Weblog outage

A new Blogger glitch has surfaced that may prove to be the last straw for my continued use of Blogger as a weblog publishing platform. Yesterday the main page and the current week's archive page for the Loosely Coupled weblog were offline most of the day, after Blogger left the pages as empty files with 0k contents. I didn't discover the problem until twelve hours later, when I saw an email from fellow-blogger and Grand Central CTO John McDowall, who'd written to let me know he'd noticed the pages were coming up blank (thanks, John).

My normal routine is to check that pages have published correctly, so it was unusual for me to miss such a catastrophic publishing glitch. It's only this morning that I've been able to piece together how this happened. Here's the story.

Ever since starting to use Blogger, I've been plagued by partial publishing errors. Maybe it's a special factor in the way their publishing engine talks to my website host (Jumpline). Who knows? Whatever the reason, it's been looked at various times, and sometimes it appeared to be fixed, but it's always come back again. The latest variation, however, is the deadliest of them all.

In the past, the problem has been that Blogger reports a transfer error when trying to FTP the finished files to my web server. The error says that Blogger hasn't been able to connect, whereas in fact it has managed to write some of the files, but not all of them. I first reported this problem, by the way, on May 9th, which was the first day of public beta testing of the current Blogger publishing engine. It finally seemed to have been fixed at the end of June, but came back again early September. It was fixed once more, but re-emerged again earlier this month.

Yesterday's problem is a nightmare mirror image of the longstanding original. In this new scenario, Blogger reports that it published successfully, whereas what it actually does is publish a blank file. What is happening is that, when Blogger FTPs to my server to write the new pages, it erases the previous contents of the file, but never actually writes the new contents, leaving a 0k file. Instead of generating a '404 page not found' error, the 0k file just comes up in the visitor's browser as a completely blank page. As you can imagine, that's not a problem most automated monitoring tools are good at spotting, so the only way to make sure it hasn't happened is to visually check the pages after publishing.

Why didn't I do that yesterday? I think it must have been because I had already gone back and republished the posting several times to make minor corrections, and I felt that the final change wasn't important enough to matter whether it had published or not. What I didn't realize was that there was a chance republishing would completely erase the whole page.

I have now had exactly the same problem this morning with another, private blog that I publish to the same server, and this time I got lots of transfer errors before I was able to successfully republish the blog instead of the blank page. So this is a problem that I don't want to live with.

All of this comes on top of the recent hacking, and coincides with my growing preference for a distributed server architecture — both of which lead me to wonder whether Blogger's architecture isn't too highly centralized to work effectively. But most of all, it comes down to the question of whether Blogger is a robust enough platform for running a business blog, and I'm reluctantly coming to the conclusion that it's not. That's not to say it's not fantastic value at $35 a year (or even $50 after its promised price rise). It's just that the quality control and reliability is not acceptable for a serious business website. Either the company has to add a higher-priced, more robust business version — and use the extra revenue to fund proper version management and problem resolution — or Blogger is going to lose customers like me to other platforms.

If we do decide to move, Movable Type will probably be the first choice alternative. But it's not going to happen yet, because the job has to go to the end of what is already quite a long list of development tasks on the site (including some long-promised RSS 2.0 features, which have already slipped down the priority list). In the meantime, I'm hoping the people at Pyra can fix these problems with Blogger. If not, at least the site's loosely coupled design principles will smooth the transfer. We already have our own server-based content management system for generating the home page, RSS feeds, archive indexes and other pages, so it is only the weblog core itself that we'd have to migrate.