feat(pageserver): add direct io pageserver config (#8622)

Part of #8130, [RFC: Direct IO For Pageserver](https://github.com/neondatabase/neon/blob/problame/direct-io-rfc/docs/rfcs/034-direct-io-for-pageserver.md) ## Description Add pageserver config for evaluating/enabling direct I/O. - Disabled: current default, uses buffered io as is. - Evaluate: still uses buffered io, but could do alignment checking and perf simulation (pad latency by direct io RW to a fake file). - Enabled: uses direct io, behavior on alignment error is configurable. Signed-off-by: Yuchen Liang <[email protected]>
neondatabase · Aug 12, 2024 · 77bb6c4 · 77bb6c4
1 parent e082226
commit 77bb6c4
Show file tree

Hide file tree

Showing 4 changed files with 68 additions and 0 deletions.
diff --git a/libs/pageserver_api/src/models.rs b/libs/pageserver_api/src/models.rs
@@ -947,6 +947,8 @@ pub struct TopTenantShardsResponse {
 }
 
 pub mod virtual_file {
+    use std::path::PathBuf;
+
     #[derive(
         Copy,
         Clone,
@@ -965,6 +967,53 @@ pub mod virtual_file {
         #[cfg(target_os = "linux")]
         TokioEpollUring,
     }
+
+    /// Direct IO modes for a pageserver.
+    #[derive(Debug, PartialEq, Eq, Clone, serde::Deserialize, serde::Serialize, Default)]
+    #[serde(tag = "mode", rename_all = "kebab-case", deny_unknown_fields)]
+    pub enum DirectIoMode {
+        /// Direct IO disabled (uses usual buffered IO).
+        #[default]
+        Disabled,
+        /// Direct IO disabled (performs checks and perf simulations).
+        Evaluate {
+            /// Alignment check level
+            alignment_check: DirectIoAlignmentCheckLevel,
+            /// Latency padded for performance simulation.
+            latency_padding: DirectIoLatencyPadding,
+        },
+        /// Direct IO enabled.
+        Enabled {
+            /// Actions to perform on alignment error.
+            on_alignment_error: DirectIoOnAlignmentErrorAction,
+        },
+    }
+
+    #[derive(Debug, PartialEq, Eq, Clone, serde::Deserialize, serde::Serialize, Default)]
+    #[serde(rename_all = "kebab-case")]
+    pub enum DirectIoAlignmentCheckLevel {
+        #[default]
+        Error,
+        Log,
+        None,
+    }
+
+    #[derive(Debug, PartialEq, Eq, Clone, serde::Deserialize, serde::Serialize, Default)]
+    #[serde(rename_all = "kebab-case")]
+    pub enum DirectIoOnAlignmentErrorAction {
+        Error,
+        #[default]
+        FallbackToBuffered,
+    }
+
+    #[derive(Debug, PartialEq, Eq, Clone, serde::Deserialize, serde::Serialize, Default)]
+    #[serde(tag = "type", rename_all = "kebab-case")]
+    pub enum DirectIoLatencyPadding {
+        /// Pad virtual file operations with IO to a fake file.
+        FakeFileRW { path: PathBuf },
+        #[default]
+        None,
+    }
 }
 
 // Wrapped in libpq CopyData

diff --git a/pageserver/src/bin/pageserver.rs b/pageserver/src/bin/pageserver.rs
@@ -123,6 +123,7 @@ fn main() -> anyhow::Result<()> {
 
     // after setting up logging, log the effective IO engine choice and read path implementations
     info!(?conf.virtual_file_io_engine, "starting with virtual_file IO engine");
+    info!(?conf.virtual_file_direct_io, "starting with virtual_file Direct IO settings");
     info!(?conf.get_impl, "starting with get page implementation");
     info!(?conf.get_vectored_impl, "starting with vectored get page implementation");
     info!(?conf.compact_level0_phase1_value_access, "starting with setting for compact_level0_phase1_value_access");

diff --git a/pageserver/src/config.rs b/pageserver/src/config.rs
@@ -300,6 +300,9 @@ pub struct PageServerConf {
     /// This flag is temporary and will be removed after gradual rollout.
     /// See <https://github.com/neondatabase/neon/issues/8184>.
     pub compact_level0_phase1_value_access: CompactL0Phase1ValueAccess,
+
+    /// Direct IO settings
+    pub virtual_file_direct_io: virtual_file::DirectIoMode,
 }
 
