The digest is calculated based on written json data, but JSONEncoder
doesn't preserve order bby default, so the digest is not consistent.
Signed-off-by: Alexey <makhov.alex@gmail.com>
## Changes
### README.md
- Fixed grammatical error: "when a process is ran" → "when a process is
run"
- Fixed plural form: "for instruction on" → "for instructions on"
### kernel/README.md
- Fixed numbering sequence: "4. Run `make`" → "3. Run `make`" in build
instructions
### Sources/Containerization/LinuxContainer.swift
- Fixed grammatical error in comment: "The container has ran" → "The
container has run"
We were doing intermediate step copies to Data objects both for push and
pull. We only need the data for the lifetime of the singular writes to
update the state of the checksums and to write to disk somewhere, so we
can use a view into the buffer from the http client to satisfy this.
We should return our own error type always for our methods. This wraps
all of the public methods on the process to return a more full featured
error than the rpcerrors we get today.
This change adds a new private method `waitIoComplete` on the
`LinuxProcess` type.
This method is called internally when the user calls `wait` for a
process - and it tries to give the IO streams some time to clear their
buffers.
Internally, this method sets up an `AsyncStream` down which an item is
sent when the vsock connection for either stdout/stderr is terminated.
We get this termination signal when the readability handler for the
associated fd fires with a no available data.
Inside the guest - once we are done relaying the IO from the process
into the socket connection, we close the socket fd which triggers the
above.
All this logic is wrapped around a timeout of 3 seconds, just to ensure
the method does not block forever.
---------
Signed-off-by: Aditya Ramani <a_ramani@apple.com>
- FileTimestamps constructor that actually let you provide values was
not public, so it wasn't possible to pass in values other than nil. This
change makes the other constructor and the underlying fields public.
- Rename EXT4+Format to EXT4+Formatter
- Rename EXT4+Export to EXT4Reader+Export
- Make the superblock publicly accessible in the reader like the docs
for the product states.
Fixes#103 (and should fix
https://github.com/apple/container/issues/108)
Today we only parse if the username field of the runtime spec is set,
but this misses if someone just supplied the integer uid and gid fields
only.
When we were testing the repo, we added authentication to most of the
registry tests so that we could access private images. Now that the
images are public, this PR removes the use of REGISTRY_TOKEN and
REGISTRY_USERNAME in tests that no longer require authenticating.
Note: REGISTRY_TOKEN, REGISTRY_USERNAME, and REGISTRY_HOST are still
required in the CI to push images and for a few registry tests. So the
env variables are not completely removed. The variables are only used in
the test tool cctl and in our swift tests.
Signed-off-by: Kathryn Baldauf <k_baldauf@apple.com>
- Corrected the typo in the method name (`Extened` → `Extended`)
- Updated all references to reflect the corrected spelling
- Introduced a `@available(*, deprecated)` alias for the original method
to maintain backward compatibility, as it was part of the public API
This PR ensures that we enter a lock inside the `@SendableProperty`
implementation as soon as we access a computed property. Additionally,
it mirrors the access level of the original property. Both changes are
required for [improved
accuracy](https://github.com/apple/container/pull/144) of progress
updates in container. Additionally, it should resolve
https://github.com/apple/containerization/issues/60 that occurs on
certain configurations.
Please tag as 0.1.1 after merging.
Use scoped access to the agent connection to ensure the vsock fds are
close(2)'d.
### Note
In a follow up I'd like to change `LinuxProcess` to dialing the agent
connection inside of `.start()`. I don't like needing to pass it an
agent connection. It has a reference to a vm so it can easily dial
inside itself if it needs to.
Fixes#85
The virtualization bool on LinuxContainer mostly just forwards to
VZVirtualMachineInstance which today would silently take your boolean
and do nothing if the underlying platform doesn't have support for it.
This is (to me) arguably worse than erroring, as it gives the client a
false security that the setting is on, and they should have virt
capabilities in the container/guest now.
This change makes it so that we throw a ContainerizationError for this
case, with a code of .unsupported so it's checkable by a user if they
want more information on the "why".
## Issue
When pulling images, the download speed appears to be slower compared to
Docker.
I found that parallel chunk generation was being performed during the
layer download process.
While individual chunks allow parallel download operations, the overall
process remains sequential between chunks.
This results less performance when chunks contain both small and large
layers mixed together.
## Changes
Discontinued chunk-based segmentation to enable more efficient parallel
downloads.
