Everything is green again

Just got back from the datacenter. Everything was running again about 60-90 minutes ago.

Since the last reboot the server in question had more disks installed. The disks are split across two 3ware raid controllers (one with the new disks and one with the old). The bios decided to try booting off the controller with the new disks – of course there's no boot record on those so it just stalled.

The serial console was working, but for some reason it'd not show me anything until it had skipped past the "press button to enter bios setup" window. Robert could get into the bios remotely, but he was too lazy to check the settings so downtown I went. Grrh.

file server reboot

We have one particular server that's hosting many of our databases, NFS for our internal software distributions etc etc. We reboot it a couple of times a year to get kernel updates and such. One of those times is in a minute (so a short outage of some services will occur). Crossed fingers it'll come up alright, we don't want to go to the datacenter right now. :-)

update: eek, it's not coming back. crap.

Overload!

You may have noticed some trouble accessing some perl.org services this morning. Our network link is being saturated, because one of the DNS services we host is getting hammered. We've shut down the server on our side, but the other side is still sending. Mad props to InterNAP for being helpful in diagnosing and putting workarounds in place to get us stable again.

nntp upgrade

Tonight we rolled out the upgrade of our nntp server (nntp.perl.org) to colobus 2.1. Previously we were running a very old, somewhat hacked up version. Now we're running a new, only slightly hacked up version. (We had to make some changes for performance and quantity -- all sent back to the author of course.) You shouldn't notice any difference -- the speed seems about the same. The new version is backed by MySQL, which conveniently gives us easy access to the overview database (which contains threading information.) This will let us do cool things in the new web interface (which has been "Coming soon" for way too long.)

Full Duplex

Tonight I went to the datacenter and changed our incoming internet pipe to full duplex mode. This will give us a few more usable megabits of bandwidth, making it much less likely that we'll max out our bandwidth. It also cuts off a few milleseconds of latency.

As always, I need to thank the wonderful support staff at InterNAP for making everything so easy! I called, and within 10 minutes we had the link forced to full duplex with no interruption of service.

Why were we in half duplex mode in the first place? Good question. Their side was set to autonegotiate as was ours. Maybe it's the media converter? Maybe its our switch? Setting our switch to full duplex (and leaving theirs on auto) only made things worse. But now both sides of the link are happy, I'm happy, and our users should be happy too!

Help us at Perl.org!

If you are reading this weblog wondering how you can help, here's your chance. :-)

RT / perlbug down today

So the issue from yesterday didn't get sorted out last night. I've been working on it today and the database is happy again. Just waiting for Robert to startup RT again. :-)

RT / perlbug issues

MySQL on our main database server (we were running an old-ish 4.1.x version) crashed and corrupted some (unrelated to the crash, I think) InnoDB tables.   We did run it with some vaguely aggressive settings, but it shouldn't lead to corruption (only losing a few updates).   Grrh!

Speaking of the database server, AMD are giving us at least one of their servers; we don't really know too much about it except that a package is supposed to arrive at my house.  One of our planned uses is a new database server which should speed up RT and several other services.

From:   Robert Spier
Subject: Perlbug Email Down, Web Interface Unstable
Date: December 6, 2005 9:52:20 PM PST

Dear p5p, p6i,

We're having some database issues.

In order to make my life easier when I try and recover from them, I've disabled incoming email to the RT instance.  (It will be queued up and
delivered later.)  You may also find the web interface to be unstable
as the database decides to disappear mid-stream.

I will be working on this tomorrow, and hope to have everything back

to normal within 24 hours.

(Details: it appears that somehow, we've made innodb unhappy.

Tomorrow, I will be doing a lot of dumps and restores to try and

straighten it out.)

-R

update 1am PST: All seems well again. Unless it goes bonkers again during the night Robert is going to turn processing of the RT emails back on in the morning. - ask

Perl Foundation Weblog

Apologies if you already saw this on use.perl, my personal weblog or one of the soon bazillion other places it's been posted.

Yesterday we announced the Perl Foundation Weblog. Robert and I will post there occasionally too. I'm not sure yet if we'll just take some posts from here or if we'll split it up and leave this weblog more for just operational stuff and have news and status reports there.

CPAN / FUnet update

It's not something we (Robert, Graham & Ask) are in control of, but if you were wondering why search.cpan.org hasn't been getting new module uploads quickly lately it's because of problems with the hardware running the FUnet.fi master CPAN mirror.  It's making all of CPAN "not move" so to speak.

PAUSE is still running and taking uploads and no data is being lost anywhere, just a little delayed.

Last I heard it should get better sometime this week.

New search engine

For a long time Search CPAN has used the WAIT module to provide searching of POD documents. Today we started testing the use of Swish-e. Try it and let us know what you think

power is back...

The power is back and the console server came up. Now for trying to sort out everything else, hopefully without having to drive there.

update: alright, I got everything beat into running again except for the mailing list server. A few of the other boxes (on really really really old hardware) had a hard time getting started too, but after an extra power cycle and a manual fsck I got them going. x6, the list server, isn't as responsive though. Grrrh.

On the radio they're talking about "45 minutes traffic" from Hollywood to downtown on the 101 (when there's no traffic it takes just 10-15 minutes) so I'm hesitating a little bit running down there to look at it, but I will have to go soon as I can't be holed up down there for too long tonight.

update 2, much too much later: The mailing list mails are flowing again... (and the console redirection on that server is working again; one of the harddrives in the RAID not so much. grrh).

Power outage in Los Angeles, perl.org down

Apparently there's a Really Big power outage here in Los Angeles.

At my house the power flicked enough for the UPS to click in and for things not on the UPS to reboot, but there's power just fine now so I didn't think anything of it.

Ask: Circuit not working. Can't breathe!

Internap NOC: Los Angeles? Big power outage there. Traffic signals not working, police on "full tactical alert" and stuff.

Ask: Uh-oh! Okay.

Ask: Generators?

Internap NOC: They said they came on but shut down due to overload.

Ask: Doh.

Our servers are in the same building but on a different floor, so it's entirely likely that they are affected too. Grrh. I'll have some fun getting everything running again when the power comes back down there (Robert is traveling today).