Support zfs page size

CONTRIBUTORS.md

@@ -13,3 +13,6 @@
 * Triss Healy (NiryaAestus)
 * Samuel Cabrero (scabrero)
 
+## Acknowledgements
+
+* M. Gerstner

kanidm_book/src/installing_the_server.md

@@ -27,6 +27,12 @@ You will also need a config file in `/data/server.toml`. It's contents should be
     # ldapbindaddress = "127.0.0.1:3636"
     # The path to the kanidm database.
     db_path = "/data/kanidm.db"
+    # If you have a known filesystem, kanidm can tune sqlite to match. Valid choices are:
+    # [zfs, other]
+    # If you are unsure about this, default to other
+    # zfs:
+    # * sets sqlite pagesize to 64k, you should set recordsize=64k on the zfs filesystem.
+    # db_fs_type = "zfs"
     # TLS ca, certificate and key in pem format. All three must be commented, or present
     # tls_ca = "/data/ca.pem"
     # tls_cert = "/data/cert.pem"

kanidmd/server.toml

@@ -1,6 +1,7 @@
 bindaddress = "127.0.0.1:8443"
 ldapbindaddress = "127.0.0.1:3636"
 db_path = "/tmp/kanidm.db"
+db_fs_type = "zfs"
 tls_ca = "../insecure/ca.pem"
 tls_cert = "../insecure/cert.pem"
 tls_key = "../insecure/key.pem"

kanidmd/src/lib/be/idl_arc_sqlite.rs

@@ -1,6 +1,6 @@
 use crate::audit::AuditScope;
 use crate::be::idl_sqlite::{
-    IdlSqlite, IdlSqliteReadTransaction, IdlSqliteTransaction, IdlSqliteWriteTransaction,
+    FsType, IdlSqlite, IdlSqliteReadTransaction, IdlSqliteTransaction, IdlSqliteWriteTransaction,
 };
 use crate::be::{IdRawEntry, IDL};
 use crate::entry::{Entry, EntryCommitted, EntrySealed};
@@ -848,8 +848,13 @@ impl<'a> IdlArcSqliteWriteTransaction<'a> {
 }
 
 impl IdlArcSqlite {
-    pub fn new(audit: &mut AuditScope, path: &str, pool_size: u32) -> Result<Self, OperationError> {
-        let db = IdlSqlite::new(audit, path, pool_size)?;
+    pub fn new(
+        audit: &mut AuditScope,
+        path: &str,
+        pool_size: u32,
+        fstype: FsType,
+    ) -> Result<Self, OperationError> {
+        let db = IdlSqlite::new(audit, path, pool_size, fstype)?;
         let entry_cache = Arc::new(
             DEFAULT_CACHE_TARGET,
             pool_size as usize,

kanidmd/src/lib/be/idl_sqlite.rs

@@ -6,6 +6,7 @@ use idlset::IDLBitRange;
 use kanidm_proto::v1::{ConsistencyError, OperationError};
 use r2d2::Pool;
 use r2d2_sqlite::SqliteConnectionManager;
+use rusqlite::OpenFlags;
 use rusqlite::OptionalExtension;
 use rusqlite::NO_PARAMS;
 use std::convert::{TryFrom, TryInto};
@@ -17,6 +18,13 @@ use uuid::Uuid;
 const DBV_ID2ENTRY: &str = "id2entry";
 const DBV_INDEXV: &str = "indexv";
 
