nydusd: add the config support of amplify_io

Add the support of `amplify_io` in the config file of nydusd
to configure read amplification.

Signed-off-by: Wenhao Ren <[email protected]>

api/src/config.rs

@@ -819,8 +819,8 @@ pub struct RafsConfigV2 {
     /// Filesystem metadata cache mode.
     #[serde(default = "default_rafs_mode")]
     pub mode: String,
-    /// Batch size to read data from storage cache layer.
-    #[serde(rename = "batch_size", default = "default_batch_size")]
+    /// Amplified user IO request batch size to read data from remote storage backend / local cache.
+    #[serde(rename = "batch_size", default = "default_user_io_batch_size")]
     pub user_io_batch_size: usize,
     /// Whether to validate data digest.
     #[serde(default)]
@@ -874,8 +874,8 @@ pub struct PrefetchConfigV2 {
     /// Number of data prefetching working threads.
     #[serde(rename = "threads", default = "default_prefetch_threads_count")]
     pub threads_count: usize,
-    /// The batch size to prefetch data from backend.
-    #[serde(rename = "batch_size", default = "default_prefetch_batch_size")]
+    /// The amplify batch size to prefetch data from backend.
+    #[serde(rename = "batch_size", default = "default_prefetch_request_batch_size")]
     pub prefetch_request_batch_size: usize,
     /// Network bandwidth rate limit in unit of Bytes and Zero means no limit.
     #[serde(default)]
@@ -1194,11 +1194,11 @@ fn default_work_dir() -> String {
     ".".to_string()
 }
 
-pub fn default_batch_size() -> usize {
-    128 * 1024
+pub fn default_user_io_batch_size() -> usize {
+    1024 * 1024
 }
 
-fn default_prefetch_batch_size() -> usize {
+pub fn default_prefetch_request_batch_size() -> usize {
     1024 * 1024
 }
 
@@ -1363,8 +1363,9 @@ struct RafsConfig {
     /// Record file name if file access trace log.
     #[serde(default)]
     pub latest_read_files: bool,
+    // Amplified user IO request batch size to read data from remote storage backend / local cache.
     // ZERO value means, amplifying user io is not enabled.
-    #[serde(rename = "amplify_io", default = "default_batch_size")]
+    #[serde(rename = "amplify_io", default = "default_user_io_batch_size")]
     pub user_io_batch_size: usize,
 }
 
@@ -1410,8 +1411,11 @@ struct FsPrefetchControl {
     #[serde(default = "default_prefetch_threads_count")]
     pub threads_count: usize,
 
-    /// Window size in unit of bytes to merge request to backend.
-    #[serde(rename = "merging_size", default = "default_batch_size")]
+    /// The amplify batch size to prefetch data from backend.
+    #[serde(
+        rename = "merging_size",
+        default = "default_prefetch_request_batch_size"
+    )]
     pub prefetch_request_batch_size: usize,
 
     /// Network bandwidth limitation for prefetching.
@@ -1449,7 +1453,7 @@ struct BlobPrefetchConfig {
     pub enable: bool,
     /// Number of data prefetching working threads.
     pub threads_count: usize,
-    /// The maximum size of a merged IO request.
+    /// The amplify batch size to prefetch data from backend.
     #[serde(rename = "merging_size")]
     pub prefetch_request_batch_size: usize,
     /// Network bandwidth rate limit in unit of Bytes and Zero means no limit.

docs/nydusd.md

@@ -130,6 +130,9 @@ We are working on enabling cloud-hypervisor support for nydus.
   "iostats_files": true,
   // Enable support of fs extended attributes
   "enable_xattr": false,
+  // Amplified user IO request batch size to read data from remote storage backend / local cache
+  // in unit of Bytes, valid values: 0-268435456, default: 1048576
+  "amplify_io": 1048576,
   "fs_prefetch": {
     // Enable blob prefetch
     "enable": false,

