kino 0.6.0 → 0.7.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 591ec1716d7d8f23c1483ba2db85f517c42da5de78343c938a27d3e00ca1a66a
-  data.tar.gz: 91a95efe00e8193be3792c4e2b8055f1b45e2b1a4dcb6acaa1a1a4525300f878
+  metadata.gz: 03f08b02b92df6b069d24b28d3abfba6147b2aa6ab854034a6e7619dd1eb771c
+  data.tar.gz: 945c95b779ea6c2729d514657b62e7a7014c30b28965122f6a62b7fb2f27ca20
 SHA512:
-  metadata.gz: ed8d00148990ce34bd73308b376f342b011de37ccf39a078a8676183b0689ad1762af4dc038d6baf2b0501cdb8556a5448e37e0f7ab0c8c755d4e719ceca62c6
-  data.tar.gz: 8305703e2c93bfbe4a00944a4773213f0414459bfdc2e85f963736767a382d85437889d9b216a845df2e314c4d339e0ddd40aebb5b6689b413c16cdf04e4cc87
+  metadata.gz: 631c65a77e3cae89796e2a7465ed11fbcec652b9bf78fe325d18f77b53f7c26d3922770cf1f42b6b6d3bf07e70ed53dbd639af4e71ba053e77b85508bc6825b4
+  data.tar.gz: 7cc673c49dcdefbb7e1f7343a9058426339d2b983507ced5e72aa573106c91219608761f84cc8dd06889292ca8761a88f85e901d49be70eb98afca6707340385

data/CHANGELOG.md

@@ -1,3 +1,24 @@
+## [0.7.0] - 2026-09-08
+
+- Experimental elastic worker pool. Set `max_workers` above `workers`
+  and the pool grows under load, one worker at a time, then shrinks
+  back to `workers` once the extra workers have sat idle for
+  `scale_down_after` seconds (default 30). Works in both modes; a
+  retiring worker finishes its request first. `stats`, `/stats` and
+  `/metrics` gain `max_workers`, `active_workers`, `scale_ups` and
+  `scale_downs`, and each `worker_status` row gains `retired`. Leave
+  `max_workers` unset and the pool is fixed, as before.
+- Ractor mode warns at boot when `max_workers` (or `workers`) exceeds
+  `RUBY_MAX_CPU` (default 8), Ruby's cap on how many ractors run Ruby
+  code at once. Set the variable to your worker count to lift it.
+- Ractor-readiness fixes, so external Ractor audits of Kino pass: the
+  env string caches root their strings through the lock-free pin slab
+  instead of per-value GC registration (unsynchronized across ractors
+  in Ruby 4.0, a crash) and no longer call into Ruby while locked (a
+  GC-barrier deadlock); the shared rack.errors and rack.input
+  singletons must be Ractor-shareable, not just frozen; the Rack
+  handler's option table is shareable. Throughput is unchanged.
+
 ## [0.6.0] - 2026-09-01
 
 - HTTP/2 support. Kino now speaks HTTP/2 natively, on by default:

data/Cargo.lock

@@ -444,7 +444,7 @@ dependencies = [
 
 [[package]]
 name = "kino"
-version = "0.6.0"
+version = "0.7.0"
 dependencies = [
  "ahash",
  "bytes",

data/README.md

@@ -259,6 +259,8 @@ server = Kino::Server.new(app,
   bind: "127.0.0.1",          # or "unix:///run/kino.sock" behind a proxy
   port: 9292,                 # 0 = ephemeral; read back via server.port
   workers: Kino.available_parallelism, # ractors (parallelism); the default
+  max_workers: nil,           # experimental: grow past workers under load (see Elastic pool)
+  scale_down_after: 30,       # seconds idle before an extra worker retires
   threads: 1,                 # per worker; ractor default 1, threaded default 3
   mode: :auto,                # :auto | :ractor | :threaded
   queue_depth: 1024,          # bounded queue; overflow → 503
@@ -418,6 +420,56 @@ Kino fires four lifecycle hooks alongside `on_error`, split by firing context.
 
 A raising hook is logged and never kills a worker.
 
+## Elastic pool (experimental)
+
+Size the pool for the quiet hours and let it grow for the busy ones:
+
+```ruby
+# kino.rb
+workers 4            # always running
+max_workers 16       # reached only under load
+scale_down_after 30  # seconds idle before an extra worker retires
+```
+
+Or `Kino::Server.new(app, workers: 4, max_workers: 16)`. Leave
+`max_workers` unset and the pool is fixed at `workers`, as before.
+
+While requests wait in the queue, Kino adds a worker every 100 ms until
+the queue clears or the pool hits `max_workers`. When the load passes,
+workers above `workers` retire one at a time after `scale_down_after`
+seconds idle, each finishing its current request first. Same behavior
+in `:ractor` and `:threaded` mode.
+
+**Use it when your app waits**: on databases, upstream services, slow
+clients. With `workers` at your core count, all workers can be blocked
+on I/O while cores sit idle; a higher ceiling puts those cores to work,
+and a ractor starts in microseconds, so the pool follows load closely.
+Pure CPU work gains nothing past the core count. In `:ractor` mode Ruby
+itself runs at most `RUBY_MAX_CPU` ractors' Ruby code at once (default
+8), and Kino warns at boot when the pool can exceed it. On a bigger box:
+
+```sh
+RUBY_MAX_CPU=16 kino
+```
+
+**Watch it breathe** in `server.stats`, `GET /stats` and `GET /metrics`:
+
+```sh
+$ curl -s localhost:9293/stats | jq '{workers, max_workers, active_workers, scale_ups, scale_downs}'
+{
+  "workers": 4,
+  "max_workers": 16,
+  "active_workers": 9,
+  "scale_ups": 12,
+  "scale_downs": 7
+}
+```
+
+Prometheus gets `kino_max_workers`, `kino_active_workers`,
+`kino_scale_ups_total` and `kino_scale_downs_total`. `after_worker_boot`
+fires for every worker the pool adds, `on_worker_exit` (with a nil
+cause) for every one it retires.
+
 ## Stuck-worker quarantine
 
 `quarantine_timeout: seconds` (or `quarantine_timeout 60` in `kino.rb`)

data/doc/architecture.md

@@ -31,6 +31,16 @@ Puma-style two-level: `workers × threads`.
 - Identical machinery either way: the flume queue is MPMC, a "worker slot"
   is per-thread, and the worker loop (`lib/kino/worker.rb`) is shared
   verbatim.
+- Elastic pool (`max_workers`): a scaler thread on the main ractor
+  samples queue depth and the per-slot sensors every 100 ms, adds one
+  worker per sample while requests wait, and retires the longest-idle
+  worker above the floor after `scale_down_after`. Retirement is a
+  per-slot flag raised under the slot's lane lock: the lane dispatcher
+  skips the slot, the take loop honors the flag at its next idle tick (a
+  request already taken finishes first, a lane worker drains its own
+  lane), and the slot is reset and reused by the next worker, so the
+  slot table never grows with churn. Both pools (the ractor supervisor
+  and the threaded pool) expose the same grow/retire/groups seam.
 - Experimental `lanes true` replaces the one shared queue with a small
   private queue per worker slot (awake-preferring dispatch, work
   stealing); see [benchmarks](benchmarks.md#lane-dispatch-experimental-lanes-true).

data/ext/kino/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "kino"
-version = "0.6.0"
+version = "0.7.0"
 edition = "2021"
 authors = ["Yaroslav Markin <yaroslav@markin.net>"]
 license = "MIT"

data/ext/kino/src/control.rs

@@ -18,6 +18,8 @@ pub struct WorkerStat {
     pub in_flight: usize,
     pub busy_ms: u64,
     pub quarantined: bool,
+    /// Sent home by the pool scaler; leaves at its next idle tick.
+    pub retired: bool,
 }
 
 /// Read every slot's per-worker sensors in one pass under the slots read
@@ -46,6 +48,7 @@ pub fn collect_worker_status(server: &ServerInner) -> Vec<WorkerStat> {
                     now.saturating_sub(started)
                 },
                 quarantined,
+                retired: slot.retired.load(Ordering::Relaxed),
             }
         })
         .collect()
@@ -70,6 +73,10 @@ pub struct StatsSnapshot {
     pub worker_status: Vec<WorkerStat>,
     pub quarantined_count: usize,
     pub quarantine_replacements: u64,
+    pub max_workers: usize,
+    pub active_workers: usize,
+    pub scale_ups: u64,
+    pub scale_downs: u64,
     pub queue_histogram: crate::registry::QueueHistogramSnapshot,
 }
 
@@ -94,6 +101,10 @@ impl StatsSnapshot {
             worker_status,
             quarantined_count,
             quarantine_replacements: server.quarantine_replacements.load(Ordering::Relaxed),
+            max_workers: server.topology.max_workers,
+            active_workers: server.active_workers.load(Ordering::Relaxed),
+            scale_ups: server.scale_ups.load(Ordering::Relaxed),
+            scale_downs: server.scale_downs.load(Ordering::Relaxed),
             queue_histogram: server.queue_histogram.snapshot(),
         }
     }
@@ -114,9 +125,10 @@ pub fn stats_json(s: &StatsSnapshot) -> String {
     let mut out = String::with_capacity(256);
     write!(
         out,
-        r#"{{"mode":"{}","lanes":{},"workers":{},"threads":{},"batch":{},"respawns":{},"queued":{},"in_flight":{},"served":{},"rejected":{},"timeouts":{}"#,
+        r#"{{"mode":"{}","lanes":{},"workers":{},"threads":{},"batch":{},"respawns":{},"queued":{},"in_flight":{},"served":{},"rejected":{},"timeouts":{},"max_workers":{},"active_workers":{},"scale_ups":{},"scale_downs":{}"#,
         s.mode, s.lanes, s.workers, s.threads, s.batch, s.respawns,
-        s.queued, s.in_flight, s.served, s.rejected, s.timeouts
+        s.queued, s.in_flight, s.served, s.rejected, s.timeouts,
+        s.max_workers, s.active_workers, s.scale_ups, s.scale_downs
     )
     .expect("writing to a String cannot fail");
     if let Some(depths) = &s.lane_depths {
@@ -134,8 +146,8 @@ pub fn stats_json(s: &StatsSnapshot) -> String {
         }
         write!(
             out,
-            r#"{{"index":{},"served":{},"in_flight":{},"busy_ms":{},"quarantined":{}}}"#,
-            w.index, w.served, w.in_flight, w.busy_ms, w.quarantined
+            r#"{{"index":{},"served":{},"in_flight":{},"busy_ms":{},"quarantined":{},"retired":{}}}"#,
+            w.index, w.served, w.in_flight, w.busy_ms, w.quarantined, w.retired
         )
         .expect("writing to a String cannot fail");
     }
@@ -247,6 +259,34 @@ pub fn metrics_text(s: &StatsSnapshot) -> String {
         "Configured threads per worker.",
         s.threads,
     );
+    metric(
+        &mut out,
+        "kino_max_workers",
+        "gauge",
+        "Worker pool ceiling (equals kino_workers for a fixed pool).",
+        s.max_workers,
+    );
+    metric(
+        &mut out,
+        "kino_active_workers",
+        "gauge",
+        "Workers currently alive and serving.",
+        s.active_workers,
+    );
+    metric(
+        &mut out,
+        "kino_scale_ups_total",
+        "counter",
+        "Workers added by the pool scaler.",
+        s.scale_ups,
+    );
+    metric(
+        &mut out,
+        "kino_scale_downs_total",
+        "counter",
+        "Idle workers retired by the pool scaler.",
+        s.scale_downs,
+    );
     metric(
         &mut out,
         "kino_ready",
@@ -641,6 +681,10 @@ mod tests {
             worker_status: vec![],
             quarantined_count: 0,
             quarantine_replacements: 0,
+            max_workers: 32,
+            active_workers: 12,
+            scale_ups: 3,
+            scale_downs: 1,
             queue_histogram: crate::registry::QueueHistogramSnapshot {
                 buckets: [0; crate::registry::QUEUE_BOUNDS_US.len()],
                 overflow: 0,
@@ -665,6 +709,10 @@ mod tests {
             r#""served":100"#,
             r#""rejected":5"#,
             r#""timeouts":6"#,
+            r#""max_workers":32"#,
+            r#""active_workers":12"#,
+            r#""scale_ups":3"#,
+            r#""scale_downs":1"#,
             r#""state":"ready""#,
             r#""version":""#,
         ] {
@@ -690,6 +738,49 @@ mod tests {
         assert!(draining.contains("kino_ready 0"));
     }
 
