane to work with now.

We're about 25% done with the conversion. We're running it pretty slowly just because there's no rush. It's been primarily converting large journals so far (since it prefers active ones to convert first), so the conversion process will speed up as it starts doing smaller and smaller ones. Also, the more people that move off the master cluster, the faster we can run the converter without adversely affecting the site.

After clustering is done, the master cluster data is so miniscule (under 2GB or so) that we'll just be running it off ramdisks on the slaves that are initialized on boot from the backup machine. That should be an awesome speedup.

Although I'm having fun doing all this architecture stuff lately, I'm running out of things to improve, so I'm almost back to working on user-level fun stuff. More on that later.
Subscribe

  • Post a new comment

    Error

    Anonymous comments are disabled in this journal

    default userpic

    Your reply will be screened

  • 67 comments
Previous
← Ctrl ← Alt
Next
Ctrl → Alt →
Previous
← Ctrl ← Alt
Next
Ctrl → Alt →