feat: add opt-in hotpath profiling (#5488)

* feat: add opt-in hotpath profiling

Co-Authored-By: heihutu <heihutu@gmail.com>

* test: fix vault kms client construction

Co-Authored-By: heihutu <heihutu@gmail.com>

---------

Co-authored-by: heihutu <heihutu@gmail.com>

crates/ecstore/src/erasure/coding/bitrot.rs

@@ -103,7 +103,7 @@ where
     /// or `out` is larger than one shard. On error `out`'s contents are
     /// unspecified but never contain bytes that failed the hash check — the copy
     /// happens only after verification.
-    #[cfg_attr(feature = "hotpath", hotpath::measure)]
+    #[hotpath::measure]
     pub async fn read(&mut self, out: &mut [u8]) -> std::io::Result<usize> {
         let want = out.len();
         self.begin_read(want)?;
@@ -303,7 +303,7 @@ where
 
     /// Write a (hash+data) block. Returns the number of data bytes written.
     /// Returns an error if called after a short write or if data exceeds shard_size.
-    #[cfg_attr(feature = "hotpath", hotpath::measure(label = "BitrotWriter::write"))]
+    #[hotpath::measure(label = "BitrotWriter::write")]
     pub async fn write(&mut self, buf: &[u8]) -> std::io::Result<usize> {
         if buf.is_empty() {
             return Ok(0);
@@ -455,7 +455,7 @@ pub fn bitrot_shard_file_size(size: usize, shard_size: usize, algo: HashAlgorith
 /// stores those as whole-file bitrot with no interleaved hash, so the size guard
 /// on the next line would reject a genuinely healthy part. Reading legacy V1
 /// whole-file-bitrot objects would need a separate verification path.
-#[cfg_attr(feature = "hotpath", hotpath::measure)]
+#[hotpath::measure]
 pub async fn bitrot_verify<R: AsyncRead + Unpin + Send>(
     mut r: R,
     want_size: usize,

crates/ecstore/src/erasure/coding/decode.rs

@@ -691,7 +691,7 @@ impl<R> ParallelReader<R>
 where
     R: crate::erasure::coding::ShardSource,
 {
-    #[cfg_attr(feature = "hotpath", hotpath::measure)]
+    #[hotpath::measure]
     pub async fn read(&mut self) -> (Vec<Option<Vec<u8>>>, Vec<Option<Error>>) {
         // On the reconstruction-verifying GET path, read every live shard reader
         // in lockstep so all readers advance one block per stripe and stay
@@ -1505,7 +1505,7 @@ where
 }
 
 impl Erasure {
-    #[cfg_attr(feature = "hotpath", hotpath::measure)]
+    #[hotpath::measure]
     pub async fn decode<W, R>(
         &self,
         writer: &mut W,

crates/ecstore/src/erasure/coding/encode.rs

@@ -504,7 +504,7 @@ impl Erasure {
         Ok((reader, total))
     }
 
-    #[cfg_attr(feature = "hotpath", hotpath::measure)]
+    #[hotpath::measure]
     pub async fn encode<R>(
         self: Arc<Self>,
         reader: R,
@@ -670,7 +670,7 @@ impl Erasure {
         Ok((reader, total))
     }
 
-    #[cfg_attr(feature = "hotpath", hotpath::measure)]
+    #[hotpath::measure]
     pub async fn encode_batched<R>(
         self: Arc<Self>,
         mut reader: R,
@@ -798,7 +798,7 @@ impl Erasure {
 
     /// Fast path for small inline objects: skip tokio::spawn + mpsc channel.
     /// Reads all data, encodes directly, writes shards sequentially.
-    #[cfg_attr(feature = "hotpath", hotpath::measure)]
+    #[hotpath::measure]
     pub async fn encode_inline_small<R>(
         self: Arc<Self>,
         reader: R,
@@ -813,7 +813,7 @@ impl Erasure {
 
     /// Fast path for single-block non-inline objects: avoids the producer/consumer
     /// pipeline in `encode()` while keeping the same writer/quorum/shutdown semantics.
-    #[cfg_attr(feature = "hotpath", hotpath::measure)]
+    #[hotpath::measure]
     pub async fn encode_single_block_non_inline<R>(
         self: Arc<Self>,
         reader: R,

crates/ecstore/src/erasure/coding/erasure.rs

@@ -640,7 +640,7 @@ impl Erasure {
     /// # Returns
     /// A vector of encoded shards as `Bytes`.
     #[tracing::instrument(level = "debug", skip_all, fields(data_len=data.len()))]
-    #[cfg_attr(feature = "hotpath", hotpath::measure)]
+    #[hotpath::measure]
     pub fn encode_data(&self, data: &[u8]) -> io::Result<Vec<Bytes>> {
         let shard_size_fn = if self.uses_legacy {
             calc_shard_size_legacy
@@ -688,7 +688,7 @@ impl Erasure {
 
     /// Encode owned data, avoiding a copy when the caller already has a heap buffer.
     /// Falls back to copying into a new buffer if zero-copy conversion fails.
-    #[cfg_attr(feature = "hotpath", hotpath::measure)]
+    #[hotpath::measure]
     pub fn encode_data_owned(&self, data: Vec<u8>) -> io::Result<Vec<Bytes>> {
         let shard_size_fn = if self.uses_legacy {
             calc_shard_size_legacy
@@ -752,7 +752,7 @@ impl Erasure {
     /// block), the `resize(need_total_size)` below stays within capacity for every
     /// `data_len <= block_size` — both shard-size formulas are monotone in
     /// `data_len` — so this function never reallocates the buffer.
-    #[cfg_attr(feature = "hotpath", hotpath::measure)]
+    #[hotpath::measure]
     pub fn encode_data_bytes_mut(&self, mut data_buffer: BytesMut, data_len: usize) -> io::Result<Vec<Bytes>> {
         let shard_size_fn = if self.uses_legacy {
             calc_shard_size_legacy
@@ -805,7 +805,7 @@ impl Erasure {
     ///
     /// # Returns
     /// Ok if reconstruction succeeds, error otherwise.
-    #[cfg_attr(feature = "hotpath", hotpath::measure)]
+    #[hotpath::measure]
     pub fn decode_data(&self, shards: &mut [Option<Vec<u8>>]) -> io::Result<()> {
         if self.parity_shards > 0 {
             if self.uses_legacy {
@@ -825,7 +825,7 @@ impl Erasure {
     }
 
     /// Decode and reconstruct missing data shards, then regenerate parity shards.
-    #[cfg_attr(feature = "hotpath", hotpath::measure)]
+    #[hotpath::measure]
     pub fn decode_data_and_parity(&self, shards: &mut [Option<Vec<u8>>]) -> io::Result<()> {
         if self.parity_shards > 0 {
             if self.uses_legacy {