+#[repr(u32)]
+#[derive(Debug, Copy, Clone)]
+pub enum FsType {
+    Generic = 4096,
+    ZFS = 65536,
+}
+
 #[derive(Debug)]
 pub struct IdSqliteEntry {
     id: i64,
@@ -1090,18 +1098,6 @@ impl IdlSqliteWriteTransaction {
     }
 
     pub fn setup(&self, audit: &mut AuditScope) -> Result<(), OperationError> {
-        // Enable WAL mode, which is just faster and better.
-        //
-        // We have to use stmt + prepare because execute can't handle
-        // the "wal" row on result when this works!
-        self.conn
-            .prepare("PRAGMA journal_mode=WAL;")
-            .and_then(|mut wal_stmt| wal_stmt.query(NO_PARAMS).map(|_| ()))
-            .map_err(|e| {
-                ladmin_error!(audit, "sqlite error {:?}", e);
-                OperationError::SQLiteError
-            })?;
-
         // This stores versions of components. For example:
         // ----------------------
         // | id       | version |
@@ -1236,8 +1232,27 @@ impl IdlSqliteWriteTransaction {
 }
 
 impl IdlSqlite {
-    pub fn new(audit: &mut AuditScope, path: &str, pool_size: u32) -> Result<Self, OperationError> {
-        let manager = SqliteConnectionManager::file(path);
+    pub fn new(
+        audit: &mut AuditScope,
+        path: &str,
+        pool_size: u32,
+        fstype: FsType,
+    ) -> Result<Self, OperationError> {
+        // If provided, set the page size to match the tuning we want. By default we use 4096. The VACUUM
+        // immediately after is so that on db create the page size takes effect.
+        //
+        // Enable WAL mode, which is just faster and better for our needs.
+        let mut flags = OpenFlags::default();
+        // Open with multi thread flags and locking options.
+        flags.insert(OpenFlags::SQLITE_OPEN_NO_MUTEX);
+        let manager = SqliteConnectionManager::file(path)
+            .with_init(move |c| {
+                c.execute_batch(
+                    format!("PRAGMA page_size={}; VACUUM; PRAGMA journal_mode=WAL;", fstype as u32)
+                    .as_str(),
+                )
+            })
+            .with_flags(flags);
         let builder1 = Pool::builder();
         let builder2 = if path == "" {
             // We are in a debug mode, with in memory. We MUST have only
@@ -1278,12 +1293,12 @@ impl IdlSqlite {
 #[cfg(test)]
 mod tests {
     use crate::audit::AuditScope;
-    use crate::be::idl_sqlite::{IdlSqlite, IdlSqliteTransaction};
+    use crate::be::idl_sqlite::{FsType, IdlSqlite, IdlSqliteTransaction};
 
     #[test]
     fn test_idl_sqlite_verify() {
         let mut audit = AuditScope::new("run_test", uuid::Uuid::new_v4(), None);
-        let be = IdlSqlite::new(&mut audit, "", 1).unwrap();
+        let be = IdlSqlite::new(&mut audit, "", 1, FsType::Generic).unwrap();
         let be_w = be.write();
         let r = be_w.verify();
         assert!(r.len() == 0);

kanidmd/src/lib/be/mod.rs

@@ -28,6 +28,9 @@ use crate::be::idl_arc_sqlite::{
     IdlArcSqliteWriteTransaction,
 };
 
+// Re-export this
+pub use crate::be::idl_sqlite::FsType;
+
 const FILTER_SEARCH_TEST_THRESHOLD: usize = 8;
 const FILTER_EXISTS_TEST_THRESHOLD: usize = 0;
 
@@ -1289,12 +1292,13 @@ impl Backend {
         audit: &mut AuditScope,
         path: &str,
         pool_size: u32,
+        fstype: FsType,
         idxmeta: Set<IdxKey>,
     ) -> Result<Self, OperationError> {
         // this has a ::memory() type, but will path == "" work?
         lperf_trace_segment!(audit, "be::new", || {
             let be = Backend {
-                idlayer: Arc::new(IdlArcSqlite::new(audit, path, pool_size)?),
+                idlayer: Arc::new(IdlArcSqlite::new(audit, path, pool_size, fstype)?),
                 idxmeta: Arc::new(CowCell::new(idxmeta)),
             };
 
