Add Support for using a native Git binary to clone repositories in Spacelift. This would be really useful for teams that want more control over the checkout process, with options for partial cloning and mirror support as well.
+1 from Fountain. This is the fix for our biggest run-time cost. We run ~1,100 runs a week from a monorepo on public workers with sparse checkout enabled, and PREPARING still spends 50-66 s per run on source download. The worker does a depth-1 fetch with no blob filter, pulling a ~415 MB pack (75,988 blobs), then applies the sparse paths afterward; git in the worker reports “this worktree is not sparse”. A stack that checks out one folder downloads as much as one that checks out all of infrastructure/. The same commit and paths fetched as a partial clone (depth 1, filter=blob:none, sparse) are 16-32 MiB, about 8-12 s on your workers. That’s 14-17 worker-hours a week for us, more than everything we’ve cut from tofu init combined. A partial-clone option, or filter=blob:none by default when sparse paths are set, would solve it.
Log in to comment and vote
Comments1
Yellow Oak
Oct 22
+1 from Fountain. This is the fix for our biggest run-time cost. We run ~1,100 runs a week from a monorepo on public workers with sparse checkout enabled, and PREPARING still spends 50-66 s per run on source download. The worker does a depth-1 fetch with no blob filter, pulling a ~415 MB pack (75,988 blobs), then applies the sparse paths afterward; git in the worker reports “this worktree is not sparse”. A stack that checks out one folder downloads as much as one that checks out all of infrastructure/. The same commit and paths fetched as a partial clone (depth 1, filter=blob:none, sparse) are 16-32 MiB, about 8-12 s on your workers. That’s 14-17 worker-hours a week for us, more than everything we’ve cut from tofu init combined. A partial-clone option, or filter=blob:none by default when sparse paths are set, would solve it.