Re: Question regarding how databases support atomicity

I am trying to sharpen my understanding of databases. Let's say there is an operation foo as part of the public API that internally translates to more than 1 operation - I am sure there are examples like this in postgres. So to do foo we have to do following in order in all or none fashion:

1. Step 1

2. Step 2

3. Step 3

The way I understand this is that if there is a failure in-between, we start undoing and reverting the previous operations one by one. But what if there is a failure and we are not able to revert an operation. How is that situation handled? e.g., something failed when we tried to do Step 3. now we revert Step 2 and succeed. but when we try to revert step 1 we fail. what happens now? To me, it seems its impossible to guarantee true atomicity in general.

S.