feat(*): init Rust port

This is a Rust port of the original Perl script, legacy cruft is removed
and it focuses on a modern Hydra deployment.

Nonetheless, it knows how to perform migrations based on the channel
versions.

Signed-off-by: Raito Bezarius <masterancpp@gmail.com>

.gitignore

@@ -1 +1,6 @@
 result
+
+
+# Added by cargo
+
+/target

Cargo.toml

@@ -0,0 +1,39 @@
+[package]
+name = "nixos-channel-scripts"
+version = "0.1.0"
+edition = "2021"
+
+[dependencies]
+clap = { version = "4.5.13", features = [ "derive" ] }
+indicatif = { version = "0.17.8", features = ["futures", "tokio"] }
+object_store = { version = "0.10.2", features = [ "aws" ] }
+regex = "1.10.6"
+reqwest = { version = "0.12.5", features = ["brotli"] }
+serde = "1.0.204"
+tempdir = "0.3.7"
+textwrap = "0.16.1"
+tokio = { version = "1.39.2", features = ["full"] }
+toml = "0.8.19"
+nix-compat = { git = "https://github.com/tvlfyi/tvix" }
+futures = "0.3.30"
+anyhow = "1.0.86"
+bytes = "1.7.1"
+tokio-util = { version = "0.7.11", features = ["io"] }
+termtree = "0.5.1"
+walkdir = "2.5.0"
+handlebars = "6.0.0"
+serde_json = "1.0.127"
+tracing = { version = "0.1.40", features = ["max_level_debug", "release_max_level_info"] }
+zstd = "0.13.2"
+tokio-stream = "0.1.15"
+async-compression = { version = "0.4.12", features = ["tokio", "zstd"] }
+digest = "0.10.7"
+blake3 = { version = "1.5.4", features = ["traits-preview"] }
+sha2 = "0.10.8"
+hex = "0.4.3"
+chrono = "0.4.38"
+tracing-subscriber = { version = "0.3.18", features = ["env-filter"] }
+opentelemetry = "0.24.0"
+opentelemetry_sdk = { version = "0.24.1", features = ["rt-tokio", "trace"] }
+tracing-opentelemetry = { version = "0.25.0", features = ["metrics"] }
+opentelemetry-otlp = "0.17.0"

default.nix

@@ -1,3 +1,11 @@
-(import (fetchTarball https://github.com/edolstra/flake-compat/archive/master.tar.gz) {
-  src = builtins.fetchGit ./.;
-}).defaultNix
+{ pkgs ? import <nixpkgs> {} }:
+{
+  shell = pkgs.mkShell {
+    buildInputs = [
+      pkgs.cargo
+      pkgs.rustc
+      pkgs.openssl
+      pkgs.pkg-config
+    ];
+  };
+}

forkos.toml

@@ -0,0 +1,6 @@
+hydra_uri = "https://hydra.forkos.org"
+binary_cache_uri = "https://cache.forkos.org"
+nixpkgs_dir = "/var/lib/nixpkgs"
+s3_release_bucket_name = "bagel-channel-scripts-test"
+s3_channel_bucket_name = "bagel-channel-scripts-test"
+base_git_uri_for_revision = "https://cl.forkos.org/plugins/gitiles/nixpkgs/+"

shell.nix

@@ -1,3 +0,0 @@
-(import (fetchTarball https://github.com/edolstra/flake-compat/archive/master.tar.gz) {
-  src = builtins.fetchGit ./.;
-}).shellNix