misc/configs/nydusd-config-v2.toml

@@ -142,7 +142,8 @@ bandwidth_limit = 10000000
 [rafs]
 # Filesystem metadata cache mode, "direct" or "cached". "direct" is almost what you want.
 mode = "direct"
-# Batch size to read data from storage cache layer, valid values: 0-0x10000000
+# Amplified user IO request batch size to read data from remote storage backend / local cache,
+# valid values: 0-0x10000000
 batch_size = 1000000
 # Whether to validate data digest.
 validate = true

service/src/fs_cache.rs

@@ -522,8 +522,8 @@ impl FsCacheHandler {
             .prefetch_request_batch_size
             .checked_next_power_of_two()
         {
-            None => nydus_api::default_batch_size() as u64,
-            Some(1) => nydus_api::default_batch_size() as u64,
+            None => nydus_api::default_prefetch_request_batch_size() as u64,
+            Some(1) => nydus_api::default_prefetch_request_batch_size() as u64,
             Some(s) => s as u64,
         };
         let size = std::cmp::max(0x4_0000u64, size);

storage/src/cache/cachedfile.rs

@@ -164,7 +164,8 @@ pub(crate) struct FileCacheEntry {
     pub(crate) dio_enabled: bool,
     // Data from the file cache should be validated before use.
     pub(crate) need_validation: bool,
-    pub(crate) batch_size: u64,
+    // Amplified user IO request batch size to read data from remote storage backend / local cache.
+    pub(crate) user_io_batch_size: u32,
     pub(crate) prefetch_config: Arc<AsyncPrefetchConfig>,
 }
 
@@ -300,19 +301,19 @@ impl FileCacheEntry {
         }
     }
 
-    fn prefetch_batch_size(&self) -> u64 {
+    fn prefetch_request_batch_size(&self) -> u64 {
         if self.prefetch_config.prefetch_request_batch_size < 0x2_0000 {
             0x2_0000
         } else {
             self.prefetch_config.prefetch_request_batch_size as u64
         }
     }
 
-    fn ondemand_batch_size(&self) -> u64 {
-        if self.batch_size < 0x2_0000 {
+    fn user_io_batch_size(&self) -> u64 {
+        if self.user_io_batch_size < 0x2_0000 {
             0x2_0000
         } else {
-            self.batch_size
+            self.user_io_batch_size as u64
         }
     }
 
@@ -559,7 +560,7 @@ impl BlobCache for FileCacheEntry {
         }
 
         // Then handle fs prefetch
-        let max_comp_size = self.prefetch_batch_size();
+        let max_comp_size = self.prefetch_request_batch_size();
         let mut bios = bios.to_vec();
         bios.sort_unstable_by_key(|entry| entry.chunkinfo.compressed_offset());
         self.metrics.prefetch_unmerged_chunks.add(bios.len() as u64);
@@ -719,7 +720,7 @@ impl BlobObject for FileCacheEntry {
         let meta = self.meta.as_ref().ok_or_else(|| enoent!())?;
         let meta = meta.get_blob_meta().ok_or_else(|| einval!())?;
         let mut chunks =
-            meta.get_chunks_compressed(offset, size, self.prefetch_batch_size(), prefetch)?;
+            meta.get_chunks_compressed(offset, size, self.prefetch_request_batch_size(), prefetch)?;
         if !chunks.is_empty() {
             if let Some(meta) = self.get_blob_meta_info()? {
                 chunks = self.strip_ready_chunks(meta, None, chunks);
@@ -745,7 +746,7 @@ impl BlobObject for FileCacheEntry {
 
         let meta = self.meta.as_ref().ok_or_else(|| einval!())?;
         let meta = meta.get_blob_meta().ok_or_else(|| einval!())?;
-        let mut chunks = meta.get_chunks_uncompressed(offset, size, self.ondemand_batch_size())?;
+        let mut chunks = meta.get_chunks_uncompressed(offset, size, self.user_io_batch_size())?;
         if let Some(meta) = self.get_blob_meta_info()? {
             chunks = self.strip_ready_chunks(meta, None, chunks);
         }
@@ -764,7 +765,7 @@ impl BlobObject for FileCacheEntry {
 
