来自Docker的Golang编译缓存

Question

I'm using the official golang alpine image to compile my source code (my host machine is a Mac), and I've noticed that even when mounting whole $GOPATH inside of the container it doesn't use cached data from previous builds. I checked that it creates it in the $GOPATH/pkg directory, but it does not affect the subsequent builds speed.

However, if I reuse the same container for several compilation, it does make use of some kind of cache, you can see the results in this experiment I did:

Using different containers, time remains around 28-30s in each build:

$ rm -r $GOPATH/pkg/linux_amd64
$ time docker run -v$GOPATH:/go -e CGO_ENABLED=0 golang:1.9-alpine3.6 go build -i github.com/myrepo/mypackage
...
0.02s user 0.08s system 0% cpu 30.914 total

$ time docker run -v$GOPATH:/go -e CGO_ENABLED=0 golang:1.9-alpine3.6 go build -i github.com/myrepo/mypackage
...
0.02s user 0.07s system 0% cpu 28.128 total

Reusing the same container, subsequent builds are much faster:

$ rm -r $GOPATH/pkg/linux_amd64    
$ docker run -d -v$GOPATH:/go -e CGO_ENABLED=0 golang:1.9-alpine3.6 tail -f /dev/null
bb4c08867bf2a28ad87facf00fa9dcf2800ad480fe1e66eb4d8a4947a6efec1d

$ time docker exec bb4c08867bf2 go build -i github.com/myrepo/mypackage
...
0.02s user 0.05s system 0% cpu 27.028 total

$ time docker exec bb4c08867bf2 go build -i github.com/myrepo/mypackage
0.02s user 0.06s system 0% cpu 7.409 total

Is Go using any kind of cache in some place outside of $GOPATH?

Answer 1

When you are building inside the golang container, it is using the directory $GOPATH/pkg inside this container. If you then start another golang container, it has an empty $GOPATH/pkg. However if you continue to use the same container (with exec), the $GOPATH/pkg is re-used.

rm -r $GOPATH/pkg/linux_amd64 will only remove this directory on your local machine. So this has no effect.

A possible alternative to re-using the same container could be

• to commit the container after the first build, or
• to mount $GOPATH/pkg as volume from your host machine or a data volume.