forked from vrnetlab/vrnetlab
-
Notifications
You must be signed in to change notification settings - Fork 86
/
makefile-install.include
48 lines (46 loc) · 3.08 KB
/
makefile-install.include
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
#
# This Makefile can be included by images that need to run an install phase,
# i.e. in addition to doing the docker build, we also want to run some stuff
# inside that image to come up with the final output image. in the case of
# JUNOS we want to do this as the first time the vMX RE boots up it detects
# that it's in a vMX RE mode and then reboots. By starting it up and letting it
# do this first check-and-reboot during the image build time we save ourselves
# from doing this on *every* run of the container image later.
#
# Since we start the virtual router we are actually running a virtual machine
# with qemu and for that we want KVM hardware acceleration, which requires
# running docker with --privileged. `docker build` doesn't have the
# --privileged argument, so instead we first run the build as normal up to the
# point where we want to start the virtual router. Then we use `docker run
# --privilged ...` do the needful and after commit it using `docker commit ...`
# to create the final output image.
#
# One of the problems with this is that normally the docker build is kind of
# idempotent in that it uses a command cache and if there are no changes to the
# Dockerfile or input files it will not rerun those commands but use a cached
# image. This greatly speeds up the build process. However, when doing this
# manual `docker run` dance we miss this opportunity since it will always be
# run.... so we worked around it. Before doing docker run we check the SHA sum
# of the built image and compare this to the ones of the previously built
# image. If they are the same it means the docker build was entirely cached and
# there's no need to run the image, otherwise if there's a change we do run it.
# When comparing the hashes we omit the last layer of the previous build. It
# contains the committed changes from the install phase of the previous build.
# Include this makefile to have your image automatically do that dance.
docker-pre-build:
-cat cidfile | xargs --no-run-if-empty docker rm -f
-rm cidfile
-docker tag $(REGISTRY)vr-$(VR_NAME):$(VERSION) $(REGISTRY)vr-$(VR_NAME):$(VERSION)-previous-build
docker-build: docker-build-common
-docker inspect --format '{{.RootFS.Layers}}' $(REGISTRY)vr-$(VR_NAME):$(VERSION)-previous-build | tr -d '][' | awk '{ $$(NF)=""; print }' > built-image-sha-previous
docker inspect --format '{{.RootFS.Layers}}' $(REGISTRY)vr-$(VR_NAME):$(VERSION) | tr -d '][' > built-image-sha-current
if [ "$$(cat built-image-sha-previous | sed -e 's/[[:space:]]*$$//')" = "$$(cat built-image-sha-current)" ]; then echo "Previous image is the same as current, retagging!"; \
docker tag $(REGISTRY)vr-$(VR_NAME):$(VERSION)-previous-build $(REGISTRY)vr-$(VR_NAME):$(VERSION) || true; \
else \
echo "Current build differ from previous, running install!"; \
docker run --cidfile cidfile --privileged $(REGISTRY)vr-$(VR_NAME):$(VERSION) --trace --install; \
docker commit --change='ENTRYPOINT ["/launch.py"]' $$(cat cidfile) $(REGISTRY)vr-$(VR_NAME):$(VERSION); \
docker rm -f $$(cat cidfile); \
fi
docker rmi -f $(REGISTRY)vr-$(VR_NAME):$(VERSION)-previous-build || true
rm built-image-sha*