I read yesterday that Microsoft is planning (releasing?) a web based version of MicroSoft Office. While I neglected to save the link, it really is not important, since this is clearly the latest both in MicroSoft’s ongoing feud with Google, and in their end-goal of subscription based software.

Google has set up an open source project to help convert between blogging software formats.¹ From the press release,² it supports Blogger, LiveJournal, MovableType, and WordPress. Very cool.

On a nearly random side note, metacity’s default keystrokes are woefully substandard to my (admittedly customized) fvwm2 keystrokes. I have yet to randomly hit upon the right keystroke to change virtual desktops.

In other news, I have spent a little bit of time modifying the css for ikiwiki to attempt to match the css I use elsewhere. It is a fustrating process. I ended up changing the theme I use for everything else in an attempt to be more ikiwiki friendly, but while I have ended up with a collection of style.css files that mostly work, I cannot be entirely happy, there is a decided lack of cohesion here.

I suspect I will need to dive into the template files for ikiwiki to make more progress.

As a correction to my previous post, libfuse2 is not in fact at fault here. I am able to mount an sshfs with the suid option. The problem is that gluster does not implement the -o option, and so I cannot pass the suid option to the mount, and thus the default of nosuid takes effect.

I am still quite disappointed.

Discovered tonight that libfuse2 forces fuse file systems, such as gluster, to mount with nodev,nosuid. This means that I cannot host an ikiwiki with CGI support on a gluster file system. This does not please me.

I would really like to see this, and read/write shared mmaps (for spamprobe and similar) to be supported in fuse file systems (and naturally gluster in particular).

Today I learned that qmail was released to the public domain back in November.¹ Vincas tells me that he informed me of this change back in November when it happened, and that I was not interested. I do not know where my head was at the time.

As we are now looking at using qpsmtpd for auth and greylisting, it is entirely feasible to put qmail behind it instead of postfix, and to ditch mailman in favor of ezmlm. I am thinking that this might be the best way to go, particularly now that there is a route forward for qmail. One of the flaws in it has historically been that its development simply does not move, at all.

For this reason I am thinking that I would look at netqmail and not qmail itself, as there are more developers for it.

Apparently OpenAFS does not really use unix accounts, and some chicanery is required to get the two to work together, and for all to be kosher. I am not a fan.

We are having issues with ocfs2. These might be solved by upgrading to a newer kernel, but it is not easy to find newer amd64 kernels with xen support. Either way, it is leaving us in a bad position, where one node will frequently find itself unable to create more than a few new files at a time. This means that monotone operations, tar -x operations, and compiling things are really out of the question in /home. Not good at all.

Vincas suggested OpenAFS, which seems similar to the glusterfs that I have been working with some. These offer a client/server model, more like an nfs replacement than an actual file system. This could work well, though it is not quite as clean, to my mind, as ocfs2. Since we are already using ldap, synchronizing UIDs isn’t a problem, so file permissions should work fine with either, or even with nfs.

OpenAFS has the advantage of having debian packages. One thing to keep in mind is that Dan dislikes it, though I do not know why. He’s certainly come down against some worthy software, while liking some less capable software, but on the other hand, his impressions are often sound.

Glusterfs is newer, and does not yet have real debian packages, though there are packages from a third party that the gluster guys have included in their wiki. On the other hand it has Dan approval. I am not really sure how which would make a more stable platform for us.

Yesterday I spent my work day converting a number of xen virtual servers into physical servers. There has got to be a better way than what I did, but I do not know of a rescue environment that has rsync installed (to my surprise, I tried several with no luck), and I did not have a usb stick or similar to install an OS on.

That is neither here nor there though. While I was converting, I came across a number of small but indicative things that lead me to believe that the administrators employed by the customer whose servers I was converting are in fact incredibly incompetent. Gah, that was a bad overly clauseful sentence.

Anyway, while nothing I found makes things certain, it was, as I said, certainly indicative. Today, I wake up, and they are complaining that files are missing from a vhost that I did not touch. At all. Sure enough, the file system is mounted but the files themselves are not there.

Looking in history, I cannot prove that they ran the rm -rf themselves. The alternative is that they were hacked. Either way, it feels like a sort of bitter “I told you so” moment, because yesterday I spent a few minutes complaining to Dan about the state of the converted machines, and suggesting he contact the customer about it. I do not think the customer realizes how incompetent his administrators are, nor that he has enough technical skills to interview a system administrator adequately. As a good business partner, I personally feel we owe it to this customer to have given him a heads up. And Dan might have done so for all I know.

Still, it is deeply ironic that we have had a catastrophe the very night after I warn Dan that they are asking for one.

bah.