build images using new hack/build-image script #135
Conversation
|
This pull request now has conflicts with the target branch. Please resolve these conflicts and force push the updated branch. |
phlogistonjohn
force-pushed
the
jjm-fqin-alt
branch
5 times, most recently
from
076fdc5 to
98fc5bf
Compare
phlogistonjohn
changed the title
|
@Mergifyio rebase |
✅ Branch has been successfully rebased |
There was a problem hiding this comment.
Thanks for doing this, @phlogistonjohn !
This is a clean improvement of the build mechanisms.
I reviewed the commits, read the new script, and played with it locally.
LGTM.
|
@phlogistonjohn needs fixes for failing checks now after rebase |
|
Seems like that job might be flaky. This is the 2nd time I just reran it and it worked on the retry. |
First, I wanted to ensure that if we ever decided to change how the tool is implemented we wouldn't have to rename it. Also, it's also very common to name executables in python without the .py extension (see commands like |
Add a python-based script `build-image` that takes on the complexities of how we build our images, including applying common tags and "fully-qualified image names". The script can be called from a makefile but doesn't require it. When used directly you can generate multiple image variants in one pass. Example: ``` ./hack/build-image -k server -p default -p nightly -a amd64 -a arm64 ``` Will produce four images, in short: 1. server with default packages on amd64 2. server with default packages on arm64 3. server with nightly packages on amd64 4. server with nightly packages on arm64 (all using the default distro, currently fedora) Signed-off-by: John Mulligan <[email protected]>
Signed-off-by: John Mulligan <[email protected]>
Signed-off-by: John Mulligan <[email protected]>
The "fake" name used in the image only works if you build the container locally first or are running in our CI. Since this is not a good general solution, I put a comment here as a reminder it should be fixed later. Signed-off-by: John Mulligan <[email protected]>
phlogistonjohn
force-pushed
the
jjm-fqin-alt
branch
from
f01475b to
087627f
Compare
Conflicts with #133#133 is closed.This PR is a working prototype of an alternate approach to #133. In this patch series I remove a lot of the complexity from the makefile and add a new scriptbuild-imagewritten in python to the hack dir.While the python ends up being longer, it allows the use of real data structures and is (IMO) a lot more straightforward and easy to read than the makefile based changes in #133. The use of python's argparse library means each command line option can be, and is, documented.
I want the team to take a look at the overall diff of both and we can debate which one is the better overall approach... or just take both as ideas and suggest a 3rd approach. ;-)Due to the recent churn related to the addition of workflow matrix in Github CI and a bunch of other changes getting this back into shape took the better part of a day. However, I think this is now in a usable and reviewable state. The current suite of tests pass.
I've tried to add a few helpful options, including
--printand--print-tagswhich displays the FQINs and FQINs plus additional tags respectively. These additional features made the script even longer but they paid off for me while I was debugging issues with the CI.Future PRs will need to change how we handle the "special tags" like latest, but first we'll need to move the scripts and such to the FQINs after this lands. Later, I plan on using manifests so that we can build and test {arm,amd}64 images seperately but then push them to qauy under the "latest" tag.