yaraskm
commented
2 years ago I know that you're busy with other things @jessfraz and @AkihiroSuda but is there anyway that you could take a quick look at this fix? It resolves a core functionality issue and I really want to avoid running a forked copy of the project long term.
The previous method would loop over the image references and try to save each one. Even if this did work, the tar archive would overwrite the previous instance on each invocation meaning that the last image specified on the CLI would be the only one in the archive. Even then, this method would hang because a filesystem lock was acquired on the underlying BoltDB during the first invocation of the loop (within
client.createWorkerOpt(...)) but never freed. Because of this, every subsequent call would only hang as seen in #332This refactor properly passes all image references to the underlying buildkit type to let it construct the image archive with all references specified.