about summary refs log tree commit diff
path: root/tvix/castore/src/import/fs.rs
diff options
context:
space:
mode:
authorFlorian Klink <flokli@flokli.de>2024-04-20T16·32+0300
committerclbot <clbot@tvl.fyi>2024-04-20T18·54+0000
commit5fc403587ffb38635e767697f1c18e7e3559aea6 (patch)
tree8779c1714af0145ca580de482a41e63821cd59f2 /tvix/castore/src/import/fs.rs
parent01239a4f6f871733231c01d6126c3ffedcc504b7 (diff)
refactor(tvix/castore): ingest filesystem entries in parallel r/7987
Rather than carrying around an Future in the IngestionEntry::Regular,
simply carry the plain B3Digest.

Code reading through a non-seekable data stream has no choice but to
read and upload blobs immediately, and code seeking through something
seekable (like a filesystem) probably knows better what concurrency to
pick when ingesting, rather than the consuming side.

(Our only) one of these seekable source implementations is now doing
exactly that. We produce a stream of futures, and then use
[StreamExt::buffered] to process more than one, concurrently.

We still keep the same order, to avoid shuffling things and violating
the stream order.

This also cleans up walk_path_for_ingestion in castore/import, as well
as ingest_dir_entries in glue/tvix_store_io.

Change-Id: I5eb70f3e1e372c74bcbfcf6b6e2653eba36e151d
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11491
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Tested-by: BuildkiteCI
Diffstat (limited to 'tvix/castore/src/import/fs.rs')
-rw-r--r--tvix/castore/src/import/fs.rs79
1 files changed, 34 insertions, 45 deletions
diff --git a/tvix/castore/src/import/fs.rs b/tvix/castore/src/import/fs.rs
index a87a0278da4d..6709d4a127e9 100644
--- a/tvix/castore/src/import/fs.rs
+++ b/tvix/castore/src/import/fs.rs
@@ -1,8 +1,8 @@
+use futures::stream::BoxStream;
+use futures::StreamExt;
 use std::os::unix::fs::MetadataExt;
 use std::os::unix::fs::PermissionsExt;
 use std::path::Path;
-
-use futures::stream::BoxStream;
 use tracing::instrument;
 use walkdir::DirEntry;
 use walkdir::WalkDir;
@@ -22,6 +22,9 @@ use super::IngestionEntry;
 /// [DirectoryService]. It returns the root node or an error.
 ///
 /// It does not follow symlinks at the root, they will be ingested as actual symlinks.
+///
+/// This function will walk the filesystem using `walkdir` and will consume
+/// `O(#number of entries)` space.
 #[instrument(skip(blob_service, directory_service), fields(path), err)]
 pub async fn ingest_path<BS, DS, P>(
     blob_service: BS,
@@ -33,61 +36,51 @@ where
     BS: BlobService + Clone,
     DS: AsRef<dyn DirectoryService>,
 {
-    let entry_stream = walk_path_for_ingestion(blob_service, path.as_ref());
-    ingest_entries(directory_service, entry_stream).await
-}
-
-/// Walk the filesystem at a given path and returns a stream of ingestion entries.
-///
-/// This is how [`ingest_path`] assembles the set of entries to pass on [`ingest_entries`].
-/// This low-level function can be used if additional filtering or processing is required on the
-/// entries.
-///
-/// It does not follow symlinks at the root, they will be ingested as actual symlinks.
-///
-/// This function will walk the filesystem using `walkdir` and will consume
-/// `O(#number of entries)` space.
-#[instrument(fields(path), skip(blob_service))]
-fn walk_path_for_ingestion<'a, BS>(
-    blob_service: BS,
-    path: &'a Path,
-) -> BoxStream<'a, Result<IngestionEntry<'a>, Error>>
-where
-    BS: BlobService + Clone + 'a,
-{
-    let iter = WalkDir::new(path)
+    let iter = WalkDir::new(path.as_ref())
         .follow_links(false)
         .follow_root_links(false)
         .contents_first(true)
         .into_iter();
 
-    dir_entry_iter_to_ingestion_stream(blob_service, iter, path)
+    let entries = dir_entries_to_ingestion_stream(blob_service, iter, path.as_ref());
+    ingest_entries(directory_service, entries).await
 }
 
 /// Converts an iterator of [walkdir::DirEntry]s into a stream of ingestion entries.
 /// This can then be fed into [ingest_entries] to ingest all the entries into the castore.
 ///
+/// The produced stream is buffered, so uploads can happen concurrently.
+///
 /// The root is the [Path] in the filesystem that is being ingested into the castore.
-pub fn dir_entry_iter_to_ingestion_stream<'a, BS, I>(
+pub fn dir_entries_to_ingestion_stream<'a, BS, I>(
     blob_service: BS,
     iter: I,
     root: &'a Path,
-) -> BoxStream<'a, Result<IngestionEntry<'a>, Error>>
+) -> BoxStream<'a, Result<IngestionEntry, Error>>
 where
     BS: BlobService + Clone + 'a,
     I: Iterator<Item = Result<DirEntry, walkdir::Error>> + Send + 'a,
 {
     let prefix = root.parent().unwrap_or_else(|| Path::new(""));
 
-    let iter = iter.map(move |entry| match entry {
-        Ok(entry) => dir_entry_to_ingestion_entry(blob_service.clone(), &entry, prefix),
-        Err(error) => Err(Error::UnableToStat(
-            root.to_path_buf(),
-            error.into_io_error().expect("walkdir err must be some"),
-        )),
-    });
-
-    Box::pin(futures::stream::iter(iter))
+    Box::pin(
+        futures::stream::iter(iter)
+            .map(move |x| {
+                let blob_service = blob_service.clone();
+                async move {
+                    match x {
+                        Ok(dir_entry) => {
+                            dir_entry_to_ingestion_entry(blob_service, &dir_entry, prefix).await
+                        }
+                        Err(e) => Err(Error::UnableToStat(
+                            prefix.to_path_buf(),
+                            e.into_io_error().expect("walkdir err must be some"),
+                        )),
+                    }
+                }
+            })
+            .buffered(50),
+    )
 }
 
 /// Converts a [walkdir::DirEntry] into an [IngestionEntry], uploading blobs to the
@@ -95,13 +88,13 @@ where
 ///
 /// The prefix path is stripped from the path of each entry. This is usually the parent path
 /// of the path being ingested so that the last element of the stream only has one component.
-fn dir_entry_to_ingestion_entry<'a, BS>(
+pub async fn dir_entry_to_ingestion_entry<BS>(
     blob_service: BS,
     entry: &DirEntry,
     prefix: &Path,
-) -> Result<IngestionEntry<'a>, Error>
+) -> Result<IngestionEntry, Error>
 where
-    BS: BlobService + 'a,
+    BS: BlobService,
 {
     let file_type = entry.file_type();
 
@@ -123,11 +116,7 @@ where
             .metadata()
             .map_err(|e| Error::UnableToStat(entry.path().to_path_buf(), e.into()))?;
 
-        // TODO: In the future, for small files, hash right away and upload async.
-        let digest = Box::pin(upload_blob_at_path(
-            blob_service,
-            entry.path().to_path_buf(),
-        ));
+        let digest = upload_blob_at_path(blob_service, entry.path().to_path_buf()).await?;
 
         Ok(IngestionEntry::Regular {
             path,