## Results(in my local env)
| image | layers | old | new |
| ---|---|---|---|
| node:latest | 12 | 1m15s | 1m10s |
| ghcr.io/norio-nomura/swift_discord_bot:main | 54 | 2m45s | 2m30s |
<details>
<summary>raw terminal log</summary>
```
❯ rm -rf ~/Library/Application\ Support/com.apple.containerization && time bin/cctl_old images pull docker.io/library/node:latest
image pulled
bin/cctl_old images pull docker.io/library/node:latest 42.37s user 6.60s system 64% cpu 1:15.94 total
❯ rm -rf ~/Library/Application\ Support/com.apple.containerization && time bin/cctl_new images pull docker.io/library/node:latest
image pulled
bin/cctl_new images pull docker.io/library/node:latest 42.27s user 6.72s system 69% cpu 1:10.58 total
❯ rm -rf ~/Library/Application\ Support/com.apple.containerization && time bin/cctl_old images pull docker.io/library/node:latest
image pulled
bin/cctl_old images pull docker.io/library/node:latest 45.65s user 7.36s system 70% cpu 1:15.64 total
❯ rm -rf ~/Library/Application\ Support/com.apple.containerization && time bin/cctl_new images pull docker.io/library/node:latest
image pulled
bin/cctl_new images pull docker.io/library/node:latest 39.76s user 6.32s system 65% cpu 1:10.50 total
❯ rm -rf ~/Library/Application\ Support/com.apple.containerization && time bin/cctl_old images pull docker.io/library/node:latest
image pulled
bin/cctl_old images pull docker.io/library/node:latest 42.47s user 6.75s system 65% cpu 1:14.72 total
❯ rm -rf ~/Library/Application\ Support/com.apple.containerization && time bin/cctl_new images pull docker.io/library/node:latest
image pulled
bin/cctl_new images pull docker.io/library/node:latest 42.28s user 6.65s system 69% cpu 1:09.93 total
❯ rm -rf ~/Library/Application\ Support/com.apple.containerization && time bin/cctl_new images pull ghcr.io/norio-nomura/swift_discord_bot:main
image pulled
bin/cctl_new images pull ghcr.io/norio-nomura/swift_discord_bot:main 103.83s user 18.71s system 81% cpu 2:30.02 total
❯ rm -rf ~/Library/Application\ Support/com.apple.containerization && time bin/cctl_old images pull ghcr.io/norio-nomura/swift_discord_bot:main
image pulled
bin/cctl_old images pull ghcr.io/norio-nomura/swift_discord_bot:main 120.79s user 20.70s system 85% cpu 2:45.26 total
```
</details>
<details>
<summary>patch for download only</summary>
```diff
diff --git a/Sources/cctl/ImageCommand.swift b/Sources/cctl/ImageCommand.swift
index 84c5218..4aa4bb8 100644
--- a/Sources/cctl/ImageCommand.swift
+++ b/Sources/cctl/ImageCommand.swift
@@ -127,6 +127,7 @@ extension Application {
}
print("image pulled")
+ return
let tempDir = FileManager.default.uniqueTemporaryDirectory(create: true)
if let platform {
```
</details>
- The kqueue handlers for stdio didn't have logic to exit on zero byte
reads.
- The agent wasn't getting closed explicitly in .delete()
- In LinuxContainer we should call delete just for sanity, if for
nothing more than ensuring the agent vsock fd is closed.
Signed-off-by: Danny Canter <danny_canter@apple.com>
This is something I forgot to update ages ago. The field names, while
they do make sense, were different than the constructor names that
eventually set the fields themselves. This is kind of awkward, and I'd
much rather fix this now while it's much easier.
We'll need to update `container` after this
Signed-off-by: Danny Canter <danny_canter@apple.com>
This creates an allocator based on a FIFO that will allocate through the
range before reusing previously released allocations.
Signed-off-by: michael crosby <michael_crosby@apple.com>
Resolves the warning
```
LocalOCILayoutClient.swift:87:37: warning: capture of non-sendable type 'T.AsyncIterator.Type' in an isolated closure
87 | for try await buffer in input {
| `- warning: capture of non-sendable type 'T.AsyncIterator.Type' in an isolated closure
88 | wrote += buffer.readableBytes
89 | try buffer.withUnsafeReadableBytes { pointer in
```
Signed-off-by: Aditya Ramani <a_ramani@apple.com>
This change adds documentation to quite a few existing public types that
didn't have a blurb before.
Additionally, this fixes a couple things that I think either didn't make
sense when going to document them:
- Rename ConnectionStream to VsockConnectionStream. This type only
functions for vsock connections.
- Deletes NsLock+Closure. This was not used anywhere.
- Rename ContainerizationOCI/Config.swift to ImageConfig.swift.
Signed-off-by: Danny Canter <danny_canter@apple.com>
Alexey Makhov