@@ -1365,7 +1369,9 @@ mod tests {
     use super::super::audit::AuditScope;
     use super::super::entry::{Entry, EntryInit, EntryNew};
     use super::IdxKey;
-    use super::{Backend, BackendTransaction, BackendWriteTransaction, OperationError, IDL};
+    use super::{
+        Backend, BackendTransaction, BackendWriteTransaction, FsType, OperationError, IDL,
+    };
     use crate::value::{IndexType, PartialValue, Value};
 
     macro_rules! run_test {
@@ -1411,7 +1417,8 @@ mod tests {
                 itype: IndexType::EQUALITY,
             });
 
-            let be = Backend::new(&mut audit, "", 1, idxmeta).expect("Failed to setup backend");
+            let be = Backend::new(&mut audit, "", 1, FsType::Generic, idxmeta)
+                .expect("Failed to setup backend");
 
             let mut be_txn = be.write();
 

kanidmd/src/lib/config.rs

@@ -20,6 +20,7 @@ pub struct Configuration {
     pub threads: usize,
     // db type later
     pub db_path: String,
+    pub db_fs_type: Option<String>,
     pub maximum_request: usize,
     pub secure_cookies: bool,
     pub tls_config: Option<TlsConfiguration>,
@@ -61,6 +62,7 @@ impl Configuration {
             ldapaddress: None,
             threads: num_cpus::get(),
             db_path: String::from(""),
+            db_fs_type: None,
             maximum_request: 262_144, // 256k
             // log type
             // log path
@@ -84,6 +86,10 @@ impl Configuration {
         self.db_path = p.to_string();
     }
 
+    pub fn update_db_fs_type(&mut self, p: &Option<String>) {
+        self.db_fs_type = p.as_ref().map(|v| v.to_lowercase());
+    }
+
     pub fn update_bind(&mut self, b: &Option<String>) {
         self.address = b
             .as_ref()

kanidmd/src/lib/core/mod.rs

@@ -33,7 +33,7 @@ use crate::actors::v1_write::{
 };
 use crate::async_log;
 use crate::audit::AuditScope;
-use crate::be::{Backend, BackendTransaction};
+use crate::be::{Backend, BackendTransaction, FsType};
 use crate::crypto::setup_tls;
 use crate::filter::{Filter, FilterInvalid};
 use crate::idm::server::IdmServer;
@@ -1270,7 +1270,24 @@ fn setup_backend(config: &Configuration, schema: &Schema) -> Result<Backend, Ope
 
     let mut audit_be = AuditScope::new("backend_setup", uuid::Uuid::new_v4(), config.log_level);
     let pool_size: u32 = config.threads as u32;
-    let be = Backend::new(&mut audit_be, config.db_path.as_str(), pool_size, idxmeta);
+    let fstype: FsType = if config
+        .db_fs_type
+        .as_ref()
+        .map(|s| s == "zfs")
+        .unwrap_or(false)
+    {
+        FsType::ZFS
+    } else {
+        FsType::Generic
+    };
+
+    let be = Backend::new(
+        &mut audit_be,
+        config.db_path.as_str(),
+        pool_size,
+        fstype,
+        idxmeta,
+    );
     // debug!
     audit_be.write_log();
     be

