fix(ecstore): reduce GET reader setup shard fanout (#4735)

crates/ecstore/src/disk/local.rs

@@ -35,7 +35,9 @@ use crate::disk::{
 use crate::erasure::coding::{self, bitrot_verify};
 use crate::runtime::sources as runtime_sources;
 use bytes::Bytes;
-use metrics::{counter, gauge};
+use metrics::counter;
+#[cfg(target_os = "linux")]
+use metrics::gauge;
 use parking_lot::{Mutex as ParkingLotMutex, RwLock as ParkingLotRwLock};
 use rustfs_filemeta::{
     Cache, FileInfo, FileInfoOpts, FileMeta, MetaCacheEntry, MetacacheWriter, ObjectPartInfo, Opts, RawFileInfo, UpdateFn,

crates/ecstore/src/ecstore_validation_blackbox.rs

@@ -205,11 +205,11 @@ async fn blackbox_get_restores_body_and_enqueues_repair_after_one_corrupt_shard(
     let mut heal_rx = rustfs_common::heal_channel::init_heal_channel()
         .expect("this must be the only ecstore test that owns the heal channel receiver");
 
-    // Force the data-blocks-first reader setup (see
-    // ENV_RUSTFS_GET_DATA_BLOCKS_FIRST_READER_SETUP in set_disk/core/io_primitives.rs):
-    // under the default all-shards strategy the corrupt shard's failed open can
-    // lose the setup-quorum race, in which case no repair is submitted and the
-    // enqueue assertion below would be a coin flip.
+    // Keep data-blocks-first reader setup explicit for this deterministic
+    // repair assertion (see ENV_RUSTFS_GET_DATA_BLOCKS_FIRST_READER_SETUP in
+    // set_disk/core/io_primitives.rs): if a caller opts back into all-shards,
+    // the corrupt shard's failed open can lose the setup-quorum race, making
+    // the enqueue assertion below a coin flip.
     temp_env::async_with_vars([("RUSTFS_GET_DATA_BLOCKS_FIRST_READER_SETUP", Some("true"))], async {
         let (_dirs, set_disks) = make_local_set_disks(4, 2).await;
         let bucket = "bb-corrupt-shard-repair";

crates/ecstore/src/set_disk/core/io_primitives.rs

@@ -70,6 +70,7 @@ use tokio::task::JoinSet;
 
 pub(in crate::set_disk) const EVENT_SET_DISK_READ: &str = "set_disk_read";
 pub(in crate::set_disk) const ENV_RUSTFS_GET_DATA_BLOCKS_FIRST_READER_SETUP: &str = "RUSTFS_GET_DATA_BLOCKS_FIRST_READER_SETUP";
+const DEFAULT_RUSTFS_GET_DATA_BLOCKS_FIRST_READER_SETUP: bool = true;
 pub(in crate::set_disk) const ENV_RUSTFS_GET_CODEC_STREAMING_DATA_BLOCKS_FIRST_READER_SETUP: &str =
     "RUSTFS_GET_CODEC_STREAMING_DATA_BLOCKS_FIRST_READER_SETUP";
 pub(in crate::set_disk) const SLOW_OBJECT_READ_LOG_THRESHOLD: Duration = Duration::from_secs(5);
@@ -990,7 +991,11 @@ pub(in crate::set_disk) fn get_bitrot_reader_setup_strategy(
 ) -> BitrotReaderSetupStrategy {
     match mode {
         BitrotReaderSetupMode::ReadQuorum
-            if prefer_data_blocks_first || rustfs_utils::get_env_bool(ENV_RUSTFS_GET_DATA_BLOCKS_FIRST_READER_SETUP, false) =>
+            if prefer_data_blocks_first
+                || rustfs_utils::get_env_bool(
+                    ENV_RUSTFS_GET_DATA_BLOCKS_FIRST_READER_SETUP,
+                    DEFAULT_RUSTFS_GET_DATA_BLOCKS_FIRST_READER_SETUP,
+                ) =>
         {
             BitrotReaderSetupStrategy::DataBlocksFirst
         }
@@ -4200,6 +4205,18 @@ mod tests {
     #[test]
     #[serial_test::serial]
     fn bitrot_reader_setup_tracks_strategy_counters_and_deferred_readers() {
+        temp_env::with_var(ENV_RUSTFS_GET_DATA_BLOCKS_FIRST_READER_SETUP, None::<&str>, || {
+            assert!(matches!(
+                get_bitrot_reader_setup_strategy(BitrotReaderSetupMode::ReadQuorum, false),
+                BitrotReaderSetupStrategy::DataBlocksFirst
+            ));
+        });
+        temp_env::with_var(ENV_RUSTFS_GET_DATA_BLOCKS_FIRST_READER_SETUP, Some("false"), || {
+            assert!(matches!(
+                get_bitrot_reader_setup_strategy(BitrotReaderSetupMode::ReadQuorum, false),
+                BitrotReaderSetupStrategy::AllShards
+            ));
+        });
         temp_env::with_var(ENV_RUSTFS_GET_DATA_BLOCKS_FIRST_READER_SETUP, Some("true"), || {
             assert!(matches!(
                 get_bitrot_reader_setup_strategy(BitrotReaderSetupMode::ReadQuorum, false),

crates/ecstore/src/set_disk/read.rs

@@ -3771,7 +3771,10 @@ mod tests {
 
         temp_env::async_with_vars(
             [
-                (ENV_RUSTFS_GET_DATA_BLOCKS_FIRST_READER_SETUP, data_blocks_first.then_some("true")),
+                (
+                    ENV_RUSTFS_GET_DATA_BLOCKS_FIRST_READER_SETUP,
+                    Some(if data_blocks_first { "true" } else { "false" }),
+                ),
                 (
                     ENV_RUSTFS_GET_CODEC_STREAMING_DATA_BLOCKS_FIRST_READER_SETUP,
                     codec_data_blocks_first.then_some("true"),