 /// We do not want to store this in a PageServerConf because the latter may be logged
@@ -408,6 +411,8 @@ struct PageServerConfigBuilder {
     l0_flush: BuilderValue<L0FlushConfig>,
 
     compact_level0_phase1_value_access: BuilderValue<CompactL0Phase1ValueAccess>,
+
+    virtual_file_direct_io: BuilderValue<virtual_file::DirectIoMode>,
 }
 
 impl PageServerConfigBuilder {
@@ -498,6 +503,7 @@ impl PageServerConfigBuilder {
             ephemeral_bytes_per_memory_kb: Set(DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB),
             l0_flush: Set(L0FlushConfig::default()),
             compact_level0_phase1_value_access: Set(CompactL0Phase1ValueAccess::default()),
+            virtual_file_direct_io: Set(virtual_file::DirectIoMode::default()),
         }
     }
 }
@@ -685,6 +691,10 @@ impl PageServerConfigBuilder {
         self.compact_level0_phase1_value_access = BuilderValue::Set(value);
     }
 
+    pub fn virtual_file_direct_io(&mut self, value: virtual_file::DirectIoMode) {
+        self.virtual_file_direct_io = BuilderValue::Set(value);
+    }
+
     pub fn build(self, id: NodeId) -> anyhow::Result<PageServerConf> {
         let default = Self::default_values();
 
@@ -743,6 +753,7 @@ impl PageServerConfigBuilder {
                 ephemeral_bytes_per_memory_kb,
                 l0_flush,
                 compact_level0_phase1_value_access,
+                virtual_file_direct_io,
             }
             CUSTOM LOGIC
             {
@@ -1018,6 +1029,9 @@ impl PageServerConf {
                 "compact_level0_phase1_value_access" => {
                     builder.compact_level0_phase1_value_access(utils::toml_edit_ext::deserialize_item(item).context("compact_level0_phase1_value_access")?)
                 }
+                "virtual_file_direct_io" => {
+                    builder.virtual_file_direct_io(utils::toml_edit_ext::deserialize_item(item).context("virtual_file_direct_io")?)
+                }
                 _ => bail!("unrecognized pageserver option '{key}'"),
             }
         }
@@ -1103,6 +1117,7 @@ impl PageServerConf {
             ephemeral_bytes_per_memory_kb: defaults::DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB,
             l0_flush: L0FlushConfig::default(),
             compact_level0_phase1_value_access: CompactL0Phase1ValueAccess::default(),
+            virtual_file_direct_io: virtual_file::DirectIoMode::default(),
         }
     }
 }
@@ -1345,6 +1360,7 @@ background_task_maximum_delay = '334 s'
                 ephemeral_bytes_per_memory_kb: defaults::DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB,
                 l0_flush: L0FlushConfig::default(),
                 compact_level0_phase1_value_access: CompactL0Phase1ValueAccess::default(),
+                virtual_file_direct_io: virtual_file::DirectIoMode::default(),
             },
             "Correct defaults should be used when no config values are provided"
         );
@@ -1420,6 +1436,7 @@ background_task_maximum_delay = '334 s'
                 ephemeral_bytes_per_memory_kb: defaults::DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB,
                 l0_flush: L0FlushConfig::default(),
                 compact_level0_phase1_value_access: CompactL0Phase1ValueAccess::default(),
+                virtual_file_direct_io: virtual_file::DirectIoMode::default(),
             },
             "Should be able to parse all basic config values correctly"
         );

diff --git a/pageserver/src/virtual_file.rs b/pageserver/src/virtual_file.rs
@@ -34,6 +34,7 @@ pub use io_engine::FeatureTestResult as IoEngineFeatureTestResult;
 mod metadata;
 mod open_options;
 use self::owned_buffers_io::write::OwnedAsyncWriter;
+pub(crate) use api::DirectIoMode;
 pub(crate) use io_engine::IoEngineKind;
 pub(crate) use metadata::Metadata;
 pub(crate) use open_options::*;