         let chunks_extended;
         let mut chunks = &range.chunks;
-        if let Some(v) = self.extend_pending_chunks(chunks, self.prefetch_batch_size())? {
+        if let Some(v) = self.extend_pending_chunks(chunks, self.prefetch_request_batch_size())? {
             chunks_extended = v;
             chunks = &chunks_extended;
         }
@@ -934,7 +935,7 @@ impl FileCacheEntry {
     fn read_iter(&self, bios: &mut [BlobIoDesc], buffers: &[FileVolatileSlice]) -> Result<usize> {
         // Merge requests with continuous blob addresses.
         let requests = self
-            .merge_requests_for_user(bios, self.ondemand_batch_size())
+            .merge_requests_for_user(bios, self.user_io_batch_size())
             .ok_or_else(|| {
                 for bio in bios.iter() {
                     self.update_chunk_pending_status(&bio.chunkinfo, false);
@@ -1100,14 +1101,14 @@ impl FileCacheEntry {
                     + region.chunks[idx].compressed_size() as u64;
                 let start = region.chunks[idx + 1].compressed_offset();
                 assert!(end <= start);
-                assert!(start - end <= self.ondemand_batch_size() >> RAFS_BATCH_SIZE_TO_GAP_SHIFT);
+                assert!(start - end <= self.user_io_batch_size() >> RAFS_BATCH_SIZE_TO_GAP_SHIFT);
                 assert!(region.chunks[idx].id() < region.chunks[idx + 1].id());
             }
         }
 
         // Try to extend requests.
         let mut region_hold;
-        if let Some(v) = self.extend_pending_chunks(&region.chunks, self.ondemand_batch_size())? {
+        if let Some(v) = self.extend_pending_chunks(&region.chunks, self.user_io_batch_size())? {
             if v.len() > r.chunks.len() {
                 let mut tag_set = HashSet::new();
                 for (idx, chunk) in region.chunks.iter().enumerate() {

storage/src/cache/filecache/mod.rs

@@ -23,7 +23,6 @@ use crate::cache::state::{
 use crate::cache::worker::{AsyncPrefetchConfig, AsyncWorkerMgr};
 use crate::cache::{BlobCache, BlobCacheMgr};
 use crate::device::{BlobFeatures, BlobInfo};
-use crate::RAFS_DEFAULT_CHUNK_SIZE;
 
 pub const BLOB_RAW_FILE_SUFFIX: &str = ".blob.raw";
 pub const BLOB_DATA_FILE_SUFFIX: &str = ".blob.data";
@@ -46,6 +45,7 @@ pub struct FileCacheMgr {
     cache_convergent_encryption: bool,
     cache_encryption_key: String,
     closed: Arc<AtomicBool>,
+    user_io_batch_size: u32,
 }
 
 impl FileCacheMgr {
@@ -55,6 +55,7 @@ impl FileCacheMgr {
         backend: Arc<dyn BlobBackend>,
         runtime: Arc<Runtime>,
         id: &str,
+        user_io_batch_size: u32,
     ) -> Result<FileCacheMgr> {
         let blob_cfg = config.get_filecache_config()?;
         let work_dir = blob_cfg.get_work_dir()?;
@@ -77,6 +78,7 @@ impl FileCacheMgr {
             cache_convergent_encryption: blob_cfg.enable_convergent_encryption,
             cache_encryption_key: blob_cfg.encryption_key.clone(),
             closed: Arc::new(AtomicBool::new(false)),
+            user_io_batch_size,
         })
     }
 
@@ -339,7 +341,7 @@ impl FileCacheEntry {
             is_zran,
             dio_enabled: false,
             need_validation,
-            batch_size: RAFS_DEFAULT_CHUNK_SIZE,
+            user_io_batch_size: mgr.user_io_batch_size,
             prefetch_config,
         })
     }

storage/src/cache/fscache/mod.rs

@@ -20,7 +20,6 @@ use crate::cache::worker::{AsyncPrefetchConfig, AsyncWorkerMgr};
 use crate::cache::{BlobCache, BlobCacheMgr};
 use crate::device::{BlobFeatures, BlobInfo, BlobObject};
 use crate::factory::BLOB_FACTORY;
-use crate::RAFS_DEFAULT_CHUNK_SIZE;
 
 use crate::cache::filecache::BLOB_DATA_FILE_SUFFIX;
 