src/actions.rs

@@ -0,0 +1,612 @@
+use std::collections::HashMap;
+use std::path::PathBuf;
+use std::str::FromStr;
+use anyhow::Context;
+use bytes::Bytes;
+use digest::DynDigest;
+use futures::future::join_all;
+use futures::AsyncReadExt;
+use sha2::Digest;
+use tokio::io::{AsyncWrite, AsyncWriteExt};
+use futures::Stream;
+use futures::TryStreamExt;
+use indicatif::MultiProgress;
+use indicatif::ProgressBar;
+use indicatif::ProgressDrawTarget;
+use indicatif::ProgressStyle;
+use tracing::debug;
+use object_store::aws::AmazonS3;
+use object_store::WriteMultipart;
+use tempdir::TempDir;
+use tracing::trace;
+use crate::html::ReleaseFileData;
+use crate::html::ReleasePageData;
+use crate::hydra;
+use crate::config;
+
+const MULTIPART_CHUNK_SIZE: usize = 10 * 1024 * 1024; // 10 Mb
+const ZSTD_COMPRESSION_LEVEL: i32 = 11; // Reasonable compression speed w.r.t. to the size of
+                                        // store-paths.
+
+#[derive(Debug, Copy, Clone, Hash, Eq, PartialEq, clap::ValueEnum)]
+pub enum Checksum {
+    Sha256,
+    Blake3
+}
+
+impl FromStr for Checksum {
+    type Err = anyhow::Error;
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        Ok(match s {
+            "sha256" => Self::Sha256,
+            "blake3" => Self::Blake3,
+            _ => anyhow::bail!("unsupported checksum")
+        })
+    }
+}
+
+impl Checksum {
+    pub fn public_name(&self) -> &'static str {
+        match self {
+            Self::Sha256 => "SHA-256",
+            Self::Blake3 => "BLAKE-3"
+        }
+    }
+
+    fn into_dyn_digest(&self) -> Box<dyn DynDigest> {
+        match self {
+            Self::Sha256 => Box::new(sha2::Sha256::new()),
+            Self::Blake3 => Box::new(blake3::Hasher::new())
+        }
+    }
+
+    fn extension(&self) -> &'static str {
+        match self {
+            Self::Sha256 => "sha256",
+            Self::Blake3 => "b3"
+        }
+    }
+}
+
+#[derive(Debug, Copy, Clone)]
+pub enum HydraProductType {
+    BrotliJson,
+    SourceDistribution,
+}
+
+#[derive(Debug)]
+pub enum Compression {
+    Zstd,
+    None
+}
+
+impl Compression {
+    pub fn http_encoding(&self) -> Option<&str> {
+        Some(match self {
+            Self::Zstd => "zstd",
+            Self::None => return None,
+        })
+    }
+
+    pub fn file_extension(&self) -> Option<&str> {
+        Some(match self {
+            Self::Zstd => "zst",
+            Self::None => return None,
+        })
+    }
+}
+
+impl ToString for HydraProductType {
+    fn to_string(&self) -> String {
+        match self {
+            Self::BrotliJson => "json-br".to_string(),
+            Self::SourceDistribution => "source-dist".to_string(),
+        }
+    }
+}
+
+#[derive(Debug)]
+pub enum Action {
+    WriteFile {
+        dst_path: PathBuf,
+        contents: Vec<u8>
+    },
+    WriteHydraProduct {
+        product_name: String,
+        dst_path: Option<PathBuf>,
+        product_type: Option<HydraProductType>,
+    },
+    WriteCompressedStorePaths {
+        compression: Compression
+    }
+}
+
+#[derive(Debug)]
+pub struct Plan {
+    pub actions: Vec<Action>,
+    pub release: hydra::Release,
+    pub channel: hydra::Channel,
+    pub target_bucket: AmazonS3,
+    pub revision: String
+}
+
+#[derive(Debug)]
+pub struct PlanOptions {
+    pub streaming: bool,
+    pub enable_progress_bar: bool,
+    pub requested_checksums: Vec<Checksum>
+}
+
+macro_rules! plan_write_file {
+    ($plan_container:ident, $dst_path:literal, $contents:expr) => {
+        $plan_container.push(Action::WriteFile {
+            dst_path: $dst_path.into(),
+            contents: $contents
+        });
+    }
+}
+
+macro_rules! plan_write_hydra_product {
+    ($plan_container:ident, $pname:expr) => {
+        $plan_container.push(Action::WriteHydraProduct {
+            product_name: $pname.into(),
+            dst_path: None,
+            product_type: None
+        });
+    };
+
+    ($plan_container:ident, $dst_path:literal, $pname:expr) => {
+        $plan_container.push(Action::WriteHydraProduct {
+            dst_path: Some($dst_path.into()),
+            product_name: $pname.into(),
+            product_type: None
+        });
+    };
+
+    ($plan_container:ident, $dst_path:literal, $pname:expr, $ptype:expr) => {
+        $plan_container.push(Action::WriteHydraProduct {
+            dst_path: Some($dst_path.into()),
+            product_name: $pname.into(),
+            product_type: Some($ptype)
+        });
+    };
+}
+
+macro_rules! plan_iso {
+    ($plan_container:ident, $iso_name:literal, $arch:literal) => {
+        plan_write_hydra_product!($plan_container, format!("nixos.iso_{}.{}", $iso_name, $arch));
+    }
+}
+
+#[tracing::instrument]
+pub async fn generate_plan(config: &config::MirrorConfig, channel: &hydra::Channel, release: &hydra::Release, evaluation: &hydra::Evaluation) -> Vec<Action> {
+    let mut plan = vec![];
+
+    plan.push(Action::WriteCompressedStorePaths {
+        compression: Compression::Zstd
+    });
+
+    if let hydra::Input::Git { revision, .. } = evaluation.jobset_eval_inputs.get("nixpkgs").unwrap() {
+        plan_write_file!(plan, "src-url", release.evaluation_url(&config.hydra_uri).into());
+        plan_write_file!(plan, "git-revision", revision.clone().into());
+        plan_write_file!(plan, "binary-cache-url", config.binary_cache_uri.clone().into());
+
+        if matches!(channel.r#type(), hydra::ChannelType::NixOS) {
+            plan_write_hydra_product!(plan, "nixexprs.tar.xz", "nixos.channel");
+            plan_write_hydra_product!(plan, "packages.json.br", "nixpkgs.tarball",
+                HydraProductType::BrotliJson
+            );
+            plan_write_hydra_product!(plan, "options.json.br", "nixos.options",
+                HydraProductType::BrotliJson
+            );
+
+            // TODO: we don't have aarch64-linux yet
+            //plan_iso!(plan, "minimal", "aarch64-linux");
+            //plan_iso!(plan, "minimal", "x86_64-linux");
+
+            if !matches!(channel.variant(), hydra::ChannelVariant::Small) {
+                // TODO: for some reason, those ISOs are not present in ForkOS hydra.
+                // TODO: implement error recovery, skipping or failure?
+                // maybe encode this in the plan (fallible, etc.)
+                // TODO: compat with nixos-20, 21, 22, 23 using plasma5?
+                // should we have plasma_latest?
+                //plan_iso!(plan, "plasma6", "aarch64-linux");
+                //plan_iso!(plan, "plasma6", "x86_64-linux");
+
+                //plan_iso!(plan, "gnome", "aarch64-linux");
+                //plan_iso!(plan, "gnome", "x86_64-linux");
+
+                // TODO: i686-linux ISO compat? for nixos ≤ 23.11
+                
+                // FIXME: weird OVA appliance here, should investigate what to do
+                //plan_write_hydra_product!(plan, "nixos.ova.x86_64-linux");
+            }
+        } else {
+            plan_write_hydra_product!(plan, "nixexprs.tar.xz", "tarball",
+                HydraProductType::SourceDistribution);
+            plan_write_hydra_product!(plan, "packages.json.br", "tarball",
+                HydraProductType::BrotliJson);
+        }
+    }
+
+    // TODO: take into account the SQLite database & debuginfo
+
+    plan
+}
+
+#[derive(Debug)]
+pub enum PlanInterpretation {
+    /// The plan has been streamed to the target object storage
+    /// It's ready to be applied by promoting atomically this temporary object store path.
+    Streamed {
+        remote_path: object_store::path::Path,
+        target_path: object_store::path::Path
+    },
+    /// The plan has been prepared locally and can now be inspected or uploaded.
+    LocallyPlanned {
+        local_path: TempDir,
+        target_path: object_store::path::Path
+    }
+}
+
+pub async fn stream_to_multipart_writer<S, F>(mut input_stream: S, mut stream_callback: F, mut writer: WriteMultipart) -> std::io::Result<usize> where 
+    S: Stream<Item = std::io::Result<Bytes>> + Unpin,
+    F: FnMut(&Bytes),
+{
+    let mut streamed = 0;
+    while let Some(bytes) = input_stream.try_next().await? {
+        streamed += bytes.len();
+        stream_callback(&bytes);
+        writer.put(bytes);
+    }
+
+    writer.finish().await?;
+
+    Ok(streamed)
+}
+
+
+pub async fn stream_to_async_writer<S, F, W>(mut input_stream: S, _size_hint: Option<usize>, mut stream_callback: F, mut target_file: W) -> std::io::Result<usize> where
+S: Stream<Item = std::io::Result<Bytes>> + Unpin,
+F: FnMut(&Bytes),
+W: AsyncWrite + Unpin
+{
+    let mut streamed = 0;
+    while let Some(bytes) = input_stream.try_next().await? {
+        streamed += bytes.len();
+        target_file.write_all(&bytes).await?;
+        stream_callback(&bytes);
+    }
+
+    Ok(streamed)
+}
+
+pub async fn stream_to_bucket<S, F, O>(input_stream: S, size_hint: Option<usize>, target_path: &object_store::path::Path, mut stream_callback: F, bucket: &O) -> std::io::Result<usize> where
+S: Stream<Item = std::io::Result<Bytes>> + Unpin,
+F: FnMut(&Bytes),
+O: object_store::ObjectStore
+{
+    let uses_multipart = if let Some(size) = size_hint {
+        size >= MULTIPART_CHUNK_SIZE 
+    } else { true }; // Worst case is that the stream is bigger than memory.
+
+    if uses_multipart {
+        // TODO: propagate error
+        let mpart = bucket.put_multipart(target_path).await.unwrap();
+        let writer = WriteMultipart::new_with_chunk_size(mpart, MULTIPART_CHUNK_SIZE);
+
+        stream_to_multipart_writer(input_stream, stream_callback, writer).await
+    } else {
+        let mut buffer = if let Some(size) = size_hint {
+            Vec::with_capacity(size)
+        } else {
+            // Constantify
+            Vec::with_capacity(8192)
+        };
+
+        // TODO: find a way to propagate the progress bar here.
+        // call the stream callback otherwise we cannot hash entire files.
+        input_stream.into_async_read().read_to_end(&mut buffer).await?;
+        let actual_length = buffer.len();
+        // TODO: generalize the stream callback argument
+        // it's not that bad because it's usually very small here.
+        stream_callback(&Bytes::copy_from_slice(&buffer));
+
+        bucket.put(target_path, buffer.into()).await?;
+
+        Ok(actual_length)
+    }
+}
+
+pub async fn stream_to_target_nock<S, F, O>(input_stream: S,
+    size_hint: Option<usize>, 
+    key_prefix: &str, key: &str, 
+    stream_callback: F, 
+    bucket: &O, streaming: bool) -> std::io::Result<usize> where
+S: Stream<Item = std::io::Result<Bytes>> + Unpin,
+F: FnMut(&Bytes),
+O: object_store::ObjectStore {
+    if streaming {
+        let bucket_path: object_store::path::Path = format!("{}/{}", key_prefix, key).into();
+        stream_to_bucket(input_stream, size_hint, &bucket_path, stream_callback, bucket).await
+    } else {
+        let fs_path: PathBuf = PathBuf::from(key_prefix).join(key);
+        let tgt_file = tokio::fs::File::create(&fs_path).await?;
+        stream_to_async_writer(input_stream, size_hint, stream_callback, tgt_file).await
+    }
+}
+
+
+pub async fn stream_to_target<S, F, O>(input_stream: S,
+    size_hint: Option<usize>, 
+    key_prefix: &str, key: &str, 
+    stream_callback: F, 
+    bucket: &O, streaming: bool,
+    requested_checksums: Vec<Checksum>) -> std::io::Result<(usize, HashMap<Checksum, Box<[u8]>>)> where
+S: Stream<Item = std::io::Result<Bytes>> + Unpin,
+F: Fn(&Bytes),
+O: object_store::ObjectStore {
+    let mut hashers: HashMap<Checksum, Box<dyn DynDigest>> = HashMap::from_iter(
+        requested_checksums.into_iter().map(|checksum| (checksum, checksum.into_dyn_digest()))
+    );
+
+    let copied = stream_to_target_nock(input_stream, size_hint, key_prefix, key, |bytes| {
+        hashers.iter_mut().for_each(|(_, hasher)| {
+            hasher.update(&bytes);
+        });
+
+        stream_callback(&bytes);
+    }, bucket, streaming).await?;
+
+    // Stream all the checksums as well.
+    let keys: HashMap<Checksum, String> = hashers.iter().map(|(ck, _)| (ck.clone(), format!("{}.{}", key, ck.extension()))).collect();
+    let checksums: HashMap<Checksum, Box<[u8]>> = hashers.into_iter().map(|(ck, hasher)| (ck, hasher.finalize())).collect();
+
+    join_all(checksums.iter().map(|(ck, digest)| {
+        let key = keys.get(ck).unwrap();
+        let hexdigest = hex::encode(digest);
+        let ck_stream = tokio_stream::once(
+            Ok(Bytes::from_iter(hexdigest.into_bytes().into_iter()))
+        );
+
+        stream_to_target_nock(ck_stream,
+            Some(digest.as_ref().len()),
+            key_prefix,
+            key,
+            |bytes| {
+                trace!("streaming the checksum ({:?}); chunk size = {}", ck.clone(), bytes.len());
+            }, bucket, streaming)
+    })).await;
+
+    Ok((copied, checksums))
+}
+
+/// Interprets the plan as a list of action.
+///
+/// It creates a staging directory where all actions are prepared.
+/// No irreversible action is committed.
+///
+/// At the end, it returns the staging directory for further post-processing, including possible
+/// uploads.
+///
+/// Enabling streaming will also directly stream the contents of the staging directory to the
+/// target bucket in a temporary suffix of the target prefix.
+/// Caller is responsible to promote that upload to a finalized upload.
+#[tracing::instrument]
+pub async fn interpret_plan(hydra_client: &hydra::HydraClient<'_>, plan: Plan, plan_opts: PlanOptions) -> PlanInterpretation {
+    // Prepare a staging directory
+    let staging_directory = TempDir::new(format!("staging-{}", &plan.channel.name).as_str()).expect("Failed to create temporary directory for staging directory");
+    let staging_prefix = staging_directory.path().file_name()
+        .expect("Failed to obtain the final component of the staging directory")
+        .to_str()
+        .expect("Failed to convert the final component into string")
+        .to_owned();
+    let streaming = plan_opts.streaming;
+
+    let key_prefix = if streaming {
+        staging_prefix.to_string()
+    } else {
+        staging_directory.path().to_str().expect("Staging directory path is not valid UTF-8").to_owned()
+    };
+
+    if streaming {
+        debug!("Streaming the update to /{}", staging_prefix);
+
+    } else {
+        debug!("Preparing locally the update in {}", staging_directory.path().display());
+    }
+
+    // Initialize multiple progress bars if we are to show progress bars.
+    let multi_pbar = MultiProgress::with_draw_target(if plan_opts.enable_progress_bar {
+        ProgressDrawTarget::stdout() 
+    } else {
+        ProgressDrawTarget::hidden()
+    });
+
+    // Pending download/upload/streaming tasks.
+    let mut pending = Vec::new();
+
+    for action in plan.actions {
+        let task = async {
+            let pbar = ProgressBar::new(0);
+            multi_pbar.add(pbar.clone());
+            pbar.set_style(ProgressStyle::with_template("[{elapsed_precise}] {bar:40.cyan/blue} {bytes:>7}/{total_bytes:>7} @ {bytes_per_sec:>7} {msg}").unwrap().progress_chars("##-"));
+            let requested_checksums = plan_opts.requested_checksums.clone();
+
+            match action {
+                Action::WriteHydraProduct { product_name, dst_path, product_type } => {
+                    let dst_path = dst_path.unwrap_or_else(|| product_name.clone().into());
+                    assert!(dst_path.is_relative(), "Unexpected absolute path in a channel script plan, all paths needs to be relative in order for them to be prefixed by the staging directory");
+
+                    let key = dst_path.to_str().expect("Failed to convert relative destination path to a UTF-8 key");
+                    let (size, stream) = hydra_client.hydra_product_bytes_stream(&plan.release, &product_name, product_type).await
+                        .context("while preparing the stream for a Hydra product")
+                        .expect("Failed to stream an Hydra product");
+
+                    pbar.set_length(size);
+                    if streaming {
+                        pbar.set_message(format!("Downloading and streaming '{}'", product_name));
+                    } else {
+                        pbar.set_message(format!("Copying '{}'", product_name));
+                    }
+
+                    let (copied, checksums) = stream_to_target(stream, Some(size as usize), &key_prefix, key, |bytes| {
