Using cache when building the image

[ondrej-fabry]

The Dockerfile seems to be split into two RUN instructions to make use of caching, hence the git clone in first and git pull in second. However build script currently uses --no-cache flag so it won't use the cache.

Any reason for this?

[unclejack]

@ondrej-fabry: That would avoid any potential caching of image. This should avoid using an outdated previously built image as that would be a bad thing.

[ondrej-fabry]

Not sure I follow. Why would it be a bad thing to cache first RUN instruction if the other one will executed
on change of VPP_COMMIT anyway.

[unclejack]

That's not the only time when the cache needs to be busted. The cache needs to be busted when any relevant packages are updated. Having built an image some months ago and using that cache isn't OK.

[ondrej-fabry]

Okay, I see your point now. However currently the cache will be always busted when using build script, right?

[unclejack]

We can set up a different build target to be used only for development which doesn't bust the cache.

[ondrej-fabry]

That sounds like good idea. It would provide both options and default could be busting the cache.

[tiewei]

This is mainly to make sure docker cloud doesn't have busted cache.
The make target currently reuse the codes, we can make it refer from env var if we want