diff --git a/Cargo.lock b/Cargo.lock
index ea7dbf6d..9482d3b0 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -533,6 +533,7 @@ dependencies = [
  "infra",
  "itertools 0.11.0",
  "lazy_static",
+ "lru",
  "mime",
  "once_cell",
  "opener",
@@ -2801,6 +2802,15 @@ version = "0.4.20"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b5e6163cb8c49088c2c36f57875e58ccd8c87c7427f7fbd50ea6710b2f3f2e8f"
 
+[[package]]
+name = "lru"
+version = "0.12.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "db2c024b41519440580066ba82aab04092b333e09066a5eb86c7c4890df31f22"
+dependencies = [
+ "hashbrown 0.14.3",
+]
+
 [[package]]
 name = "mac"
 version = "0.1.1"
diff --git a/Cargo.toml b/Cargo.toml
index b9b75484..01e19fc7 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -65,6 +65,7 @@ casbin = { version = "2.1.0" }
 dotenvy = "0.15.7"
 url = "2.5.0"
 brotli = "3.4.0"
+lru.workspace = true
 
 # collab
 collab = { version = "0.1.0", features = ["async-plugin"] }
@@ -148,6 +149,7 @@ tracing = { version = "0.1"}
 collab-entity = { version = "0.1.0" }
 gotrue = { path = "libs/gotrue" }
 redis = "0.24.0"
+lru = "0.12.2"
 
 [profile.release]
 lto = true
diff --git a/libs/database/src/collab/collab_db_ops.rs b/libs/database/src/collab/collab_db_ops.rs
index b1880ddf..f4acc1d8 100644
--- a/libs/database/src/collab/collab_db_ops.rs
+++ b/libs/database/src/collab/collab_db_ops.rs
@@ -11,6 +11,7 @@ use crate::pg_row::AFSnapshotRow;
 use app_error::AppError;
 use chrono::{Duration, Utc};
 use futures_util::stream::BoxStream;
+
 use sqlx::postgres::PgRow;
 use sqlx::{Error, Executor, PgPool, Postgres, Row, Transaction};
 use std::collections::HashMap;
@@ -178,11 +179,14 @@ pub async fn insert_into_af_collab(
 }
 
 #[inline]
-pub async fn select_blob_from_af_collab(
-  pg_pool: &PgPool,
+pub async fn select_blob_from_af_collab<'a, E>(
+  conn: E,
   collab_type: &CollabType,
   object_id: &str,
-) -> Result<Vec<u8>, sqlx::Error> {
+) -> Result<Vec<u8>, sqlx::Error>
+where
+  E: Executor<'a, Database = Postgres>,
+{
   let partition_key = collab_type.value();
   sqlx::query_scalar!(
     r#"
@@ -193,7 +197,7 @@ pub async fn select_blob_from_af_collab(
     object_id,
     partition_key,
   )
-  .fetch_one(pg_pool)
+  .fetch_one(conn)
   .await
 }
 
@@ -330,18 +334,13 @@ pub async fn should_create_snapshot<'a, E: Executor<'a, Database = Postgres>>(
 /// of snapshots stored for the specified `oid` does not exceed the provided `snapshot_limit`. If the limit
 /// is exceeded, the oldest snapshots are deleted to maintain the limit.
 ///
-pub(crate) async fn create_snapshot_and_maintain_limit(
-  pg_pool: &PgPool,
+pub(crate) async fn create_snapshot_and_maintain_limit<'a>(
+  mut transaction: Transaction<'a, Postgres>,
   oid: &str,
   encoded_collab_v1: &[u8],
   workspace_id: &Uuid,
   snapshot_limit: i64,
 ) -> Result<AFSnapshotMeta, AppError> {
-  let mut tx = pg_pool
-    .begin()
-    .await
-    .context("acquire transaction to insert collab snapshot")?;
-
   let snapshot_meta = sqlx::query_as!(
     AFSnapshotMeta,
     r#"
@@ -355,7 +354,7 @@ pub(crate) async fn create_snapshot_and_maintain_limit(
     0,
     workspace_id,
   )
-  .fetch_one(tx.deref_mut())
+  .fetch_one(transaction.deref_mut())
   .await?;
 