+                        // This is not really called when the multipart
+                        // has completed pushing this part.
+                        // In practice, if we are pushing as soon as we read,
+                        // we expect the bytes chunk to be small
+                        // and therefore, this is fairly reasonable.
+                        trace!("multipart streaming: chunk size = {}", bytes.len());
+                        pbar.inc(bytes.len() as u64);
+                    }, &plan.target_bucket, streaming, requested_checksums).await
+                    .context(
+                        format!("while streaming to the bucket the product {}", product_name))
+                    .unwrap();
+
+                    pbar.finish_with_message(format!("Streamed '{}'", product_name));
+
+                    debug!("Copied {} bytes", copied);
+
+                    return ReleaseFileData::new(key, size, checksums);
+                },
+                Action::WriteFile { dst_path, contents } => {
+                    // We don't need a progress bar here.
+                    assert!(dst_path.is_relative(), "Unexpected absolute path in a channel script plan, all paths needs to be relative in order for them to be prefixed by the staging directory");
+                    let key = dst_path.to_str().expect("Failed to convert a relative path to a UTF-8 key");
+                    let size = contents.len();
+                    let stream = tokio_stream::once(Ok(Bytes::from_iter(contents.into_iter())));
+                    pbar.set_length(size as u64);
+                    if streaming {
+                        pbar.set_message(format!("Streaming '{}'", key));
+                    } else {
+                        pbar.set_message(format!("Writing '{}'", key));
+                    }
+                    let (copied, checksums) = stream_to_target(stream, Some(size), &key_prefix, key, |bytes| {
+                       // This is not really called when the multipart
+                        // has completed pushing this part.
+                        // In practice, if we are pushing as soon as we read,
+                        // we expect the bytes chunk to be small
+                        // and therefore, this is fairly reasonable.
+                        trace!("multipart streaming: chunk size = {}", bytes.len());
+                        pbar.inc(bytes.len() as u64);
+                    }, &plan.target_bucket, streaming, requested_checksums).await
+                    .context(
+                        format!("while streaming to the bucket the following key '{}'", key)
+                    ).unwrap();
+
+                    pbar.finish_with_message(format!("Written '{}'", key));
+
+                    debug!("Written {} bytes", copied);
+
+                    return ReleaseFileData::new(key, size as u64, checksums);
+                },
+                Action::WriteCompressedStorePaths { compression } => {
+                    let spaths = hydra_client.fetch_store_paths(&plan.release).await.context("while fetching store paths").expect("Failed to fetch store paths from Hydra");
+                    let mut serialized = serde_json::to_vec(&spaths).context("while serializing store paths to JSON").expect("failed to serialize");
+
+                    let key;
+
+                    if matches!(compression, Compression::Zstd) {
+                        let compressed_serialization = zstd::bulk::compress(&serialized, ZSTD_COMPRESSION_LEVEL).expect("Failed to compress the store-paths");
+                        let _ = std::mem::replace(&mut serialized, compressed_serialization);
+                        key = "store-paths.zst";
+                    } else {
+                        key = "store-paths";
+                    }
+
+                    let size = serialized.len();
+                    let stream = tokio_stream::once(Ok(Bytes::from_iter(serialized.into_iter())));
+
+                    pbar.set_length(size as u64);
+                    if streaming {
+                        pbar.set_message("Streaming compressed store paths");
+                    } else {
+                        pbar.set_message("Writing compressed store paths");
+                    }
+
+                    let (copied, checksums) = stream_to_target(stream, Some(size), &key_prefix, key, |bytes| {
+                       // This is not really called when the multipart
+                        // has completed pushing this part.
+                        // In practice, if we are pushing as soon as we read,
+                        // we expect the bytes chunk to be small
+                        // and therefore, this is fairly reasonable.
+                        trace!("multipart streaming: chunk size = {}", bytes.len());
+                        pbar.inc(bytes.len() as u64);
+                    }, &plan.target_bucket, streaming, requested_checksums).await
+                    .context(
+                        format!("while streaming to the bucket the following key '{}'", key)
+                    ).unwrap();
+
+                    pbar.finish_with_message("Compressed store paths streamed");
+                    debug!("Written {} bytes", copied); 
+
+                    return ReleaseFileData::new(key, size as u64, checksums);
+                },
+            };
+        };
+
+        pending.push(task);
+    }
+
+    let files: Vec<ReleaseFileData> = join_all(pending).await;
+    let target_path = plan.release.prefix(&plan.channel);
+    let rendered_page = crate::html::render_release_page(ReleasePageData {
+        channel_name: plan.channel.name.clone(),
+        release_name: plan.release.nix_name.clone(),
+        release_date: chrono::Utc::now().format("%Y-%m-%d %H-%M UTC").to_string(),
+        git_commit_link: format!("{}/{}", hydra_client.config.base_git_uri_for_revision, plan.revision),
+        git_revision: plan.revision,
+        // TODO: mega evil, clean up and expose config properly.
+        hydra_eval_link: plan.release.evaluation_url(&hydra_client.config.hydra_uri),
+        hydra_eval_id: plan.release.evaluation_id(),
+        checksums: plan_opts.requested_checksums.iter().map(|ck| ck.public_name().to_string()).collect(),
+        files
+    }).expect("Failed to render a release page");
+
+
+    let pbar = ProgressBar::new(0);
+    multi_pbar.add(pbar.clone());
+    pbar.set_style(ProgressStyle::with_template("[{elapsed_precise}] {bar:40.cyan/blue} {bytes:>7}/{total_bytes:>7} @ {bytes_per_sec:>7} {msg}").unwrap().progress_chars("##-"));
+    let key = if streaming {
+        plan.release.nix_name
+    } else {
+        format!("{}.html", plan.release.nix_name)
+    };
+    let contents = rendered_page.into_bytes();
+    let size = contents.len();
+    let stream = tokio_stream::once(Ok(Bytes::from_iter(contents.into_iter())));
+    pbar.set_length(size as u64);
+    if streaming {
+        pbar.set_message("Streaming the release HTML page");
+    } else {
+        pbar.set_message("Writing the release HTML page");
+    }
+    let (copied, _) = stream_to_target(stream, Some(size), &key_prefix, &key, |bytes| {
+       // This is not really called when the multipart
+        // has completed pushing this part.
+        // In practice, if we are pushing as soon as we read,
+        // we expect the bytes chunk to be small
+        // and therefore, this is fairly reasonable.
+        trace!("multipart streaming: chunk size = {}", bytes.len());
+        pbar.inc(bytes.len() as u64);
+    }, &plan.target_bucket, streaming, Vec::new()).await
+    .context(
+        format!("while streaming to the bucket the following key '{}'", key)
+    ).unwrap();
+
+    debug!("Release HTML page sent; copied {} bytes", copied);
+    pbar.finish_with_message("Release HTML page streamed");
+
+    if streaming {
+        staging_directory.close().expect("Failed to get rid of the staging directory");
+        PlanInterpretation::Streamed {
+            remote_path: staging_prefix.into(),
+            target_path
+        }
+    } else {
+        PlanInterpretation::LocallyPlanned {
+            local_path: staging_directory,
+            target_path
+        }
+    }
+}

