Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

The way I've heard it used is thus:

In a complex system you are attempting to simplify (or deprecate components of), and where you cannot find an owner or stakeholder over some component system -- unplug the component and wait for errors or user complaints. Thus you will find out who is affected by the component and who likely stakeholders should be.

For example, old graybeards like myself will almost all have a story of finding some lost Sun or Irix server sitting under a pile of gear in an old closet, happily turned on with nobody current employed who knows anything about it.

What does it do? What's the user/password? How did it get there? all lost to the sands of time.

Turning it off will let you know if you can retire it or if somebody emerges with a problem then they can own it.



And then you accidentally crashed the whole business, for months, also known as a career limiting move. The stuff The Daily WTF published for failed companies.

Which is why only independent contractors can safely pull it off, and even then not always.

The correct test is targeted error injection. Make the thing a bit flaky see who complains or notices. Fan favorite is explicitly laggy or busted routers in the path for networked hardware. It's still not without risks, crappy old glue or scripts tend to fail catastrophically. So as usual, backups and spares. Documentation before messing with the thing...

Generally also found that it is safer to send excess garbage downstream from the putative server than induce outages. Spam gets quick results. Assumes of course you know what it is supposed to send... Or that garbage sent from there will get noticed.

The last thing is an actual smoke test.


At my company, we call this a scream test


There are only certain classes of systems this works for. If the system generates infrequent alerts, for example, the stakeholder won't know there's an issue until they independently discover that they missed an alert. That said, I've resorted to this approach before, and it works.


And then you suddenly torpedoed security notifications for the whole company. Nobody noticed for a year that updates were not applied. The rest of the story writes itself.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: