chore: missing 0.4.9 changelog, de-flake timing tests, small cleanups

CHANGELOG.md

@@ -1,5 +1,37 @@
 # changelog
 
+## 0.4.9 (2026-02-27)
+
+### features
+- bitmap commands: `GETBIT`, `SETBIT`, `BITCOUNT`, `BITPOS`, `BITOP` (#316, #330)
+- `LMPOP` / `ZMPOP` multi-key pops (#317)
+- `EXPIREAT`, `PEXPIREAT`, `EXPIRETIME`, `PEXPIRETIME`, `GETSET`, `MSETNX`, `SMOVE`, `SINTERCARD`, count args for `LPOP`/`RPOP` (#315, #318)
+- `HRANDFIELD` and `ZRANDMEMBER` (#319)
+- Redis 6.2+ commands: `LMOVE`, `GETDEL`, `GETEX`, `ZDIFF`, `ZINTER`, `ZUNION` (#299)
+- `COMMAND`, `HINCRBYFLOAT`, `ZDIFFSTORE`/`ZINTERSTORE`/`ZUNIONSTORE`, `FLUSHALL`, `MEMORY USAGE`, `WAIT` (#312, #324, #331)
+- keyspace notifications (`__keyevent@0__:expired` and write events) (#313)
+- automatic snapshot scheduling (#300)
+- `CONFIG SET` live-apply for maxmemory and maxmemory-policy (#312)
+- ember-client: typed async RESP3 API with pipelining; CLI now reuses it (#297, #302, #303)
+- ember-ts client with full API parity; npm package renamed to `emberdb` with OIDC trusted publishing (#308, #309, #332)
+- grpc: 39 new rpcs closing the RESP/gRPC api gap; 17 new commands surfaced across cli, grpc, and clients (#306, #320)
+- CLI watch mode and batch mode (#298)
+
+### fixes
+- BITOP cross-shard correctness (#330)
+- cluster reports `cluster_state:ok` on bootstrap (#339)
+- eliminated production panics in client decoder and server startup (#322)
+- auth failure metrics, command memory budget, migration progress reporting (#323)
+- replication send-failure counter; go client subscribe error propagation (#325)
+- raft log persistence switched from bincode to postcard; dropped atomic-polyfill (#296, #329)
+- prometheus counters, crash recovery hardening, TLS tests (#314)
+- go client: regenerated protobuf stubs for the 17 new rpcs (#337)
+
+### docs
+- command count updated to 190+; compatibility guide refreshed (#335)
+
+---
+
 ## 0.4.8 (2026-02-25)
 
 ### performance
@@ -13,7 +45,7 @@
 - rate-limited ENOSPC handling for AOF writes (#249)
 
 ### docs
-- removed concurrent mode references — sharded is now the only execution mode (#275, #291)
+- pruned concurrent mode from the docs (#275, #291) — correction: the mode itself was not removed and is still available behind `--concurrent`; sharded remains the default execution mode
 - refreshed all benchmark numbers from 2026-02-25 GCP run (#288-290)
 - added documentation section, code of conduct, performance tuning guide, production checklist (#270, #281-282)
 

crates/ember-core/src/shard/blocking.rs

@@ -180,12 +180,10 @@ mod tests {
             })
             .await;
 
-        // give the shard a moment to process
-        tokio::time::sleep(Duration::from_millis(50)).await;
-
-        let result = rx.try_recv();
-        assert!(result.is_ok());
-        let (key, data) = result.unwrap();
+        // recv with a generous timeout instead of a fixed sleep so a busy
+        // runner can't flake the test
+        let result = tokio::time::timeout(Duration::from_secs(5), rx.recv()).await;
+        let (key, data) = result.expect("timed out waiting for BLPop").unwrap();
         assert_eq!(key, "mylist");
         assert_eq!(data, Bytes::from("hello"));
     }
@@ -267,8 +265,8 @@ mod tests {
             })
             .await;
 
-        tokio::time::sleep(Duration::from_millis(50)).await;
-        let (key, data) = rx.try_recv().unwrap();
+        let result = tokio::time::timeout(Duration::from_secs(5), rx.recv()).await;
+        let (key, data) = result.expect("timed out waiting for BRPop").unwrap();
         assert_eq!(key, "mylist");
         assert_eq!(data, Bytes::from("b"));
     }

crates/ember-core/src/shard/persistence.rs

@@ -429,8 +429,9 @@ mod tests {
             // drop handle to shut down shard
         }
 
-        // give it a moment to flush
-        tokio::time::sleep(Duration::from_millis(50)).await;
+        // the final AOF sync runs in the shard task after the handle drops;
+        // wait generously so a loaded runner can't observe a partial flush
+        tokio::time::sleep(Duration::from_millis(500)).await;
 
         // start a new shard with the same config — should recover
         {

crates/ember-core/src/types/sorted_set.rs

@@ -93,6 +93,9 @@ impl SortedSet {
 
     /// Adds or updates a member with ZADD flag semantics.
     pub fn add_with_flags(&mut self, member: &str, score: f64, flags: &ZAddFlags) -> AddResult {
+        // the unwraps below rely on the score map and sorted vec staying in
+        // lockstep; catch any desync early in debug/test builds
+        debug_assert_eq!(self.scores.len(), self.sorted.len());
         let new_score = OrderedFloat(score);
 
         if let Some(&old_score) = self.scores.get(member) {
@@ -137,6 +140,7 @@ impl SortedSet {
 
     /// Removes a member from the sorted set. Returns `true` if it existed.
     pub fn remove(&mut self, member: &str) -> bool {
+        debug_assert_eq!(self.scores.len(), self.sorted.len());
         if let Some((name, score)) = self.scores.remove_entry(member) {
             let idx = self.search_idx(score, &name).unwrap();
             self.sorted.remove(idx);

crates/ember-server/src/connection/exec/strings.rs

@@ -343,7 +343,9 @@ pub(in crate::connection) async fn msetnx(pairs: Vec<(String, Bytes)>, cx: &Exec
     // but matches Redis cluster semantics where MSETNX pairs must
     // share a hash slot. For single-node mode it is correct.
     if pairs.is_empty() {
-        return Frame::Error("ERR wrong number of arguments for 'MSETNX'".into());
+        // unreachable in practice (the parser enforces arity); message kept
+        // in the canonical WrongArity format for consistency
+        return Frame::Error("ERR wrong number of arguments for 'MSETNX' command".into());
     }
 
     let keys: Vec<String> = pairs.iter().map(|(k, _)| k.clone()).collect();