CI/CD

Say four students make a class newspaper together. Each one writes a separate article at home. You only find out how the others are writing on deadline day.

The night before the deadline, you put the four drafts together for the first time, and problems pour out. Two people described the same event differently, and the headline sizes are all over the place. You stay up all night tracing where things first went out of sync.

The longer the pieces stay apart, the harder they are to join again. So people came up with a different idea: don't wait for the deadline, combine a little every day. This habit of merging often so you catch mismatches early is called continuous integration.

Software teams are in the same boat. If many people each write their own part separately for weeks and then merge it all at once, the same thing happens as with the class newspaper the night before the deadline. So developers merge their own drafts into the shared pile of code many times a day.

Merging often isn't enough if people still have to hunt for mismatches by eye. So you attach a checker that runs by itself at every merge. In the newspaper, it's like a spell checker switching on the moment you upload an article to the shared folder.

In software, the thing that plays the checker's role is the automated test. A machine goes down a list of checks written in advance, one by one. If even one check fails, it stops the merge and immediately tells you whose line, and where, tripped it.

If the checks pass, the next stage follows. Just as the finished newspaper gets pinned to the hallway bulletin board without anyone carrying it, code that passes goes up to the live service server. This latter part is called continuous deployment.

This flow, where merging, checking, and shipping are linked in a line and run by themselves, is called a pipeline. People focus on writing, and the key idea is to leave the repeated checking and delivery to machines.

Passing the checker doesn't mean the newspaper has no flaws. A spell checker catches typos, but it can't tell whether the date written in an article is right. Automated tests also check only the items a person wrote down in advance.

So what you fill the check list with matters more than which tool you use. If the checks are thin, the pipeline lets bad code straight through. A fast-moving flow does its job only when good checks are in place.

One more thing: whether to hand the last step to machines too differs from team to team. Some teams only carry the newspaper up to the bulletin board, and a person decides after checking whether to pin it up. A setup that keeps a human approval step like this is called continuous delivery, as a separate term.

Both approaches aim at the same thing. If you merge small and often and check on the spot, you find mismatches early, and fixing them takes less effort. That means fewer all-nighters the night before the deadline.