src/config.rs

@@ -0,0 +1,36 @@
+use object_store::aws::{AmazonS3, AmazonS3Builder};
+use serde::Deserialize;
+use std::path::PathBuf;
+
+#[derive(Debug, Deserialize)]
+pub struct MirrorConfig {
+    /// URI to Hydra instance
+    pub hydra_uri: String,
+    /// URI to the binary cache
+    pub binary_cache_uri: String,
+    /// Base URI to access to a web interface pertaining to a specific Git revision of the nixpkgs
+    /// input
+    pub base_git_uri_for_revision: String,
+    /// A path to a checkout of nixpkgs
+    nixpkgs_dir: PathBuf,
+    /// S3 releases bucket name
+    pub s3_release_bucket_name: String,
+    /// S3 channels bucket name
+    s3_channel_bucket_name: String,
+}
+
+impl MirrorConfig {
+    pub fn release_bucket(&self) -> AmazonS3 {
+        AmazonS3Builder::from_env()
+            .with_bucket_name(&self.s3_release_bucket_name)
+            .build()
+            .expect("Failed to connect to the S3 release bucket")
+    }
+
+    pub fn channel_bucket(&self) -> AmazonS3 {
+        AmazonS3Builder::from_env()
+            .with_bucket_name(&self.s3_channel_bucket_name)
+            .build()
+            .expect("Failed to connect to the S3 channel bucket")
+    }
+}

src/git.rs

@@ -0,0 +1 @@
+

src/html.rs

@@ -0,0 +1,49 @@
+use std::collections::HashMap;
+
+use handlebars::Handlebars;
+use indicatif::HumanBytes;
+use serde::Serialize;
+
+use crate::actions::Checksum;
+
+
+#[derive(Debug, Serialize)]
+pub struct ReleaseFileData {
+    pub name: String,
+    pub size: String,
+    pub checksums: HashMap<String, String>
+}
+
+impl ReleaseFileData {
+    pub fn new(name: &str, size: u64, checksums: HashMap<Checksum, Box<[u8]>>) -> Self {
+        ReleaseFileData {
+            name: name.to_string(),
+            size: HumanBytes(size).to_string(),
+            checksums: checksums.into_iter().map(|(ck, digest)| {
+                (ck.public_name().to_string(),
+                hex::encode(digest))
+            }).collect()
+        }
+    }
+}
+
+#[derive(Debug, Serialize)]
+#[serde(rename_all = "camelCase")]
+pub struct ReleasePageData {
+    pub channel_name: String,
+    pub release_name: String,
+    pub release_date: String,
+    pub git_commit_link: String,
+    pub git_revision: String,
+    pub hydra_eval_link: String,
+    pub hydra_eval_id: u64,
+    pub files: Vec<ReleaseFileData>,
+    pub checksums: Vec<String>
+}
+
+pub fn render_release_page(page_data: ReleasePageData) -> Result<String, handlebars::RenderError> {
+    let reg = Handlebars::new();
+
+    let release_template = include_str!("./release.tpl");
+    reg.render_template(&release_template, &page_data)
+}

