Pi deploy follow-ups: XDG default_path + parallel restore #28

Merged
osobh merged 1 commits from pi-followups into main 2026-07-12 14:43:12 +00:00
4 changed files with 259 additions and 6 deletions
+32 -5
View File
@@ -46,8 +46,9 @@ use crate::cluster::build_cache::{
};
use crate::cluster::client_config::{ClientConfig, ResolvedClientConfig};
use crate::cluster::rpc::{
call_blob_get_stream, call_blob_put_stream, call_blob_stat, call_delete_tag, call_get_metrics,
call_peer_status, prewarm_missing_chunks_between_parallel,
call_blob_get_parallel, call_blob_get_stream, call_blob_put_stream, call_blob_stat,
call_delete_tag, call_get_metrics, call_peer_status,
prewarm_missing_chunks_between_parallel,
call_get_ref, call_get_tag, call_list_tags, call_put_ref, call_put_tag,
};
use crate::cluster::transport::{NodeIdentity, QuicClient};
@@ -219,6 +220,13 @@ struct BuildArgs {
/// Skip capture + upload on a miss. Useful for read-only cache use.
#[arg(long)]
no_upload: bool,
/// Field finding 2026-07-12: parallel chunk fetch on cache HIT.
/// Sequential BlobGetStream restore of 947 MiB on Pi 5 took 18s
/// (~53 MiB/s); parallel chunk fetches stack their per-stream
/// throughputs. Set to 1 to force sequential (matches pre-fix
/// behavior).
#[arg(long, default_value_t = 8)]
parallel_restore: usize,
/// Extra args passed verbatim to `cargo build` (after `--`).
#[arg(last = true)]
cargo_args: Vec<String>,
@@ -558,10 +566,29 @@ async fn cmd_build(args: BuildArgs) -> Result<()> {
let mut outcome = CacheOutcome::Miss;
match peer_lookup(&conn, &fp).await? {
Some((blob_id, stat)) => {
tracing::info!("cache HIT — blob {} ({} bytes)", blob_id, stat.total_size);
let mut buf = Vec::with_capacity(stat.total_size as usize);
let ok = call_blob_get_stream(&conn, &blob_id, &mut buf).await?;
tracing::info!(
"cache HIT — blob {} ({} bytes, parallel={})",
blob_id,
stat.total_size,
args.parallel_restore
);
// Field finding 2026-07-12: single-stream BlobGetStream
// capped Pi restore at ~53 MiB/s. Parallel chunk fetches
// let per-stream throughputs stack. `parallel_restore <= 1`
// falls back to the sequential BlobGetStream path (kept
// for diagnostic parity).
let buf = if args.parallel_restore <= 1 {
let mut b = Vec::with_capacity(stat.total_size as usize);
let ok = call_blob_get_stream(&conn, &blob_id, &mut b).await?;
if ok {
Some(b)
} else {
None
}
} else {
call_blob_get_parallel(&conn, &blob_id, args.parallel_restore).await?
};
if let Some(buf) = buf {
std::fs::create_dir_all(&target_dir).with_context(|| {
format!("creating {}", target_dir.display())
})?;
+89
View File
@@ -677,6 +677,95 @@ pub async fn prewarm_missing_chunks_between(
Ok((uploaded, total))
}
/// Field finding 2026-07-12 (Pi restore = 18s single-stream): fetch a
/// blob by pulling its chunks in parallel and reassembling in memory.
/// Faster than `call_blob_get_stream` on connections with per-stream
/// throughput ceilings — parallel streams stack their contributions.
///
/// Flow:
/// 1. `LoadManifest` upstream (small).
/// 2. Spawn N concurrent `GetChunk` tasks bounded by a semaphore.
/// 3. Assemble the results in manifest order into a `Vec<u8>` sized
/// to `manifest.total_size`.
///
/// `concurrency <= 1` degrades to sequential (matches
/// `call_blob_get_stream` semantics but keeps the code path uniform).