+    #[test]
+    fn metrics_text_reports_the_elastic_pool() {
+        let text = metrics_text(&snapshot(crate::registry::STATE_READY));
+        assert!(text.contains("# TYPE kino_max_workers gauge"));
+        assert!(text.contains("kino_max_workers 32"));
+        assert!(text.contains("# TYPE kino_active_workers gauge"));
+        assert!(text.contains("kino_active_workers 12"));
+        assert!(text.contains("# TYPE kino_scale_ups_total counter"));
+        assert!(text.contains("kino_scale_ups_total 3"));
+        assert!(text.contains("# TYPE kino_scale_downs_total counter"));
+        assert!(text.contains("kino_scale_downs_total 1"));
+    }
+
+    #[test]
+    fn worker_status_reports_retired_slots() {
+        let server = crate::registry::test_server(false, 4);
+        server.register_worker();
+        server.register_worker();
+        server.slots.read()[1].retire();
+
+        let status = collect_worker_status(&server);
+
+        assert_eq!(
+            status.iter().map(|w| w.retired).collect::<Vec<_>>(),
+            vec![false, true]
+        );
+        assert!(status.iter().all(|w| w.busy_ms == 0));
+    }
+
+    #[test]
+    fn snapshot_reads_the_pool_counters() {
+        let server = crate::registry::test_server(false, 4);
+        server.active_workers.store(3, Ordering::Relaxed);
+        server.scale_ups.fetch_add(2, Ordering::Relaxed);
+        server.scale_downs.fetch_add(1, Ordering::Relaxed);
+
+        let s = StatsSnapshot::take(&server);
+
+        assert_eq!(s.active_workers, 3);
+        assert_eq!(s.max_workers, server.topology.max_workers);
+        assert_eq!((s.scale_ups, s.scale_downs), (2, 1));
+    }
+
     #[test]
     fn metrics_text_includes_lane_depth_samples_when_lanes_are_on() {
         let mut s = snapshot(crate::registry::STATE_READY);
@@ -757,6 +848,7 @@ mod tests {
                 in_flight: 1,
                 busy_ms: 4,
                 quarantined: false,
+                retired: false,
             },
             WorkerStat {
                 index: 1,
@@ -764,10 +856,11 @@ mod tests {
                 in_flight: 0,
                 busy_ms: 0,
                 quarantined: false,
+                retired: true,
             },
         ];
         let json = stats_json(&s);
-        assert!(json.contains(r#""worker_status":[{"index":0,"served":10,"in_flight":1,"busy_ms":4,"quarantined":false},{"index":1,"served":7,"in_flight":0,"busy_ms":0,"quarantined":false}]"#), "got {json}");
+        assert!(json.contains(r#""worker_status":[{"index":0,"served":10,"in_flight":1,"busy_ms":4,"quarantined":false,"retired":false},{"index":1,"served":7,"in_flight":0,"busy_ms":0,"quarantined":false,"retired":true}]"#), "got {json}");
     }
 
     #[test]
@@ -786,6 +879,7 @@ mod tests {
                 in_flight: 1,
                 busy_ms: 4,
                 quarantined: false,
+                retired: false,
             },
             WorkerStat {
                 index: 1,
@@ -793,6 +887,7 @@ mod tests {
                 in_flight: 0,
                 busy_ms: 0,
                 quarantined: false,
+                retired: true,
             },
         ];
         let text = metrics_text(&s);
@@ -835,6 +930,7 @@ mod tests {
                 in_flight: 1,
                 busy_ms: 0,
                 quarantined: true,
+                retired: false,
             },
             WorkerStat {
                 index: 1,
@@ -842,6 +938,7 @@ mod tests {
                 in_flight: 1,
                 busy_ms: 5,
                 quarantined: false,
+                retired: false,
             },
         ];
         let json = stats_json(&s);
@@ -850,7 +947,7 @@ mod tests {
             "top-level count: {json}"
         );
         assert!(
-            json.contains(r#"{"index":0,"served":3,"in_flight":1,"busy_ms":0,"quarantined":true}"#),
+            json.contains(r#"{"index":0,"served":3,"in_flight":1,"busy_ms":0,"quarantined":true,"retired":false}"#),
             "{json}"
         );
         assert!(json.contains(r#""quarantined":false"#));