@@ -40,6 +39,7 @@ pub struct FsCacheMgr {
     need_validation: bool,
     blobs_check_count: Arc<AtomicU8>,
     closed: Arc<AtomicBool>,
+    user_io_batch_size: u32,
 }
 
 impl FsCacheMgr {
@@ -49,6 +49,7 @@ impl FsCacheMgr {
         backend: Arc<dyn BlobBackend>,
         runtime: Arc<Runtime>,
         id: &str,
+        user_io_batch_size: u32,
     ) -> Result<FsCacheMgr> {
         if config.cache_compressed {
             return Err(enosys!("fscache doesn't support compressed cache mode"));
@@ -73,6 +74,7 @@ impl FsCacheMgr {
             need_validation: config.cache_validate,
             blobs_check_count: Arc::new(AtomicU8::new(0)),
             closed: Arc::new(AtomicBool::new(false)),
+            user_io_batch_size,
         })
     }
 
@@ -290,7 +292,7 @@ impl FileCacheEntry {
             is_zran,
             dio_enabled: true,
             need_validation,
-            batch_size: RAFS_DEFAULT_CHUNK_SIZE,
+            user_io_batch_size: mgr.user_io_batch_size,
             prefetch_config,
         })
     }

storage/src/cache/worker.rs

@@ -25,7 +25,7 @@ pub(crate) struct AsyncPrefetchConfig {
     pub enable: bool,
     /// Number of working threads.
     pub threads_count: usize,
-    /// Window size to merge/amplify requests.
+    /// The amplify batch size to prefetch data from backend.
     pub prefetch_request_batch_size: usize,
     /// Network bandwidth for prefetch, in unit of Bytes and Zero means no rate limit is set.
     #[allow(unused)]

storage/src/factory.rs

@@ -17,7 +17,7 @@ use std::sync::{Arc, Mutex};
 use std::time::Duration;
 
 use lazy_static::lazy_static;
-use nydus_api::{BackendConfigV2, ConfigV2};
+use nydus_api::{default_user_io_batch_size, BackendConfigV2, ConfigV2};
 use tokio::runtime::{Builder, Runtime};
 use tokio::time;
 
@@ -117,6 +117,10 @@ impl BlobFactory {
     ) -> IOResult<Arc<dyn BlobCache>> {
         let backend_cfg = config.get_backend_config()?;
         let cache_cfg = config.get_cache_config()?;
+        let user_io_batch_size = config
+            .get_rafs_config()
+            .map_or_else(|_| default_user_io_batch_size(), |v| v.user_io_batch_size)
+            as u32;
         let key = BlobCacheMgrKey {
             config: config.clone(),
         };
@@ -128,7 +132,13 @@ impl BlobFactory {
         let backend = Self::new_backend(backend_cfg, &blob_info.blob_id())?;
         let mgr = match cache_cfg.cache_type.as_str() {
             "blobcache" | "filecache" => {
-                let mgr = FileCacheMgr::new(cache_cfg, backend, ASYNC_RUNTIME.clone(), &config.id)?;
+                let mgr = FileCacheMgr::new(
+                    cache_cfg,
+                    backend,
+                    ASYNC_RUNTIME.clone(),
+                    &config.id,
+                    user_io_batch_size,
+                )?;
                 mgr.init()?;
                 Arc::new(mgr) as Arc<dyn BlobCacheMgr>
             }
@@ -139,6 +149,7 @@ impl BlobFactory {
                     backend,
                     ASYNC_RUNTIME.clone(),
                     &config.id,
+                    user_io_batch_size,
                 )?;
                 mgr.init()?;
                 Arc::new(mgr) as Arc<dyn BlobCacheMgr>