/// Memory ceiling: `total_size + 4 MiB × in-flight` — dominated by
/// the reassembled blob buffer itself.
pub async fn call_blob_get_parallel(
conn: &Connection,
id: &BlobId,
concurrency: usize,
) -> Result<Option<Vec<u8>>> {
let manifest = match call_blob_load_manifest(conn, id).await? {
Some(m) => m,
None => return Ok(None),
};
let concurrency = concurrency.max(1);
let sem = std::sync::Arc::new(tokio::sync::Semaphore::new(concurrency));
let mut set = tokio::task::JoinSet::new();
for (idx, hash) in manifest.chunks.iter().copied().enumerate() {
let permit = sem
.clone()
.acquire_owned()
.await
.context("acquiring get_parallel semaphore permit")?;
let conn = conn.clone();
set.spawn(async move {
let _permit = permit;
let bytes = call_get_chunk(&conn, &hash).await?.with_context(|| {
format!(
"manifest referenced chunk {} but GetChunk returned NotFound",
hash.to_hex()
)
})?;
Ok::<(usize, Vec<u8>), anyhow::Error>((idx, bytes))
});
}
// Assemble in manifest order. Pre-size the outer vec so we can
// slot each chunk's bytes at the right offset without copying.
let mut out = vec![0u8; manifest.total_size as usize];
// Chunk boundaries: chunk i starts at i * CHUNK_SIZE.
let chunk_size = crate::cluster::blob::CHUNK_SIZE;
let mut first_err: Option<anyhow::Error> = None;
while let Some(join) = set.join_next().await {
match join {
Ok(Ok((idx, bytes))) => {
let start = idx * chunk_size;
let end = start + bytes.len();
if end > out.len() {
if first_err.is_none() {
first_err = Some(anyhow::anyhow!(
"chunk {} at idx {} would overflow reassembly buffer \
(end {}, total_size {})",
manifest.chunks[idx].to_hex(),
idx,
end,
manifest.total_size
));
}
continue;
}
out[start..end].copy_from_slice(&bytes);
}
Ok(Err(e)) => {
if first_err.is_none() {
first_err = Some(e);
}
}
Err(join_err) => {
if first_err.is_none() {
first_err = Some(anyhow::Error::from(join_err));
}
}
}
}
if let Some(e) = first_err {
return Err(e).context("parallel chunk fetch failed");
}
Ok(Some(out))
}
/// Phase 5k: parallel-fanout variant of [`prewarm_missing_chunks_between`].
///
/// Runs the has→get→put pipeline for each chunk concurrently, bounded
@@ -1204,3 +1204,72 @@ async fn end_to_end_parallel_prewarm_copies_chunks_and_matches_sequential() {
acc_a.abort();
acc_c.abort();
}
#[tokio::test]
async fn parallel_blob_get_reassembles_multi_chunk_blob_byte_equal() {
// Field finding 2026-07-12: parallel chunk fetch on restore.
// Verifies (1) reassembly byte-equals a sequential BlobGetStream,
// (2) tail chunks (not full CHUNK_SIZE) land at the right offset,
// (3) NotFound path returns None.
use crate::cluster::blob::CHUNK_SIZE;
let (id_a, id_b) = NodeIdentity::generate_test_pair("a", "b").unwrap();
let (_tmp_a, router_a) = router_with_full_stack("a", next_port()).await;
let payload: Vec<u8> = (0..(3 * CHUNK_SIZE + CHUNK_SIZE / 5))
.map(|i| ((i * 13) % 251) as u8)
.collect();
let blob_id = router_a
.blob_store()
.unwrap()
.put_bytes(&payload)
.await
.unwrap();
let server_a = QuicServer::bind(loopback(0), id_a).unwrap();
let addr_a = server_a.local_addr().unwrap();
let ra = router_a.clone();
let acc_a = tokio::spawn(async move {
while let Some(Ok(conn)) = server_a.accept().await {
let r = ra.clone();
tokio::spawn(async move {
let _ = serve_connection(conn, r).await;
});
}
});
let client = QuicClient::new(loopback(0), id_b).unwrap();
let conn = client.connect(addr_a, "a").await.unwrap();
// Sequential reference: BlobGetStream.
let mut seq = Vec::new();
let ok = call_blob_get_stream(&conn, &blob_id, &mut seq).await.unwrap();
assert!(ok);
assert_eq!(seq, payload, "sequential fetch must be byte-equal to source");
// Parallel with concurrency = 4.
