merge pingcap/release-3.1 until (tikv#6617) (#10)

* [3.1] external_storage: add S3 support (tikv#6209) (tikv#6536) * external_storage: add S3 support (tikv#6209) Signed-off-by: Yi Wu <[email protected]> Signed-off-by: kennytm <[email protected]> * lock_manager: more metrics (tikv#6392) (tikv#6422) (tikv#6444) Signed-off-by: youjiali1995 <[email protected]> * lock_manager: update default config (tikv#6426) (tikv#6429) (tikv#6446) Signed-off-by: youjiali1995 <[email protected]> * Update rocksdb and titan. Update rust-rocksdb url (tikv#6554) Signed-off-by: Yi Wu <[email protected]> * raft_client: connect to other stores using peer_address (tikv#5569) (tikv#6491) * *: pick threads statistics for regions (tikv#6605) Co-authored-by: jiyingtk <[email protected]> Co-authored-by: disksing <[email protected]> * fix test_replica_read::test_read_after_cleanup_range_for_snap (tikv#6493) Signed-off-by: 5kbpers <[email protected]> * deadlock: more solid role change observer (tikv#6415) (tikv#6431) (tikv#6445) Signed-off-by: youjiali1995 <[email protected]> * raftstore: read index message count metric (tikv#6579) (tikv#6610) Signed-off-by: 5kbpers <[email protected]> * raftstore: speed up conf change (tikv#6421) (tikv#6432) Signed-off-by: Jay Lee <[email protected]> * raftstore: fix a panic in read index queue (tikv#6609) (tikv#6613) Signed-off-by: qupeng <[email protected]> * pick tikv#6414, tikv#6487, tikv#6477 and tikv#6539 to release-3.1 (tikv#6564) * raftstore: extract batch system (tikv#6414) * apply: support yield (tikv#6487) * batch-system: add benchmark (tikv#6477) * Introduce pre transfer leader (tikv#6539) Signed-off-by: Jay Lee <[email protected]> * raftstore: set wait_merge_state to none after resuming pending state (tikv#6621) Signed-off-by: Liqi Geng <[email protected]> * raftstore: learner load merge target & fix a merge network recovery bug (tikv#6623) Signed-off-by: Liqi Geng <[email protected]> * change merge flow path (tikv#6617) Signed-off-by: Liqi Geng <[email protected]> Co-authored-by: kennytm <[email protected]> Co-authored-by: Lei Zhao <[email protected]> Co-authored-by: yiwu-arbug <[email protected]> Co-authored-by: disksing <[email protected]> Co-authored-by: ShuNing <[email protected]> Co-authored-by: jiyingtk <[email protected]> Co-authored-by: pingcap-github-bot <[email protected]> Co-authored-by: 5kbpers <[email protected]> Co-authored-by: Jay <[email protected]> Co-authored-by: gengliqi <[email protected]>
pingcap · Feb 15, 2020 · 16f7f7a · 16f7f7a
1 parent 7883f74
commit 16f7f7a
Show file tree

Hide file tree

Showing 64 changed files with 5,713 additions and 3,165 deletions.
diff --git a/Cargo.lock b/Cargo.lock
diff --git a/Cargo.toml b/Cargo.toml
@@ -105,6 +105,7 @@ engine = { path = "components/engine" }
 tikv_util = { path = "components/tikv_util" }
 farmhash = "1.1.5"
 external_storage = { path = "components/external_storage" }
+batch-system = { path = "components/batch-system" }
 
 [dependencies.murmur3]
 git = "https://github.com/pingcap/murmur3.git"
@@ -122,7 +123,6 @@ default-features = false
 [replace]
 "log:0.3.9" = { git = "https://github.com/busyjay/log", branch = "use-static-module" }
 "log:0.4.6" = { git = "https://github.com/busyjay/log", branch = "revert-to-static" }
-"https://github.com/pingcap/kvproto.git#kvproto:0.0.1" = { git = "https://github.com/solotzg/kvproto.git", branch = "learner-merge" }
 "prometheus:0.4.2" = { git = "https://github.com/birdstorm/rust-prometheus", branch = "0.4.2" }
 
