What Time is It?

So for a while there I was getting really confused. I kept getting the output from daily cronjobs on Chevette mailed to me in the afternoon but with a timestamp of the middle of the night. Lo and behold, Chevette's clock was just plain wrong.

So naturally, I needed to get me some of that NTP that the kids keep talking about. It turns out the folks over at http://www.pool.ntp.org/ are trying to organize a large number of NTP servers from volunteers. It's a really cool project and worth taking a look at since they discuss specific usage issues and how to best configure your client to update time from the 'net.

So initially I setup NTP on chevette and then I realized that neither siona nor friday were synching time so I did those too. After a little bit of reading on the above website as well as the ISC's site, I ended up deciding to try to run my own NTP server on the LAN. Okay, not really exciting, but overall, the setup is really easy and definately cool. Siona updates against the ca.pool.ntp.org servers and then chevette and friday both update against her. Kinda neet to have that working finally :D

Okay, back to work and my ever-growing TODO list.

Chevette is Loaded and She's Online Live!

Hot hot hot! Chevette is online. Took an afternoon of downloading packages and updates and about an hour of PAM/LDAP configuration and she's live and running the Icecast stream. The stream was running off Friday before but given that's the only usable computer in the house, that was rapidly getting dysfunctional.

On the fun side, the single ices feed including transcoding takes ~30-33% of her CPU. w007! She's already loaded!

In other news, I changed the google search shortcut in Konqueror (which can be done in Firefox too) from the boring "gg" to "grep". That's right, I can now grep the 'net. Who's your daddy? Or should I say:

grep: your daddy

Google returned 1 hit: archangel

Adding a test server and some rantin'

After rebuilding Siona a couple weeks ago, I went out and (finally) bought a battery backup. I got the smallest APC UPS I could get from Futureshop which cost 40CDN. I brought it home, hooked it up, and installed "apcupsd" on Siona. It's pretty bitchin. With apcupsd, I set the tolerences, e.g. if the battery level drops below 10% or 3 minutes remaining, and the battery will notify apcupsd that power is failing so apcupsd can shutdown the server.

The really fun part is that I already got to "test" the power failure events. I had set the threshold to 10 minutes but the estimated battery life, which is just w/ Siona mind you, is only 8.4 total. So the thing with the UPS is that if there are any "significant" fluctuations in the power from the line, then it cuts to battery mode. These dirty power fluctuations happen, oh, every other day or so. So even though it only cuts to battery for a second and then restores normal power, my threshold was high enough that it just issued a shutdown at the first sign of trouble. So I've now confirmed proper operation of the UPS during a "power event" and I've "tuned" the parameters (back to the mfr default).

In other cool application news, I'm finding that AmaroK really rocks! It builds a database of your music from the song meta-data and has a really fabulous interface both for playing music (queues and playlists) and for modifying the metadata (e.g. you can update many songs at once). AmaroK is pretty damned good.

So for upcoming events, I'm thinking of finally putting Chevette to use again but this time as just a test server. Her display is hopelessly foobar and upgrading RAM would also require more $$ then I want to invest in her (sadly). So I figure she'll make a good test server so I can setup stuff like a secondary mail server and test fail-over for mail delivery. Or inter-domain operation with the XMPP server. Or just try different applications. I plan to load her with Debian Sarge on the weekend and go from there.

Anyhow, back to work for me.

Destroy system files? Don't mind if I do!

So last week I was in Ottawa. The day I left for Ottawa, there was a power failure at home. I was 3 megametres from being able to troubleshoot and wouldn't be back for the rest of the week. Just typical.

So after getting home and restoring power, we were still without service. I ran fsck on the drives and mostly they were okay... Except the root partition. After a more agressive fsck, the partition mounted but lots of system stuff was missing. It was just borked.

So I pulled everything off the server (all day yesterday) and rebuilt the system (all day today). There's a couple tricks since my IP address changed during the downtime so it will be a bit before things are all back to normal but we're doing good otherwise.

The one good thing is that I got to switch the drives to an LVM scheme. I went with partitioning the primary drive into boot, root, swap, and var partitions (regular-style). The two 80GB drives got added to a volume group and I just divied up the space between user space, share (media), project (including uro forums), temp, and unallocated space. It's pretty spiffy! I can grow any of the LVM partitions live!

Okay, enough chit-chat. Clearly apache, php, and mysql are working ;)

Time to Ubuntu

So after many weeks and months of my attempts to destroy the Debian/Sarge installation on Friday, I finally gave up. I had muddled and muddled until basically every other application was failing. Yay for me.

So on the weekend I dumped that install and started loading Ubuntu. It went well be very slowly. I'm still not sure why. It kind of had the feel of a bottleneck around drive reading/writing but hdparm verified that the drives are in DMA mode and running pretty derned fast.

Nonetheless, the major applications (except UT2004) have been reinstalled including the music server. Channel Skyhook is back online! W007!

Enough chitter-chatter... Time to go back to getting things done. Know whattamsayin?

Some news old and new

Google Talk recently joined the open XMPP (Jabber) federation so their users can now message users on any of the other open networks including the one I'm running (dl.nibble.bz). Bitchin! I dropped my Google Talk login and added all my contacts to my own login. So far, there's been no problems. Anyone with compatible clients will be able to any of the (pseudo) peer-to-peer stuff like voice and video. That includes multi-platform clients like... Well it's coming. Gaim has been working hard to introduce voice and video (vv) based on Google's technology and Gaim expects to be able to interoperate across platforms and domains very soon. We're all very excited at the possibility :D

In other "old" news, my luck has been holding with KDE and KDE PIM (Personal Information Management). I have been saving my calendar and contacts as files on a USB key and it works great! I just plug in the key, point KMail at the calendar and contact files on there and *bam* it works!

The other thing with mail is managing filters. Well it turns out KMail doesn't let me filter messages to IMAP folders (at least with the versions I have installed) so I ended up looking at Procmail which runs on the server side. A bit of poking around and it actually works really well! A little bit obtuse to be doing just text-based configs with yet another file format, but it's fairly simple and does the job and naturally no filters need "synching" on the client-side.

The one last issue is that I want to maintain a unified list of syndicated feeds. I can manually export/import my feed list with aKgregator (which is a good RSS/Atom client btw) but it's hardly unified. Oh well.

Syslog: Good for info, good for DOS attacks

So after getting some odd complaints from the mail server this morning about "insufficient space", I had to take a closer look at Siona's var partition which was appearing to fluctuate by ~600MB per day. A quick du -sh * pointed to /var/log being the culprit.

So on first inspection, the mysql-bin logs were not being compressed. A gzip on the old logs there gave a couple hundred megs back. Not a world of difference, but a lot.

Next up, it turns out that syslog.0 and debug.0 were over 300MB each and today's syslog/debug were already over 10MB (having been just rotated). So I take a quick look and I see the usual chatter from various services... But it turns out that "the usual chatter" from slapd was 99.96% of the log. Other then enabling it for debugging, I've never found the stuff getting logged to be especially useful to have around so I've disabled that. It was recording a half dozen lines every time PAM or NSS wanted user info (e.g. perpetually).

In summary, the Squid cache (256MB) didn't make the top 3 disk usage problems under var. It was Syslog that was doing a DOS attack.