src/hydra.rs

@@ -0,0 +1,338 @@
+use std::{collections::HashMap, path::PathBuf};
+use anyhow::Context;
+use bytes::Bytes;
+use futures::Stream;
+use indicatif::ProgressBar;
+use nix_compat::store_path::{StorePath, StorePathRef};
+use regex::Regex;
+
+use serde::{de::Error, Deserialize, Deserializer};
+use tokio::io::AsyncWrite;
+
+use crate::{actions::HydraProductType, config::MirrorConfig};
+
+pub type ReleaseId = u64;
+pub type EvaluationId = u64;
+pub type BuildId = u64;
+
+fn deser_bool_as_int<'de, D>(deserializer: D) -> Result<bool, D::Error>
+    where D: Deserializer<'de>
+{
+    u8::deserialize(deserializer).map(|b| if b == 1 { true } else { false })
+}
+
+fn deser_bool_as_string<'de, D>(deserializer: D) -> Result<bool, D::Error>
+where D: Deserializer<'de>
+{
+    let s: &str = Deserialize::deserialize(deserializer)?;
+
+    match s {
+        "false" => Ok(false),
+        "true" => Ok(true),
+        _ => Err(Error::unknown_variant(s, &["false", "true"]))
+    }
+}
+
+#[derive(Debug, Deserialize)]
+pub struct OutputPath {
+    #[serde(rename="path")]
+    store_path: String
+}
+
+#[derive(Debug, Deserialize)]
+pub struct BuildProduct {
+    #[serde(rename="path")]
+    store_path: String,
+    name: String,
+    subtype: String,
+    sha256hash: String,
+    r#type: String,
+    #[serde(rename="filesize")]
+    file_size: u64
+}
+
+#[derive(Debug, Deserialize)]
+pub struct BuildInfo {
+    jobset: String,
+    job: String,
+    project: String,
+    #[serde(deserialize_with = "deser_bool_as_int", rename="buildstatus")]
+    finished: bool,
+    #[serde(rename="releasename")]
+    release_name: Option<String>,
+    #[serde(rename="stoptime")]
+    stop_time: u64,
+    #[serde(rename="starttime")]
+    start_time: u64,
+    system: String,
+    id: BuildId,
+    #[serde(rename="buildoutputs")]
+    build_outputs: HashMap<String, OutputPath>,
+    #[serde(rename="buildproducts")]
+    build_products: HashMap<String, BuildProduct>,
+    #[serde(rename="nixname")]
+    nix_name: String,
+    #[serde(rename="drvpath")]
+    drv_path: String,
+}
+
+#[derive(Debug, Clone)]
+pub struct Channel {
+    pub name: String
+}
+
+#[non_exhaustive]
+pub enum ChannelType {
+    NixOS,
+    Nixpkgs,
+}
+
+#[non_exhaustive]
+pub enum ChannelVariant {
+    Normal,
+    Small
+}
+
+impl Channel {
+    pub fn version(&self) -> String {
+        let re = Regex::new("([a-z]+)-(?<ver>.*)").unwrap();
+        let caps = re.captures(&self.name).expect("Failed to parse the channel name");
+
+        caps["ver"].to_string()
+    }
+
+    pub fn prefix(&self) -> String {
+        let re = Regex::new("(?<name>[a-z]+)-(.*)").unwrap();
+        let caps = re.captures(&self.name).expect("Failed to parse the channel name");
+
+        caps["name"].to_string()
+    }
+
+    // TODO: just regex match?
+    pub fn r#type(&self) -> ChannelType {
+        ChannelType::NixOS
+    }
+
+    pub fn variant(&self) -> ChannelVariant {
+        ChannelVariant::Normal
+    }
+}
+
+#[derive(Deserialize, Debug)]
+pub struct Release {
+    pub id: ReleaseId,
+    job: String,
+    #[serde(rename = "releasename")]
+    release_name: Option<String>,
+    #[serde(rename = "starttime")]
+    start_time: u64,
+    #[serde(rename = "stoptime")]
+    stop_time: u64,
+    #[serde(rename = "nixname")]
+    pub nix_name: String,
+    #[serde(rename = "jobsetevals")]
+    jobset_evals: Vec<EvaluationId>,
+    jobset: String,
+    #[serde(deserialize_with = "deser_bool_as_int")]
+    finished: bool,
+    priority: u64,
+    system: String,
+    timestamp: u64,
+    project: String,
+    #[serde(rename = "drvpath")]
+    derivation_path: String,
+    // ignored: buildproducts, buildoutputs, buildmetrics, buildstatus
+}
+
+#[derive(Debug)]
+pub struct GitInput {
+    uri: String,
+    revision: String
+}
+
+// FIXME(Raito): for some reason, #[serde(tag = "type"), rename_all = "lowercase"] doesn't behave
+// correctly, and causes deserialization failures in practice. `untagged` is suboptimal but works
+// because of the way responses works...
+#[derive(Debug, Deserialize)]
+#[serde(untagged, expecting = "An valid jobset input")]
+pub enum Input {
+    Boolean {
+        #[serde(deserialize_with = "deser_bool_as_string")]
+        value: bool
+    },
+    Git { uri: String, revision: String },
+    Nix { value: String },
+}
+
+#[derive(Deserialize, Debug)]
+pub struct Evaluation {
+    pub id: EvaluationId,
+    #[serde(rename="checkouttime")]
+    checkout_time: u64,
+    #[serde(rename="evaltime")]
+    eval_time: u64,
+    flake: Option<String>,
+    #[serde(rename="jobsetevalinputs", default)]
+    pub jobset_eval_inputs: HashMap<String, Input>,
+    timestamp: u64,
+    builds: Vec<BuildId>,
+}
+
+impl Release {
+    pub fn version(&self) -> String {
+        let re = Regex::new(".+-(?<ver>[0-9].+)").unwrap();
+        let caps = re.captures(&self.nix_name).expect("Failed to parse the release name");
+
+        caps["ver"].to_string()
+    }
+
+    pub fn evaluation_id(&self) -> u64 {
+        *self.jobset_evals.first().expect("Failed to obtain the corresponding evaluation, malformed release?")
+    }
+
+    pub fn evaluation_url(&self, hydra_base_uri: &str) -> String {
+        let eval_id = self.evaluation_id();
+        format!("{}/eval/{}", hydra_base_uri, eval_id)
+    }
+
+    pub fn job_url(&self, hydra_base_uri: &str, job_name: &str) -> String {
+        let eval_id = self.evaluation_id();
+        format!("{}/eval/{}/job/{}", hydra_base_uri, eval_id, job_name)
+    }
+
+    pub fn store_paths_url(&self, hydra_base_uri: &str) -> String {
+        let eval_id = self.evaluation_id();
+        format!("{}/eval/{}/store-paths", hydra_base_uri, eval_id)
+    }
+
+    /// Directory related to this release.
+    fn directory(&self, channel: &Channel) -> String {
+        match channel.name.as_str() {
+            "nixpkgs-unstable" => "nixpkgs".to_string(),
+            _ => format!("{}/{}", channel.prefix(), channel.version())
+        }
+    }
+
+    pub fn prefix(&self, channel: &Channel) -> object_store::path::Path {
+        format!("{}/{}", self.directory(channel), self.nix_name).into()
+    }
+}
+
+pub fn release_uri(hydra_uri: &str, job_name: &str) -> String {
+    format!("{}/job/{}/latest", hydra_uri, job_name)
+}
+
+#[derive(Debug)]
+pub struct HydraClient<'a> {
+    pub config: &'a MirrorConfig,
+}
+
+impl HydraClient<'_> {
+    pub async fn fetch_release(&self, job_name: &str) -> reqwest::Result<Release> {
+        let client = reqwest::Client::new();
+        let resp = client.get(release_uri(&self.config.hydra_uri, job_name))
+            .header("Accept", "application/json")
+            // TODO: put a proper version
+            .header("User-Agent", "nixos-channel-scripts (rust)")
+            .send()
+            .await?;
+
+        resp.json().await
+    }
+
+    pub async fn fetch_evaluation(&self, release: &Release) -> reqwest::Result<Evaluation> {
+        let client = reqwest::Client::new();
+        let resp = client.get(release.evaluation_url(&self.config.hydra_uri))
+            .header("Accept", "application/json")
+            .header("User-Agent", "nixos-channel-scripts (rust)")
+            .send()
+            .await?;
+
+        resp.json().await
+    }
+
+    pub async fn fetch_store_paths(&self, release: &Release) -> anyhow::Result<Vec<StorePath<String>>> {
+        let client = reqwest::Client::new();
+
+        Ok(client.get(release.store_paths_url(&self.config.hydra_uri))
+            .header("Accept", "application/json")
+            .header("User-Agent", "nixos-channel-scripts (rust)")
+            .send()
+            .await?
+            .json()
+            .await
+            .context(format!("while downloading store-paths information for an evaluation"))?
+        )
+    }
+
+    pub async fn hydra_product_bytes_stream(&self, release: &Release, job_name: &str, product_type: Option<HydraProductType>) -> anyhow::Result<(u64, impl Stream<Item = std::io::Result<Bytes>>)> {
+        let client = reqwest::Client::new();
+        let build_info: BuildInfo = client.get(release.job_url(&self.config.hydra_uri, job_name))
+            .header("Accept", "application/json")
+            .header("User-Agent", "nixos-channel-scripts (rust)")
+            .send()
+            .await?
+            .json()
+            .await
+            .context(format!("while downloading build information from {}", release.job_url(&self.config.hydra_uri, job_name)))?;
+
+        let mut products_by_subtype: HashMap<String, BuildProduct> = HashMap::new();
+        for (_, product) in build_info.build_products {
+            if products_by_subtype.contains_key(&product.subtype) {
+                todo!("Job {} has multiple products of the same subtype {:?}. This is not supported yet.", job_name, product_type);
+            }
+
+            products_by_subtype.insert(product.subtype.clone(), product);
+        }
+
+        if products_by_subtype.len() > 1 && product_type.is_none() {
+            panic!("Job {} has {} build products. Select the right product via the subtypes.", job_name, products_by_subtype.len());
+        }
+
+        let product: BuildProduct = if let Some(ptype) = product_type {
+            products_by_subtype.remove(&ptype.to_string()).expect(&format!("Expected product type '{}' but not found in the list of products", &ptype.to_string()))
+        } else {
+            products_by_subtype.into_iter().last().expect(&format!("Expected at least one build product in job {}, found zero", job_name)).1
+        };
+
+        // 3. FIXME: validate sha256hash during read?
+        let (store_path, rel_path) = StorePathRef::from_absolute_path_full(&product.store_path).expect("Failed to parse the product's store path");
+        crate::nar::file_in_nar_bytes_stream(&self.config.binary_cache_uri, &nix_compat::nixbase32::encode(store_path.digest()), rel_path.to_str().unwrap()).await.context("while copying the NAR to the target")
+    }
+
+    pub async fn copy_hydra_product<W: AsyncWrite + Unpin>(&self, release: &Release, job_name: &str, product_type: Option<HydraProductType>, pbar: ProgressBar, out: &mut W) -> anyhow::Result<u64> {
+        // TODO: dry me?
+        let client = reqwest::Client::new();
+        let build_info: BuildInfo = client.get(release.job_url(&self.config.hydra_uri, job_name))
+            .header("Accept", "application/json")
+            .header("User-Agent", "nixos-channel-scripts (rust)")
+            .send()
+            .await?
+            .json()
+            .await
+            .context(format!("while downloading build information from {}", release.job_url(&self.config.hydra_uri, job_name)))?;
+
+        let mut products_by_subtype: HashMap<String, BuildProduct> = HashMap::new();
+        for (_, product) in build_info.build_products {
+            if products_by_subtype.contains_key(&product.subtype) {
+                todo!("Job {} has multiple products of the same subtype {:?}. This is not supported yet.", job_name, product_type);
+            }
+
+            products_by_subtype.insert(product.subtype.clone(), product);
+        }
+
+        if products_by_subtype.len() > 1 && product_type.is_none() {
+            panic!("Job {} has {} build products. Select the right product via the subtypes.", job_name, products_by_subtype.len());
+        }
+
+        let product: BuildProduct = if let Some(ptype) = product_type {
+            products_by_subtype.remove(&ptype.to_string()).expect(&format!("Expected product type '{}' but not found in the list of products", &ptype.to_string()))
+        } else {
+            products_by_subtype.into_iter().last().expect(&format!("Expected at least one build product in job {}, found zero", job_name)).1
+        };
+
+        // 3. FIXME: validate sha256hash during read?
+        let (store_path, rel_path) = StorePathRef::from_absolute_path_full(&product.store_path).expect("Failed to parse the product's store path");
+        crate::nar::copy_file_in_nar(&self.config.binary_cache_uri, &nix_compat::nixbase32::encode(store_path.digest()), rel_path.to_str().unwrap(), pbar, out).await.context("while copying the NAR to the target")
+    }
+}