let par = call_blob_get_parallel(&conn, &blob_id, 4)
.await
.unwrap()
.unwrap();
assert_eq!(par, payload, "parallel fetch must match sequential");
assert_eq!(par, seq, "parallel and sequential must agree");
// concurrency = 1 falls through to still-parallel (with just one
// in-flight) but must still be correct.
let ser_via_par = call_blob_get_parallel(&conn, &blob_id, 1)
.await
.unwrap()
.unwrap();
assert_eq!(ser_via_par, payload);
// Unknown blob → None.
let missing = crate::cluster::blob::BlobId::from_bytes([0u8; 32]);
let none = call_blob_get_parallel(&conn, &missing, 4).await.unwrap();
assert!(none.is_none(), "NotFound must surface as None");
conn.close(quinn::VarInt::from_u32(0), b"done");
client.shutdown().await;
tokio::time::sleep(Duration::from_millis(50)).await;
acc_a.abort();
}
+68
View File
@@ -127,7 +127,31 @@ impl Manifest {
}
}
/// Field finding 2026-07-12 (Pi deploy on vision-02): the old
/// hardcoded `/var/lib/claw-store/projects.toml` broke under
/// `ProtectSystem=strict` in the user-mode systemd unit because
/// /var is read-only. Follow the XDG Base Directory spec so a
/// user-mode install writes under `$HOME/.local/state`, and only
/// root installs land in `/var/lib`.
///
/// Precedence:
/// 1. `$XDG_STATE_HOME/claw-store/projects.toml` (per spec)
/// 2. `$HOME/.local/state/claw-store/projects.toml` (XDG default)
/// 3. `/var/lib/claw-store/projects.toml` (system fallback)
pub fn default_path() -> PathBuf {
if let Ok(xdg) = std::env::var("XDG_STATE_HOME") {
if !xdg.is_empty() {
return PathBuf::from(xdg)
.join("claw-store")
.join("projects.toml");
}
}
if let Ok(home) = std::env::var("HOME") {
if !home.is_empty() {
return PathBuf::from(home)
.join(".local/state/claw-store/projects.toml");
}
}
PathBuf::from("/var/lib/claw-store/projects.toml")
}
}
@@ -207,6 +231,50 @@ mod tests {
use super::*;
use tempfile::NamedTempFile;
#[test]
fn default_path_honours_xdg_state_home() {
// Field finding 2026-07-12: precedence XDG_STATE_HOME →
// $HOME/.local/state → /var/lib. Guard: an env-set XDG wins
// over HOME; empty XDG is treated as unset.
// NOTE: env mutation is process-global, so this test does its
// own setup/teardown and doesn't run in parallel with another
// that touches the same vars.
let saved_xdg = std::env::var("XDG_STATE_HOME").ok();
let saved_home = std::env::var("HOME").ok();
// Case 1: XDG_STATE_HOME wins.
std::env::set_var("XDG_STATE_HOME", "/tmp/xdg-fake");
std::env::set_var("HOME", "/tmp/home-fake");
assert_eq!(
Manifest::default_path(),
PathBuf::from("/tmp/xdg-fake/claw-store/projects.toml")
);
// Case 2: XDG unset → HOME/.local/state.
std::env::remove_var("XDG_STATE_HOME");
assert_eq!(
Manifest::default_path(),
PathBuf::from("/tmp/home-fake/.local/state/claw-store/projects.toml")
);
// Case 3: empty XDG treated as unset.
std::env::set_var("XDG_STATE_HOME", "");
assert_eq!(
Manifest::default_path(),
PathBuf::from("/tmp/home-fake/.local/state/claw-store/projects.toml")
);
// Restore.
match saved_xdg {
Some(v) => std::env::set_var("XDG_STATE_HOME", v),
None => std::env::remove_var("XDG_STATE_HOME"),
}
match saved_home {
Some(v) => std::env::set_var("HOME", v),
None => std::env::remove_var("HOME"),
}
}
#[test]
fn test_roundtrip_manifest() {
let mut m = Manifest::default();