 [dev-dependencies]

diff --git a/components/backup/src/writer.rs b/components/backup/src/writer.rs
@@ -69,14 +69,18 @@ impl Writer {
             .observe(sst_info.file_size() as f64);
         let file_name = format!("{}_{}.sst", name, cf);
 
-        let reader = Sha256Reader::new(sst_reader)
+        let (reader, hasher) = Sha256Reader::new(sst_reader)
             .map_err(|e| Error::Other(box_err!("Sha256 error: {:?}", e)))?;
-        let mut reader = limiter.limit(AllowStdIo::new(reader));
-        storage.write(&file_name, &mut reader)?;
-        let sha256 = reader
-            .into_inner()
-            .into_inner()
-            .hash()
+        storage.write(
+            &file_name,
+            Box::new(limiter.limit(AllowStdIo::new(reader))),
+            sst_info.file_size(),
+        )?;
+        let sha256 = hasher
+            .lock()
+            .unwrap()
+            .finish()
+            .map(|digest| digest.to_vec())
             .map_err(|e| Error::Other(box_err!("Sha256 error: {:?}", e)))?;
 
         let mut file = File::new();

diff --git a/components/batch-system/Cargo.toml b/components/batch-system/Cargo.toml
@@ -0,0 +1,29 @@
+[package]
+name = "batch-system"
+version = "0.1.0"
+edition = "2018"
+
+[dependencies]
+crossbeam = "0.7"
+tikv_util = { path = "../tikv_util" }
+slog = { version = "2.3", features = ["max_level_trace", "release_max_level_debug"] }
+slog-global = { version = "0.1", git = "https://github.com/breeswish/slog-global.git", rev = "91904ade" }
+derive_more = "0.15.0"
+
+[dev-dependencies]
+criterion = "0.3"
+
+[[test]]
+name = "tests"
+path = "tests/cases/mod.rs"
+required-features = ["test-runner"]
+
+[[bench]]
+name = "router"
+harness = false
+required-features = ["test-runner"]
+
+[[bench]]
+name = "batch-system"
+harness = false
+required-features = ["test-runner"]
diff --git a/components/batch-system/benches/batch-system.rs b/components/batch-system/benches/batch-system.rs
@@ -0,0 +1,142 @@
+// Copyright 2020 TiKV Project Authors. Licensed under Apache-2.0.
+
+#![feature(test)]
+
+extern crate test;
+
+use batch_system::test_runner::*;
+use batch_system::*;
+use criterion::*;
+use std::sync::atomic::*;
+use std::sync::Arc;
+
+fn end_hook(tx: &std::sync::mpsc::Sender<()>) -> Message {
+    let tx = tx.clone();
+    Message::Callback(Box::new(move |_| {
+        tx.send(()).unwrap();
+    }))
+}
+
+/// Benches how it performs when many messages are sent to the bench system.
+///
+/// A better router and lightweight batch scheduling can lead to better result.
+fn bench_spawn_many(c: &mut Criterion) {
+    let (control_tx, control_fsm) = Runner::new(100000);
+    let (router, mut system) = batch_system::create_system(2, 2, control_tx, control_fsm);
+    system.spawn("test".to_owned(), Builder::new());
+    const ID_LIMIT: u64 = 32;
+    const MESSAGE_LIMIT: usize = 256;
+    for id in 0..ID_LIMIT {
+        let (normal_tx, normal_fsm) = Runner::new(100000);
+        let normal_box = BasicMailbox::new(normal_tx, normal_fsm);
+        router.register(id, normal_box);
+    }
+
+    let (tx, rx) = std::sync::mpsc::channel();
+    c.bench_function("spawn_many", |b| {
+        b.iter(|| {
+            for id in 0..ID_LIMIT {
+                for i in 0..MESSAGE_LIMIT {
+                    router.send(id, Message::Loop(i)).unwrap();
+                }
+                router.send(id, end_hook(&tx)).unwrap();
+            }
+            for _ in 0..ID_LIMIT {
+                rx.recv().unwrap();
+            }
+        })
+    });
+    system.shutdown();
+}
+
+/// Bench how it performs if two hot FSMs are shown up at the same time.
+///
+/// A good scheduling algorithm should be able to spread the hot FSMs to
+/// all available threads as soon as possible.
+fn bench_imbalance(c: &mut Criterion) {