   // When a new snapshot is created that surpasses the preset limit, older snapshots will be deleted to maintain the limit
@@ -367,10 +366,11 @@ pub(crate) async fn create_snapshot_and_maintain_limit(
     )
     .bind(oid)
     .bind(snapshot_limit)
-    .execute(tx.deref_mut())
+    .execute(transaction.deref_mut())
     .await?;
 
-  tx.commit()
+  transaction
+    .commit()
     .await
     .context("fail to commit the transaction to insert collab snapshot")?;
 
diff --git a/libs/database/src/collab/collab_storage.rs b/libs/database/src/collab/collab_storage.rs
index 7c94ae9c..4dcf4710 100644
--- a/libs/database/src/collab/collab_storage.rs
+++ b/libs/database/src/collab/collab_storage.rs
@@ -4,7 +4,6 @@ use app_error::AppError;
 use async_trait::async_trait;
 use collab::core::collab::MutexCollab;
 use collab::core::collab_plugin::EncodedCollab;
-
 use database_entity::dto::{
   AFAccessLevel, AFRole, AFSnapshotMeta, AFSnapshotMetas, CollabParams, CreateCollabParams,
   InsertSnapshotParams, QueryCollab, QueryCollabParams, QueryCollabResult, SnapshotData,
@@ -14,7 +13,9 @@ use sqlx::types::Uuid;
 use sqlx::{Executor, PgPool, Postgres, Transaction};
 use std::collections::HashMap;
 use std::sync::{Arc, Weak};
-use tracing::{debug, event, warn};
+use std::time::Duration;
+use tokio::time::sleep;
+use tracing::{debug, event, warn, Level};
 use validator::Validate;
 
 pub const COLLAB_SNAPSHOT_LIMIT: i64 = 15;
@@ -26,7 +27,7 @@ pub type DatabaseResult<T, E = AppError> = core::result::Result<T, E>;
 #[async_trait]
 pub trait CollabStorageAccessControl: Send + Sync + 'static {
   /// Checks if the user with the given ID can access the [Collab] with the given ID.
-  async fn get_collab_access_level<'a, E: Executor<'a, Database = Postgres>>(
+  async fn get_or_refresh_collab_access_level<'a, E: Executor<'a, Database = Postgres>>(
     &self,
     uid: &i64,
     oid: &str,
@@ -246,28 +247,50 @@ impl CollabStoragePgImpl {
     &self,
     _uid: &i64,
     params: QueryCollabParams,
-  ) -> DatabaseResult<EncodedCollab> {
+  ) -> Result<EncodedCollab, AppError> {
     event!(
-      tracing::Level::INFO,
+      Level::INFO,
       "Get encoded collab:{} from disk",
       params.object_id
     );
-    match collab_db_ops::select_blob_from_af_collab(
-      &self.pg_pool,
-      &params.collab_type,
-      &params.object_id,
-    )
-    .await
-    {
-      Ok(data) => EncodedCollab::decode_from_bytes(&data)
-        .map_err(|err| AppError::Internal(anyhow!("fail to decode data to EncodedDoc: {:?}", err))),
-      Err(e) => match e {
-        sqlx::Error::RowNotFound => {
-          let msg = format!("Can't find the row for query: {:?}", params);
-          Err(AppError::RecordNotFound(msg))
+
+    const MAX_ATTEMPTS: usize = 3;
+    let mut attempts = 0;
+
+    loop {
+      let result = collab_db_ops::select_blob_from_af_collab(
+        &self.pg_pool,
+        &params.collab_type,
+        &params.object_id,
+      )
+      .await;
+
+      match result {
+        Ok(data) => {
+          return tokio::task::spawn_blocking(move || {
+            EncodedCollab::decode_from_bytes(&data).map_err(|err| {
+              AppError::Internal(anyhow!("fail to decode data to EncodedCollab: {:?}", err))
+            })
+          })
+          .await?;
         },
-        _ => Err(e.into()),
-      },
+        Err(e) => {
+          // Handle non-retryable errors immediately
+          if matches!(e, sqlx::Error::RowNotFound) {
+            let msg = format!("Can't find the row for query: {:?}", params);
+            return Err(AppError::RecordNotFound(msg));
+          }
+
+          // Increment attempts and retry if below MAX_ATTEMPTS and the error is retryable
+          if attempts < MAX_ATTEMPTS - 1 && matches!(e, sqlx::Error::PoolTimedOut) {
+            attempts += 1;
+            sleep(Duration::from_millis(500 * attempts as u64)).await;
+            continue;
+          } else {
+            return Err(e.into());
+          }
+        },
+      }
     }
   }
 