kanidmd/src/lib/macros.rs

@@ -2,7 +2,7 @@
 macro_rules! run_test_no_init {
     ($test_fn:expr) => {{
         use crate::audit::AuditScope;
-        use crate::be::Backend;
+        use crate::be::{Backend, FsType};
         use crate::schema::Schema;
         use crate::server::QueryServer;
         use crate::utils::duration_from_epoch_now;
@@ -22,7 +22,7 @@ macro_rules! run_test_no_init {
             let schema_txn = schema_outer.write();
             schema_txn.reload_idxmeta()
         };
-        let be = match Backend::new(&mut audit, "", 1, idxmeta) {
+        let be = match Backend::new(&mut audit, "", 1, FsType::Generic, idxmeta) {
             Ok(be) => be,
             Err(e) => {
                 audit.write_log();
@@ -46,7 +46,7 @@ macro_rules! run_test_no_init {
 macro_rules! run_test {
     ($test_fn:expr) => {{
         use crate::audit::AuditScope;
-        use crate::be::Backend;
+        use crate::be::{Backend, FsType};
         use crate::schema::Schema;
         use crate::server::QueryServer;
         use crate::utils::duration_from_epoch_now;
@@ -66,7 +66,7 @@ macro_rules! run_test {
             let schema_txn = schema_outer.write();
             schema_txn.reload_idxmeta()
         };
-        let be = match Backend::new(&mut audit, "", 1, idxmeta) {
+        let be = match Backend::new(&mut audit, "", 1, FsType::Generic, idxmeta) {
             Ok(be) => be,
             Err(e) => {
                 audit.write_log();
@@ -116,7 +116,7 @@ macro_rules! entry_str_to_account {
 macro_rules! run_idm_test {
     ($test_fn:expr) => {{
         use crate::audit::AuditScope;
-        use crate::be::Backend;
+        use crate::be::{Backend, FsType};
         use crate::idm::server::IdmServer;
         use crate::schema::Schema;
         use crate::server::QueryServer;
@@ -137,7 +137,8 @@ macro_rules! run_idm_test {
             let schema_txn = schema_outer.write();
             schema_txn.reload_idxmeta()
         };
-        let be = Backend::new(&mut audit, "", 1, idxmeta).expect("Failed to init be");
+        let be =
+            Backend::new(&mut audit, "", 1, FsType::Generic, idxmeta).expect("Failed to init be");
 
         let test_server = QueryServer::new(be, schema_outer);
         test_server

kanidmd/src/lib/plugins/macros.rs

@@ -19,7 +19,7 @@ macro_rules! setup_test {
             let schema_txn = schema_outer.write();
             schema_txn.reload_idxmeta()
         };
-        let be = Backend::new($au, "", 1, idxmeta).expect("Failed to init BE");
+        let be = Backend::new($au, "", 1, FsType::Generic, idxmeta).expect("Failed to init BE");
 
         let qs = QueryServer::new(be, schema_outer);
         qs.initialise_helper($au, duration_from_epoch_now())
@@ -48,7 +48,7 @@ macro_rules! run_create_test {
         $check:expr
     ) => {{
         use crate::audit::AuditScope;
-        use crate::be::Backend;
+        use crate::be::{Backend, FsType};
         use crate::event::CreateEvent;
         use crate::schema::Schema;
         use crate::server::QueryServer;
@@ -103,7 +103,7 @@ macro_rules! run_modify_test {
         $check:expr
     ) => {{
         use crate::audit::AuditScope;
-        use crate::be::Backend;
+        use crate::be::{Backend, FsType};
         use crate::event::ModifyEvent;
         use crate::schema::Schema;
         use crate::server::QueryServer;
@@ -165,7 +165,7 @@ macro_rules! run_delete_test {
         $check:expr
     ) => {{
         use crate::audit::AuditScope;
-        use crate::be::Backend;
+        use crate::be::{Backend, FsType};
         use crate::event::DeleteEvent;
         use crate::schema::Schema;
         use crate::server::QueryServer;

kanidmd/src/server/main.rs

@@ -33,6 +33,7 @@ struct ServerConfig {
     pub ldapbindaddress: Option<String>,
     // pub threads: Option<usize>,
     pub db_path: String,
+    pub db_fs_type: Option<String>,
     pub tls_ca: Option<String>,
     pub tls_cert: Option<String>,
     pub tls_key: Option<String>,
@@ -273,6 +274,7 @@ async fn main() {
 
     config.update_log_level(ll);
     config.update_db_path(&sconfig.db_path.as_str());
+    config.update_db_fs_type(&sconfig.db_fs_type);
     config.update_tls(&sconfig.tls_ca, &sconfig.tls_cert, &sconfig.tls_key);
     config.update_bind(&sconfig.bindaddress);
     config.update_ldapbind(&sconfig.ldapbindaddress);