src/main.rs

@@ -0,0 +1,261 @@
+mod config;
+mod actions;
+mod hydra;
+mod git;
+mod nar;
+mod html;
+
+use std::path::PathBuf;
+
+use actions::{interpret_plan, Plan, PlanInterpretation, PlanOptions};
+use clap::{Subcommand, Parser, Args};
+use hydra::{Channel, Evaluation, HydraClient, Release};
+use tracing::{info, trace, warn};
+use object_store::{aws::AmazonS3, ObjectStore};
+use tempdir::TempDir;
+
+use tracing_opentelemetry::OpenTelemetryLayer;
+use tracing_subscriber::layer::SubscriberExt;
+use opentelemetry::trace::TracerProvider;
+
+#[derive(Debug, Args)]
+struct ChannelArgs {
+    /// Channel name to update
+    channel_name: String,
+    /// Job name to fetch from the Hydra instance configured
+    job_name: String,
+}
+
+#[derive(Debug, Args)]
+struct RemoteArgs {
+    /// Channel name to update
+    channel_name: String,
+    /// Staging prefix
+    staging_prefix: String
+}
+
+#[derive(Debug, Args)]
+struct GlobalOpts {
+    /// TOML configuration file for channel updates
+    #[arg(short, long)]
+    config_file: PathBuf,
+    /// Whether to execute no remote side effects (S3 uploads, redirections), etc.
+    #[arg(short, long, default_value_t = false)]
+    dry_run: bool,
+    /// Whether to bypass all preflight checks
+    #[arg(short, long, default_value_t = false)]
+    bypass_preflight_checks: bool,
+    /// Enable or disable progress bar reports
+    #[arg(short, long, default_value_t = true)]
+    enable_progress_bar: bool,
+    /// Stream the plan instead of preparing it in a local directory
+    #[arg(short, long, default_value_t = true)]
+    stream: bool,
+    #[arg(long, default_values = &["sha256", "blake3"])]
+    checksum: Vec<actions::Checksum>
+}
+
+#[derive(Debug, Parser)]
+#[command(version, about, long_about = None)]
+struct App {
+    #[command(flatten)]
+    global_opts: GlobalOpts,
+    #[command(subcommand)]
+    command: Commands
+}
+
+#[derive(Debug, Subcommand)]
+enum Commands {
+    /// Print the plan for the given channel name and job name
+    Plan(ChannelArgs),
+    /// Apply the plan that would be generated for the given channel name and job name
+    Apply(ChannelArgs),
+    /// Apply an existing remote plan that was planned in the past
+    ApplyRemote(RemoteArgs),
+}
+
+#[derive(Debug)]
+struct PreflightCheckContext<'a> {
+    target_channel: &'a Channel,
+    new_release: &'a Release,
+    new_evaluation: &'a Evaluation,
+    current_release_path: Option<object_store::path::Path>,
+}
+
+#[tracing::instrument]
+async fn run_preflight_checks(channel: &Channel, release: &Release, evaluation: &Evaluation, channel_bucket: AmazonS3) -> bool {
+    info!("Running pre-flight checks...");
+    let channel_name = object_store::path::Path::parse(&channel.name).expect("Channel name should be a valid S3 path");
+    let mut context = PreflightCheckContext {
+        target_channel: channel,
+        new_release: release,
+        new_evaluation: evaluation,
+        current_release_path: None
+    };
+    match channel_bucket.get_opts(&channel_name, object_store::GetOptions { head: true, ..Default::default() }).await {
+        Ok(object_store::GetResult { attributes, meta, .. }) => {
+            info!("Release found: {:?}", meta);
+            trace!("Attributes: {:#?}", attributes);
+            if let Some(redirection_target) = attributes.get(&object_store::Attribute::Metadata("x-amz-website-redirect-location".into())) {
+                context.current_release_path = Some(redirection_target.to_string().into());
+            }
+        },
+        Err(err) => {
+            warn!("Error while asking the channel bucket: {}", err);
+            todo!();
+        }
+    }
+
+    trace!("Preflight context check assembled: {:?}", context);
+    // TODO: run anti rollback protection
+    true
+}
+
+enum PlanResult {
+    AlreadyExecuted {
+        remote_path: object_store::path::Path
+    },
+    LocallyPrepared {
+        local_path: TempDir,
+        target_path: object_store::path::Path
+    },
+    StreamedRemotely {
+        remote_path: object_store::path::Path,
+        target_path: object_store::path::Path
+    },
+}
+
+impl From<PlanInterpretation> for PlanResult {
+    fn from(value: PlanInterpretation) -> Self {
+        match value {
+            PlanInterpretation::Streamed { remote_path, target_path } => PlanResult::StreamedRemotely {
+                remote_path,
+                target_path
+            },
+            PlanInterpretation::LocallyPlanned { local_path, target_path } => PlanResult::LocallyPrepared {
+                local_path,
+                target_path,
+            },
+        }
+    }
+}
+
+#[tracing::instrument]
+async fn plan(hydra_client: &HydraClient<'_>, channel_name: &str, job_name: String, global_opts: &GlobalOpts, config: &config::MirrorConfig) -> std::io::Result<PlanResult> {
+    let chan = Channel {
+        name: channel_name.to_string()
+    };
+    info!("Planning for channel {} using job {}", &chan.name, &job_name);
+    let release = hydra_client.fetch_release(&job_name)
+        .await.expect("Failed to fetch release");
+    trace!("{:#?}", release);
+    let evaluation = hydra_client.fetch_evaluation(&release)
+        .await.expect("Failed to fetch evaluation");
+    trace!("{:?}", evaluation.jobset_eval_inputs);
+
+    let plan_options = PlanOptions {
+        streaming: global_opts.stream,
+        enable_progress_bar: global_opts.enable_progress_bar,
+        requested_checksums: global_opts.checksum.clone()
+    };
+
+    if let hydra::Input::Git { revision, .. } = evaluation.jobset_eval_inputs.get("nixpkgs").expect("Expected a nixpkgs repository") {
+        info!("Release information:\n- Release is: {} (build {})\n- Eval is: {}\n- Prefix is: {}\n- Git commit is {}\n", 
+            release.nix_name, release.id, evaluation.id, release.prefix(&chan), revision);
+
+        let release_bucket = config.release_bucket();
+
+        // If the release already exists, skip it.
+        if release_bucket.head(&release.prefix(&chan)).await.is_ok() {
+            tracing::warn!("Release already exists, skipping");
+            return Ok(PlanResult::AlreadyExecuted {
+                remote_path: release.prefix(&chan)
+            });
+        }
+
+        let channel_bucket = config.channel_bucket();
+        if global_opts.bypass_preflight_checks || run_preflight_checks(&chan, &release, &evaluation, channel_bucket).await {
+            let actions = actions::generate_plan(&config, &chan, &release, &evaluation).await;
+            tracing::debug!("Action plan: {:#?}", actions);
+            
+            let interpretation = interpret_plan(&hydra_client, Plan {
+                actions,
+                release,
+                channel: chan.clone(),
+                target_bucket: config.release_bucket(),
+                revision: revision.to_string(),
+            }, plan_options).await;
+            info!("Plan interpreted: {:?}", interpretation);
+
+            Ok(interpretation.into())
+        } else {
+            tracing::error!("Preflight check failed, cannot continue, pass `--force` if you want to bypass preflight checks");
+            Err(std::io::Error::other("preflight check failure"))
+        }
+    } else {
+        panic!("Nixpkgs input is not of type Git");
+    }
+}
+
+
+#[tokio::main]
+async fn main() -> std::io::Result<()> {
+    let args = App::parse();
+
+    let config: config::MirrorConfig = toml::from_str(&std::fs::read_to_string(&args.global_opts.config_file)
+        .expect("Failed to read the configuration file"))
+        .expect("Failed to deserialize the configuration file");
+
+    let hydra_client: HydraClient = HydraClient {
+        config: &config
+    };
+
+    let trace_provider = opentelemetry_otlp::new_pipeline()
+        .tracing()
+        .with_trace_config(opentelemetry_sdk::trace::Config::default().with_resource(opentelemetry_sdk::Resource::new(vec![
+            opentelemetry::KeyValue::new("service.name", env!("CARGO_PKG_NAME")),
+            opentelemetry::KeyValue::new("service.version", env!("CARGO_PKG_VERSION")),
+        ])))
+        .with_exporter(opentelemetry_otlp::new_exporter().tonic())
+        .install_batch(opentelemetry_sdk::runtime::Tokio)
+        .expect("Couldn't create OTLP tracer");
+
+    opentelemetry::global::set_tracer_provider(trace_provider.clone());
+    let fmt_layer = tracing_subscriber::fmt::layer();
+
+    tracing_subscriber::registry()
+        .with(tracing_subscriber::EnvFilter::from_default_env())
+        .with(fmt_layer)
+        .with(OpenTelemetryLayer::new(trace_provider.tracer("channel-scripts")));
+
+    match args.command {
+        Commands::Plan(channel) => {
+            plan(&hydra_client, &channel.channel_name, channel.job_name, &args.global_opts, &config).await?;
+        },
+        Commands::Apply(channel) => {
+            let plan = plan(&hydra_client, &channel.channel_name, channel.job_name, &args.global_opts, &config).await?;
+
+            match plan {
+                PlanResult::AlreadyExecuted { .. } => {
+                    info!("Plan was already executed, no further action required");
+                },
+                PlanResult::LocallyPrepared { .. } => todo!("Uploading a plan is a todo; use streaming!"),
+                PlanResult::StreamedRemotely { remote_path, target_path } => {
+                    info!("Plan was streamed remotely successfully to '{}@{}', replacing {}", config.s3_release_bucket_name, remote_path, target_path);
+
+                    config.release_bucket().rename(&remote_path, &target_path)
+                        .await.expect("Failed to atomically promote the streamed channel, aborting");
+                }
+            }
+        },
+        Commands::ApplyRemote(remote) => {
+            let release = config.release_bucket();
+            // TODO: fetch the release and obtain the right name.
+
+            release.rename(&remote.staging_prefix.try_into().unwrap(), &remote.channel_name.try_into().unwrap())
+                .await.expect("Failed to atomically rename the staging prefix into the channel name");
+        }
+    }
+
+    Ok(())
+}