@@ -302,15 +325,23 @@ impl CollabStoragePgImpl {
     params.validate()?;
 
     debug!("create snapshot for object:{}", params.object_id);
-    let meta = collab_db_ops::create_snapshot_and_maintain_limit(
-      &self.pg_pool,
-      &params.object_id,
-      &params.encoded_collab_v1,
-      &params.workspace_id.parse::<Uuid>()?,
-      COLLAB_SNAPSHOT_LIMIT,
-    )
-    .await?;
-    Ok(meta)
+    match self.pg_pool.try_begin().await {
+      Ok(Some(transaction)) => {
+        let meta = collab_db_ops::create_snapshot_and_maintain_limit(
+          transaction,
+          &params.object_id,
+          &params.encoded_collab_v1,
+          &params.workspace_id.parse::<Uuid>()?,
+          COLLAB_SNAPSHOT_LIMIT,
+        )
+        .await?;
+
+        Ok(meta)
+      },
+      _ => Err(AppError::Internal(anyhow!(
+        "fail to acquire transaction to create snapshot",
+      ))),
+    }
   }
 
   pub async fn get_collab_snapshot(&self, snapshot_id: &i64) -> DatabaseResult<SnapshotData> {
diff --git a/src/api/workspace.rs b/src/api/workspace.rs
index 01e3a359..d9109684 100644
--- a/src/api/workspace.rs
+++ b/src/api/workspace.rs
@@ -312,7 +312,6 @@ async fn create_collab_handler(
   };
 
   params.validate().map_err(AppError::from)?;
-  // TODO(nathan): should override the existing collab if it exists
   state.collab_storage.upsert_collab(&uid, params).await?;
   Ok(Json(AppResponse::Ok()))
 }
@@ -391,7 +390,6 @@ async fn batch_create_collab_handler(
     .await
     .context("acquire transaction to upsert collab")
     .map_err(AppError::from)?;
-
   for params in collab_params_list {
     state
       .collab_storage
diff --git a/src/biz/collab/access_control.rs b/src/biz/collab/access_control.rs
index 07fcf4ff..5fa9f688 100644
--- a/src/biz/collab/access_control.rs
+++ b/src/biz/collab/access_control.rs
@@ -68,7 +68,7 @@ where
   CollabAC: CollabAccessControl,
   WorkspaceAC: WorkspaceAccessControl,
 {
-  async fn get_collab_access_level<'a, E: Executor<'a, Database = Postgres>>(
+  async fn get_or_refresh_collab_access_level<'a, E: Executor<'a, Database = Postgres>>(
     &self,
     uid: &i64,
     oid: &str,
diff --git a/src/biz/collab/mem_cache.rs b/src/biz/collab/mem_cache.rs
index 349cf3bd..aa50fc21 100644
--- a/src/biz/collab/mem_cache.rs
+++ b/src/biz/collab/mem_cache.rs
@@ -1,92 +1,52 @@
 use crate::state::RedisClient;
-
 use collab::core::collab_plugin::EncodedCollab;
-
-use redis::AsyncCommands;
-
-use rand::{thread_rng, Rng};
+use lru::LruCache;
+use std::num::NonZeroUsize;
 use std::sync::Arc;
 use tokio::sync::Mutex;
-use tracing::{error, trace};
-
-const ENCODED_COLLAB_KEY_PREFIX: &str = "encoded_collab";
+use tracing::error;
 
