# The Backup That Was Never Tested

I ran the backup nightly for the better part of a year,
and never once restored it, which is why I write this here.

The job would finish quietly at twenty past the two,
and mail a line of green to say the nightly job was through.

But green meant "the script exited," and nothing more than that.
It did not mean the archive held a single useful byte.

A copy is a promise. A restore is a proof.
Until you pull it back again you have a paper roof.

So one wet Tuesday morning I went looking for a file,
and found nine months of archives that had failed in the same style.

A minor version update had revised the header spec.
The writer wrote it happily. The reader would not check.

Now let me give the other side its honest bit of ground:
a monthly restore drill is dull, and dull work goes unfound.

It costs an hour, it proves a thing you mostly think you know,
and every hour proving it is one you cannot grow.

I still say run it. Not because the odds are all that steep,
but because the day you need it is the day you cannot sleep.

And here is what will get you when the restore finally lands:
the data will come back just fine, the config will not stand.

You saved the rows, you saved the files, you saved the whole database.
You did not save the one small file that told them all their place.