src/nar.rs

@@ -0,0 +1,117 @@
+use std::borrow::{Borrow, BorrowMut};
+
+use anyhow::{bail, Context};
+use async_compression::tokio::bufread::ZstdDecoder;
+use bytes::Bytes;
+use futures::{Stream, TryStreamExt};
+use indicatif::ProgressBar;
+use tracing::debug;
+use nix_compat::nar::listing::{Listing, ListingEntry};
+use nix_compat::narinfo::NarInfo;
+use tokio::io::{AsyncReadExt, AsyncWrite};
+use tokio_util::io::{ReaderStream, StreamReader};
+
+/// Returns a NarInfo from a HTTP binary cache.
+#[tracing::instrument]
+pub async fn get_narinfo(binary_cache_uri: &str, narinfo_hash: &str) -> reqwest::Result<String> {
+    let client = reqwest::Client::new();
+    let resp = client.get(format!("{}/{}.narinfo", binary_cache_uri, narinfo_hash))
+        .header("User-Agent", "nixos-channel-scripts (rust)")
+        .send()
+        .await?;
+
+    resp.text().await
+}
+
+/// Returns a listing from a HTTP binary cache.
+#[tracing::instrument]
+pub async fn get_listing(binary_cache_uri: &str, narinfo_hash: &str) -> reqwest::Result<Listing> {
+    let client = reqwest::Client::builder().brotli(true).build().unwrap();
+    let resp = client.get(format!("{}/{}.ls", binary_cache_uri, narinfo_hash))
+        .header("User-Agent", "nixos-channel-scripts (rust)")
+        .send()
+        .await?;
+
+    resp.json().await
+}
+
+/// Returns a `Content-Range`-aware HTTP body reader of the NAR.
+pub async fn ranged_nar_reader(binary_cache_uri: &str, narinfo: NarInfo<'_>, start: u64, end: u64) -> reqwest::Result<impl Stream<Item=reqwest::Result<bytes::Bytes>>> {
+    let client = reqwest::Client::new();
+    // TODO: handle compression?
+    let resp = client.get(format!("{}/{}", binary_cache_uri, narinfo.url))
+        .header("User-Agent", "nixos-channel-scripts (rust)")
+        .header("Content-Range", format!("bytes {}-{}/*", start, end))
+        .send()
+        .await?;
+
+    Ok(resp.bytes_stream())
+}
+
+#[tracing::instrument]
+pub async fn nar_reader(binary_cache_uri: &str, narinfo: NarInfo<'_>) -> reqwest::Result<impl Stream<Item=reqwest::Result<bytes::Bytes>>> {
+    let client = reqwest::Client::new();
+    // TODO: handle compression?
+    let resp = client.get(format!("{}/{}", binary_cache_uri, narinfo.url))
+        .header("User-Agent", "nixos-channel-scripts (rust)")
+        .send()
+        .await?;
+
+    Ok(resp.bytes_stream())
+}
+
+/// Streams a binary cache file contained in a NAR
+/// by using narinfo + listing information.
+#[tracing::instrument]
+pub async fn file_in_nar_bytes_stream(binary_cache_uri: &str, narinfo_hash: &str, path: &str) -> anyhow::Result<(u64, impl Stream<Item = std::io::Result<Bytes>>)> {
+    let narinfo_str: String = get_narinfo(binary_cache_uri, narinfo_hash).await?;
+    let narinfo: NarInfo = NarInfo::parse(narinfo_str.as_ref()).context("while parsing the narinfo").expect("Failed to parse narinfo from HTTP binary cache server");
+    let listing: Listing = get_listing(binary_cache_uri, narinfo_hash).await.context("while parsing the listing of that NAR")?;
+    match listing {
+        Listing::V1 { root, .. } => {
+            let entry = root.locate(path)
+                .expect("Invalid relative path to the NAR")
+                .ok_or(tokio::io::Error::new(tokio::io::ErrorKind::NotFound, format!("Entry {} not found in the NAR listing", path)))?;
+
+            if let ListingEntry::Regular {
+                size, nar_offset, ..
+            } = entry {
+                // TODO: this is ugly.
+                let mut file_reader = ZstdDecoder::new(
+                    StreamReader::new(nar_reader(binary_cache_uri, narinfo).await?.map_err(std::io::Error::other))
+                );
+
+                // Discard *nar_offset bytes
+                let discarded = tokio::io::copy(&mut file_reader.borrow_mut().take(*nar_offset), &mut tokio::io::sink()).await
+                    .context("while discarding the start of the NAR").unwrap();
+
+                debug!("discarded {} bytes and taking at most {} bytes", discarded, *size);
+
+                // Let the consumer copy at most *size bytes of data.
+                Ok((*size, ReaderStream::new(file_reader.take(*size))))
+            } else {
+                bail!("Expected a file, obtained either a symlink or a directory");
+            }
+
+        },
+        _ => bail!("Unsupported listing version")
+    }
+}
+
+/// Hits the binary cache with a narinfo request
+/// and copies a specific file to the provided writer
+/// asynchronously.
+pub async fn copy_file_in_nar<W: AsyncWrite + Unpin>(binary_cache_uri: &str, narinfo_hash: &str, path: &str, pbar: ProgressBar, out: &mut W) -> anyhow::Result<u64> {
+    let (size, stream) = file_in_nar_bytes_stream(binary_cache_uri, narinfo_hash, path).await?;
+    pbar.set_length(size);
+
+    let copied = tokio::io::copy(
+        &mut StreamReader::new(stream),
+        &mut pbar.wrap_async_write(out)
+    ).await.context("while copying the file in the NAR").unwrap();
+
+    assert!(copied == size, "mismatch in the copy sizes");
+    pbar.finish();
+
+    return Ok(copied)
+}

src/release.html

@@ -0,0 +1,72 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>{{channelName}} - NixOS Channel</title>
+    <style>
+        body {
+            font-family: Arial, sans-serif;
+            margin: 20px;
+        }
+        .header {
+            margin-bottom: 20px;
+        }
+        .header h1 {
+            margin: 0;
+            font-size: 1.5em;
+        }
+        .header p {
+            margin: 5px 0;
+            color: #666;
+        }
+        .links a {
+            color: #0066cc;
+            text-decoration: none;
+        }
+        .file-list {
+            width: 100%;
+            border-collapse: collapse;
+            margin-top: 20px;
+        }
+        .file-list th, .file-list td {
+            border: 1px solid #ddd;
+            padding: 8px;
+            text-align: left;
+        }
+        .file-list th {
+            background-color: #f2f2f2;
+        }
+    </style>
+</head>
+<body>
+    <div class="header">
+        <h1>Channel: {{channelName}}</h1>
+        <p>Release: {{releaseName}}</p>
+        <p>Date of Release: {{releaseDate}}</p>
+        <div class="links">
+            <a href="{{gitCommitLink}}">Git Commit</a> |
+            <a href="{{hydraEvalLink}}">Hydra Evaluation</a>
+        </div>
+    </div>
+
+    <table class="file-list">
+        <thead>
+            <tr>
+                <th>File Name</th>
+                <th>Size</th>
+                <th>SHA-256 Hash</th>
+            </tr>
+        </thead>
+        <tbody>
+            {{#each files}}
+            <tr>
+                <td>{{name}}</td>
+                <td>{{size}}</td>
+                <td>{{sha256}}</td>
+            </tr>
+            {{/each}}
+        </tbody>
+    </table>
+</body>
+</html>

src/release.tpl

@@ -0,0 +1,78 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>{{channelName}} - NixOS Channel</title>
+    <style>
+        body {
+            font-family: Arial, sans-serif;
+            margin: 20px;
+        }
+        .header {
+            margin-bottom: 20px;
+        }
+        .header h1 {
+            margin: 0;
+            font-size: 1.5em;
+        }
+        .header p {
+            margin: 5px 0;
+            color: #666;
+        }
+        .links a {
+            color: #0066cc;
+            text-decoration: none;
+        }
+        .file-list {
+            width: 100%;
+            border-collapse: collapse;
+            margin-top: 20px;
+        }
+        .file-list th, .file-list td {
+            border: 1px solid #ddd;
+            padding: 8px;
+            text-align: left;
+        }
+        .file-list th {
+            background-color: #f2f2f2;
+        }
+    </style>
+</head>
+<body>
+    <div class="header">
+        <h1>Channel: {{channelName}}</h1>
+        <p>Release: {{releaseName}}</p>
+        <p>Date of Release: {{releaseDate}}</p>
+        <div class="links">
+            <a href="{{gitCommitLink}}">Git Commit</a> |
+            <a href="{{hydraEvalLink}}">Hydra Evaluation</a>
+        </div>
+    </div>
+
+    <table class="file-list">
+        <thead>
+            <tr>
+                <th>File Name</th>
+                <th>Size</th>
+                {{#if checksums}}
+                    {{#each checksums as |checksum|}}
+                        <th>{{checksum}} hash</th>
+                    {{/each}}
+                {{/if}}
+            </tr>
+        </thead>
+        <tbody>
+            {{#each files}}
+            <tr>
+                <td><a href="./{{name}}">{{name}}</a></td>
+                <td>{{size}}</td>
+                {{#each checksums as |checksum|}}
+                    <td>{{this}}</td>
+                {{/each}}
+            </tr>
+            {{/each}}
+        </tbody>
+    </table>
+</body>
+</html>