 #[derive(Clone)]
 pub struct CollabMemCache {
-  /// Workaround for Redis cache
-  ///   This cache instance has a unique identifier. Use only the current ID with the cache to ensure its relevance.
-  ///   Using this ID to avoid outdated cache data after server restarts.
-  id: u32,
-  redis_client: Arc<Mutex<RedisClient>>,
+  lru_cache: Arc<Mutex<LruCache<String, Vec<u8>>>>,
 }
 
 impl CollabMemCache {
-  pub fn new(redis_client: RedisClient) -> Self {
-    let mut rng = thread_rng();
-    let id: u32 = rng.gen();
-
+  pub fn new(_redis_client: RedisClient) -> Self {
+    let lru = LruCache::new(NonZeroUsize::new(5000).unwrap());
     Self {
-      id,
-      redis_client: Arc::new(Mutex::new(redis_client)),
+      lru_cache: Arc::new(Mutex::new(lru)),
     }
   }
 
   pub async fn get_encoded_collab(&self, object_id: &str) -> Option<EncodedCollab> {
-    let key = encoded_collab_key(self.id, object_id);
-    let result = self
-      .redis_client
-      .lock()
-      .await
-      .get::<_, Option<Vec<u8>>>(&key)
-      .await;
-    match result {
-      Ok(Some(bytes)) => match EncodedCollab::decode_from_bytes(&bytes) {
-        Ok(encoded_collab) => Some(encoded_collab),
-        Err(err) => {
-          error!("Failed to decode collab from redis cache bytes: {:?}", err);
-          None
-        },
-      },
-      Ok(None) => {
-        trace!(
-          "No encoded collab found in cache for object_id: {}",
-          object_id
-        );
-        None
-      },
+    let cache = self.lru_cache.lock().await.get(object_id)?.clone();
+    tokio::task::spawn_blocking(move || match EncodedCollab::decode_from_bytes(&cache) {
+      Ok(encoded_collab) => Some(encoded_collab),
       Err(err) => {
-        error!("Failed to get encoded collab from redis: {:?}", err);
+        error!("Failed to decode collab from redis cache bytes: {:?}", err);
         None
       },
-    }
+    })
+    .await
+    .ok()?
   }
 
-  pub async fn cache_encoded_collab(&self, object_id: &str, encoded_collab: &EncodedCollab) {
-    match encoded_collab.encode_to_bytes() {
+  pub fn cache_encoded_collab(&self, object_id: String, encoded_collab: &EncodedCollab) {
+    let encoded_collab = encoded_collab.clone();
+    let cache = self.lru_cache.clone();
+    tokio::task::spawn_blocking(move || match encoded_collab.encode_to_bytes() {
       Ok(bytes) => {
-        self.cache_encoded_collab_bytes(object_id, bytes).await;
+        tokio::spawn(async move { cache.lock().await.put(object_id, bytes) });
       },
       Err(e) => {
         error!("Failed to encode collab to bytes: {:?}", e);
       },
-    }
+    });
   }
 
-  pub async fn cache_encoded_collab_bytes(&self, object_id: &str, bytes: Vec<u8>) {
-    let key = encoded_collab_key(self.id, object_id);
-    if let Err(err) = self
-      .redis_client
-      .lock()
-      .await
-      .set::<_, Vec<u8>, ()>(&key, bytes)
-      .await
-    {
-      error!("Failed to cache encoded collab: {:?}", err);
-    }
+  pub fn cache_encoded_collab_bytes(&self, object_id: String, bytes: Vec<u8>) {
+    let cache = self.lru_cache.clone();
+    tokio::spawn(async move { cache.lock().await.put(object_id, bytes) });
   }
 }
-
-#[inline]
-fn encoded_collab_key(id: u32, object_id: &str) -> String {
-  format!("{}:{}:{}", id, ENCODED_COLLAB_KEY_PREFIX, object_id)
-}
diff --git a/src/biz/collab/ops.rs b/src/biz/collab/ops.rs
index 8d55c497..f9258b93 100644
--- a/src/biz/collab/ops.rs
+++ b/src/biz/collab/ops.rs
@@ -31,20 +31,24 @@ pub async fn create_collab_member(
 ) -> Result<(), AppError> {
   params.validate()?;
 
-  let mut txn = pg_pool
+  let mut transaction = pg_pool
     .begin()
     .await
     .context("acquire transaction to insert collab member")?;
 
-  if !database::collab::is_collab_exists(&params.object_id, txn.deref_mut()).await? {
+  if !database::collab::is_collab_exists(&params.object_id, transaction.deref_mut()).await? {
     return Err(AppError::RecordNotFound(format!(
       "Fail to insert collab member. The Collab with object_id {} does not exist",
       params.object_id
     )));
   }
 
