The God of Backups. ================== By Greg Rose. There is a God whose prerogatives include backups. I don't know this God's name, but I know He (or She, or perhaps It, after all I wouldn't want to get sexist when talking about this God...) is definitely a vengeful God. Vengeful, angry, full of wrath, innovative in devising punishments and with a warped sense of humour. Altogether, He (She, It) is not a fun Guy (Gal, Goo) to have around. Take a case in point. This is a true story, I know the names of the individuals and companies involved, but I'm not going to tell you. Anyway, there is this computer vendor, let's call them Vendor Inc., who sold a big, expensive computer to The Customer Corporation (say). This computer runs Customer's absolutely vital, not to mention huge, corporate everything database. This system and database have been installed for over ten years, and the Data Processing department of Customer Corp. has been doing its job quietly and happily for that time. Every six months, in addition to their regular backup schedule, they took a full system image dump, ending up with more than 10 tapes. Then, being the professionals that they are, Customer's Data Processing people sent the tapes to Vendor and paid a lot of money to have them loaded and verified. Vendor actually had to assemble a big enough machine and restore and run the system on it. Then the tapes were locked into a fireproof safe at Vendor's headquarters in case they were needed. Well, late in December, Customer's mainframe lost a cabinet full of drives to a small fire. "No Worries," said the D.P. Manager, a Mr. Lamb. "We'll just get new drives under maintenance and reload from the backups!" The replacement drives were installed within a day, after all Customer Corp was big and important, and they paid a lot for maintenance too. The tapes were retrieved from Vendor's safe, and the reload commenced. As it happens, this major backup was done in November, only a couple of weeks earlier. Tape 7 was the problem. Tape 7 couldn't be read. To be slightly more precise, what was there could be read, but there wasn't anything there. Nothing at all. It had been a brand new tape before the backup was done, of course, no expense was spared, and it had all the characteristics of a brand new tape now. It looked like some sort of procedural error had occured, and one of the other tapes had been written twice, or some such irrelevant thing. At this point, who cares what went wrong. Or, I should say, Who cares? It was, of course, recrimination time. Poor Mr. Lamb was called up before the Board of Customer Corp, and asked to explain why the system has been down for two weeks. Fortunately, he was able to point the finger; after all Vendor was supposed to ensure that the backups were all there. The CEO of Customer called the CEO of Vendor Inc, and after the conversation had settled down a bit, asked why the backups had been useless after customer had spent the last 13 years paying (a lot) for them to be verified? "I'll get back to you," said Vendor's CEO, with a sort of quaver in his voice. Vendor's CEO was noticeably more confident when he called back. "I have some bad news for you -- I hope you're sitting down," he said to Customer's Head Honcho. "When the tapes came last month there was a note attached. It said that you were fed up with paying so much to have the backups verified, after all they were never used, and you just wanted us to store them. It was signed by a Mr. B. Counter. Of course we did as requested. I'm sorry." Mr. Lamb was fired for failing to verify that the verification had happened. Somewhere, I'm sure, there was a sound of Hysterical Laughter. Many people think that "backup" is a noun. That it is an object that you refer back to when you need to. Others think that "backup" is a verb, that it is the act of making a copy of data. That is a bit closer. Any noun can be verbed. But here is the Truth. A backup is a religious ceremony that propitiates the God of Backups. It's obvious really. Everyone knows that if you take a good backup it will never be used. Even Murphy knew that. (Archives get used a lot. They are not backups. If you ever let your users know how easily you can restore files, they will delete important ones whenever they need temporary storage, then just ask for them back. That's an archive, not a backup.) The converse is also true though. If you forget to take a backup, that isn't so bad, the God has other people He can Hassle, for a while anyway. But if you take a backup onto bad media, or have a power outage in the middle of one, or take a backup with a script that expects to be run in "/" when your current directory was "/tmp", or any of these sorts of things, well, you asked for it. KerPow. The God of Backups wants his ceremonies done right, Or Else. Just to establish my credentials (and show how old I am) I'll give another example. We'd been running Version 5 Unix (no, that is not UNIX System V) for about six months when Version 6 came in. Doing a disk-to-disk copy of an RK05 using the block device took about 15 minutes. But V6 introduced raw devices! We had three drives, not surprisingly numbered 0, 1 and 2. No tape, but removable packs. All backups were disk-to-disk. Well, we went to single user mode, put the system backup pack in drive 1, and started to dd from /dev/rk0 to /dev/rk1. (I told you about nouns and verbs, didn't I?) About halfway, our administrator stopped the copying, and said "I'll just use the raw devices to make it faster." Of course, the moment the copying was interrupted, the God's beeper went off, and He (She, It) was watching in fascination. Our admin person quickly mknoded three times to create the raw devices. When the dd was restarted, this time on the raw devices, the lights on drives 0 and 2 (!) came on, and stayed on for a pleasing amount of time as entire tracks were copied. It was only about thirty seconds before we noticed that the wrong drive was lighting up. The DecWriter console still showed the typo that had mixed up the drive minor numbers, but it was too late. The system disk backup had the beginning of the current system disk, but the end of the old one. The actual system disk had the end of itself intact, but the beginning of it was clobbered by the front of the user disk. The situation might have been recovered at that point, since the whole system disk image existed, just in two pieces. Of course, since the system was running off that system disk, it promptly crashed. Maniacal Laughter. So when you are performing a backup, remember the proper sense of gravity in the situation. Your backup ceremony should be systematic. Label the tapes neatly, immediately after they come out of the drive. Store them properly. Don't interrupt a backup, and don't let anything stop you from finishing one. Establish a schedule of Worship, and stick to it. And don't ever start giggling during a backup, or tell backup jokes...