v4.0: feat: enable direct-io in accounts-db snapshot operations with opt-out flag (backport of #10790)

CHANGELOG.md

@@ -89,6 +89,7 @@ prerelease version. The new interpretation is as follows:
 #### Changes
 * `agave-validator exit` now saves bank state before exiting. This enables restarts from local state when snapshot generation is disabled.
 * Added `--accounts-index-limit` to specify the memory limit of the accounts index.
+* Snapshot archive unpacking now uses direct I/O by default to improve performance by bypassing the OS page cache. Use `--no-accounts-db-snapshots-direct-io` to opt out if your file system does not support `O_DIRECT`. Direct I/O will be extended to snapshot creation in a future release.
 ### CLI
 #### Breaking
 * Removed deprecated arguments

ledger-tool/src/args.rs

@@ -150,6 +150,15 @@ pub fn accounts_db_args<'a, 'b>() -> Box<[Arg<'a, 'b>]> {
             .takes_value(true)
             .help("The number of ancient storages the ancient slot combining should converge to.")
             .hidden(hidden_unless_forced()),
+        Arg::with_name("no_accounts_db_snapshots_direct_io")
+            .long("no-accounts-db-snapshots-direct-io")
+            .help("Disable direct I/O use for accounts-db snapshot operations")
+            .long_help(
+                "Do *not* use direct I/O for accounts-db file operations related to snapshot \
+                 processsing. Direct I/O can improve performance by bypassing OS page cache, but \
+                 requires the file systems hosting snapshots and accounts-db directories to \
+                 support files opened with the O_DIRECT flag.",
+            ),
     ]
     .into_boxed_slice()
 }
@@ -364,6 +373,7 @@ pub fn get_accounts_db_config(
         use_registered_io_uring_buffers: resource_limits::check_memlock_limit_for_disk_io(
             solana_accounts_db::accounts_db::TOTAL_IO_URING_BUFFERS_SIZE_LIMIT,
         ),
+        snapshots_use_direct_io: !arg_matches.is_present("no_accounts_db_snapshots_direct_io"),
         ..AccountsDbConfig::default()
     }
 }

validator/src/commands/run/args.rs

@@ -1027,6 +1027,17 @@ pub fn add_args<'a>(app: App<'a, 'a>, default_args: &'a DefaultArgs) -> App<'a,
             )
             .hidden(hidden_unless_forced()),
     )
+    .arg(
+        Arg::with_name("no_accounts_db_snapshots_direct_io")
+            .long("no-accounts-db-snapshots-direct-io")
+            .help("Disable direct I/O use for accounts-db snapshot operations")
+            .long_help(
+                "Do *not* use direct I/O for accounts-db file operations related to snapshot \
+                 processsing. Direct I/O can improve performance by bypassing OS page cache, but \
+                 requires the file systems hosting snapshots and accounts-db directories to \
+                 support files opened with the O_DIRECT flag.",
+            ),
+    )
     .arg(
         Arg::with_name("accounts_index_scan_results_limit_mb")
             .long("accounts-index-scan-results-limit-mb")

validator/src/commands/run/execute.rs

@@ -707,7 +707,7 @@ pub fn execute(
         use_registered_io_uring_buffers: resource_limits::check_memlock_limit_for_disk_io(
             solana_accounts_db::accounts_db::TOTAL_IO_URING_BUFFERS_SIZE_LIMIT,
         ),
-        snapshots_use_direct_io: false,
+        snapshots_use_direct_io: !matches.is_present("no_accounts_db_snapshots_direct_io"),
     };
 
     let on_start_geyser_plugin_config_files = if matches.is_present("geyser_plugin_config") {