-  if database::collab::is_collab_member_exists(params.uid, &params.object_id, txn.deref_mut())
-    .await?
+  if database::collab::is_collab_member_exists(
+    params.uid,
+    &params.object_id,
+    transaction.deref_mut(),
+  )
+  .await?
   {
     return Err(AppError::RecordAlreadyExists(format!(
       "Collab member with uid {} and object_id {} already exists",
@@ -57,11 +61,11 @@ pub async fn create_collab_member(
     params.uid,
     &params.object_id,
     &params.access_level,
-    &mut txn,
+    &mut transaction,
   )
   .await?;
 
-  txn
+  transaction
     .commit()
     .await
     .context("fail to commit the transaction to insert collab member")?;
@@ -74,12 +78,12 @@ pub async fn upsert_collab_member(
   params: &UpdateCollabMemberParams,
 ) -> Result<(), AppError> {
   params.validate()?;
-  let mut txn = pg_pool
+  let mut transaction = pg_pool
     .begin()
     .await
     .context("acquire transaction to upsert collab member")?;
 
-  if !database::collab::is_collab_exists(&params.object_id, txn.deref_mut()).await? {
+  if !database::collab::is_collab_exists(&params.object_id, transaction.deref_mut()).await? {
     return Err(AppError::RecordNotFound(format!(
       "Fail to upsert collab member. The Collab with object_id {} does not exist",
       params.object_id
@@ -90,11 +94,11 @@ pub async fn upsert_collab_member(
     params.uid,
     &params.object_id,
     &params.access_level,
-    &mut txn,
+    &mut transaction,
   )
   .await?;
 
-  txn
+  transaction
     .commit()
     .await
     .context("fail to commit the transaction to upsert collab member")?;
diff --git a/src/biz/collab/storage.rs b/src/biz/collab/storage.rs
index fbf58656..d7e5c667 100644
--- a/src/biz/collab/storage.rs
+++ b/src/biz/collab/storage.rs
@@ -140,7 +140,7 @@ where
       // If the collab already exists, check if the user has enough permissions to update collab
       let level = self
         .access_control
-        .get_collab_access_level(uid, &params.object_id, transaction.deref_mut())
+        .get_or_refresh_collab_access_level(uid, &params.object_id, transaction.deref_mut())
         .await
         .context(format!(
           "Can't find the access level when user:{} try to insert collab",
@@ -172,15 +172,17 @@ where
         uid, params.object_id
       )));
     }
-
-    self
-      .mem_cache
-      .cache_encoded_collab_bytes(&params.object_id, params.encoded_collab_v1.clone())
-      .await;
+    let object_id = params.object_id.clone();
+    let encoded_collab = params.encoded_collab_v1.clone();
     self
       .disk_cache
       .upsert_collab_with_transaction(workspace_id, uid, params, transaction)
-      .await
+      .await?;
+
+    self
+      .mem_cache
+      .cache_encoded_collab_bytes(object_id, encoded_collab);
+    Ok(())
   }
 
   async fn get_collab_encoded(
@@ -191,7 +193,7 @@ where
     params.validate()?;
     self
       .access_control
-      .get_collab_access_level(uid, &params.object_id, &self.disk_cache.pg_pool)
+      .get_or_refresh_collab_access_level(uid, &params.object_id, &self.disk_cache.pg_pool)
       .await?;
     let object_id = params.object_id.clone();
 
@@ -232,8 +234,7 @@ where
         let encoded_collab = self.disk_cache.get_collab_encoded(uid, params).await?;
         self
           .mem_cache
-          .cache_encoded_collab(&object_id, &encoded_collab)
-          .await;
+          .cache_encoded_collab(object_id, &encoded_collab);
         Ok(encoded_collab)
       },
     }
@@ -285,7 +286,7 @@ where
   async fn delete_collab(&self, uid: &i64, object_id: &str) -> DatabaseResult<()> {
     if !self
       .access_control
-      .get_collab_access_level(uid, object_id, &self.disk_cache.pg_pool)
+      .get_or_refresh_collab_access_level(uid, object_id, &self.disk_cache.pg_pool)
       .await
       .context(format!(
         "Can't find the access level when user:{} try to delete {}",