On Call: a telco programmer swapped 2,650 lines of C for a six-line FTP shell script, and its three-strikes pager alert kept him up

Tim worked for a phone company and was asked to fix a data transfer program. It ran overnight, pulled data from the telco’s switches, tidied it up and uploaded it to a mainframe. Inside it sat a custom implementation of FTP, which is a lot of C for moving files.

He fixed it by deleting it. The 2,650 lines of C became a six-line shell script, according to The Register’s On Call column, written by Simon Sharwood. (The headline says 2,600; the body says 2,650.) The column puts the effort at eight hours: “Writing the script took eight hours, most of it tracking down the source code of FTP.”

That detail dates the story. This was before the web, and before Linux existed. AT&T had licensed Unix source to universities with strict conditions, so FTP’s code was not sitting on a public repo. Tim hunted for it with WAIS, the old search tool for remote databases. “It took me quite some time using WAIS to track it down – the security could have been better – and read the code,” he recalled. Eight hours to write six lines is a rate consultants dream about, and nearly all of it went to reading someone else’s FTP so he could replace it with the real thing.

The script did its job, and it also did what he told it to do when things went wrong. “It would automatically page me if it couldn’t connect to a switch or mainframe after three attempts,” the column says. The telco’s networks did not cooperate. The pager was “soon going off two or three times a night, every night.” Three strikes, and you’re up. (Mostly you.)

After two weeks without sleep, his manager told him to hand the pager to a colleague. That did not go well either, because the colleague “habitually left the pager at work, forgot to turn it on, or conveniently placed it out of earshot.” Tim took the pager back and kept the nightly interruptions for several more weeks.

No one in the column is shown fixing the networks that kept tripping the alert.

The pattern outlived the pager. Any alert triggered by a retry count fires at exactly the rate its network fails, and it reaches whoever wrote the check. Retries are cheap to add, and the rate of failures a person can absorb at 3 a.m. rarely gets measured before the pager goes live. Modern tooling mostly repeats the trick with a phone app, which is easier to leave on a desk and harder to ignore, so expect the same story to keep turning up with newer gear. The mechanism hasn’t changed: whoever writes the script inherits its noise.

Tim got the pager back.