+    let (control_tx, control_fsm) = Runner::new(100000);
+    let (router, mut system) = batch_system::create_system(2, 2, control_tx, control_fsm);
+    system.spawn("test".to_owned(), Builder::new());
+    const ID_LIMIT: u64 = 10;
+    const MESSAGE_LIMIT: usize = 512;
+    for id in 0..ID_LIMIT {
+        let (normal_tx, normal_fsm) = Runner::new(100000);
+        let normal_box = BasicMailbox::new(normal_tx, normal_fsm);
+        router.register(id, normal_box);
+    }
+
+    let (tx, rx) = std::sync::mpsc::channel();
+    c.bench_function("imbalance", |b| {
+        b.iter(|| {
+            for i in 0..MESSAGE_LIMIT {
+                for id in 0..2 {
+                    router.send(id, Message::Loop(i)).unwrap();
+                }
+            }
+            for id in 0..2 {
+                router.send(id, end_hook(&tx)).unwrap();
+            }
+            for _ in 0..2 {
+                rx.recv().unwrap();
+            }
+        })
+    });
+    system.shutdown();
+}
+
+/// Bench how it performs when scheduling a lot of quick tasks during an long-polling
+/// tasks.
+///
+/// A good scheduling algorithm should not starve the quick tasks.
+fn bench_fairness(c: &mut Criterion) {
+    let (control_tx, control_fsm) = Runner::new(100000);
+    let (router, mut system) = batch_system::create_system(2, 2, control_tx, control_fsm);
+    system.spawn("test".to_owned(), Builder::new());
+    for id in 0..10 {
+        let (normal_tx, normal_fsm) = Runner::new(100000);
+        let normal_box = BasicMailbox::new(normal_tx, normal_fsm);
+        router.register(id, normal_box);
+    }
+
+    let (tx, _rx) = std::sync::mpsc::channel();
+    let running = Arc::new(AtomicBool::new(true));
+    let router1 = router.clone();
+    let running1 = running.clone();
+    let handle = std::thread::spawn(move || {
+        while running1.load(Ordering::SeqCst) {
+            // Using 4 to ensure all worker threads are busy spinning.
+            for id in 0..4 {
+                let _ = router1.send(id, Message::Loop(16));
+            }
+        }
+        tx.send(()).unwrap();
+    });
+
+    let (tx2, rx2) = std::sync::mpsc::channel();
+    c.bench_function("fairness", |b| {
+        b.iter(|| {
+            for _ in 0..10 {
+                for id in 4..6 {
+                    router.send(id, Message::Loop(10)).unwrap();
+                }
+            }
+            for id in 4..6 {
+                router.send(id, end_hook(&tx2)).unwrap();
+            }
+            for _ in 4..6 {
+                rx2.recv().unwrap();
+            }
+        })
+    });
+    running.store(false, Ordering::SeqCst);
+    system.shutdown();
+    let _ = handle.join();
+}
+
+criterion_group!(fair, bench_fairness);
+criterion_group!(
+    name = load;
+    config = Criterion::default().sample_size(30);
+    targets = bench_imbalance, bench_spawn_many
+);
+criterion_main!(fair, load);
diff --git a/components/batch-system/benches/router.rs b/components/batch-system/benches/router.rs
@@ -0,0 +1,24 @@
+// Copyright 2020 TiKV Project Authors. Licensed under Apache-2.0.
+
+use batch_system::test_runner::*;
+use batch_system::*;
+use criterion::*;
+
+fn bench_send(c: &mut Criterion) {
+    let (control_tx, control_fsm) = Runner::new(100000);
+    let (router, mut system) = batch_system::create_system(2, 2, control_tx, control_fsm);
+    system.spawn("test".to_owned(), Builder::new());
+    let (normal_tx, normal_fsm) = Runner::new(100000);
+    let normal_box = BasicMailbox::new(normal_tx, normal_fsm);
+    router.register(1, normal_box);
+
+    c.bench_function("router::send", |b| {
+        b.iter(|| {
+            router.send(1, Message::Loop(0)).unwrap();
+        })
+    });
+    system.shutdown();
+}
+
+criterion_group!(benches, bench_send);
+criterion_main!(benches);