chore(server): rdb save can now save into tcp socket directly. (#317)

In more detail, RdbSaver uses AlignedBuffer that writes into io::Sink in chunks of 4KB. It's great for the direct file I/O, but bad for sockets that receive blocks of 4KB with garbage at the end. I improved the code around this and actually simplified the logic, so now AlignedBuffer is just another Sink that is passed into serializer when writing into files. When sending to sockets a socket sink is passed instead. Also many other unrelated changes grouped into this pretty big cr. 1. dashtable readability improvements. 2. Move methods from facade::ConnectionContext - into facade::Service, make ConnectionContext a dumb object. 3. Optionally allow journal to be memory only (not backed up by a disk) by using a ring buffer to store last k entries in each journal slice. Also renamed journal_shard into journal_slice because journal has presence in each DF thread and not only in its shards. 4. Introduce journal::Entry that will consolidate any store change that happens in the thread. 5. Introduce GetRandomHex utility function. 6. Introduce two hooks: ServerFamily::OnClose that is called when a connection is closed, and ServerFamily::BreakOnShutdown that is called when process exits and any background fibers neet to break early. 7. Pull some noisy info logs out of rdb_load class. 8. Snapshot class now has the ability to subscribe to journal changes, thus it can include concurrent changes into the snapshot. Currently only journal::Op::VAL is supported (it's part of RDB format anyway). Signed-off-by: Roman Gershman <roman@dragonflydb.io>

Author: romange

src/core/dash.h

@@ -716,12 +716,12 @@ template <typename U, typename V, typename EvictionPolicy>
 auto DashTable<_Key, _Value, Policy>::InsertInternal(U&& key, V&& value, EvictionPolicy& ev)
  -> std::pair<iterator, bool> {
  uint64_t key_hash = DoHash(key);
- uint32_t seg_id = SegmentId(key_hash);
+ uint32_t target_seg_id = SegmentId(key_hash);
 
  while (true) {
  // Keep last global_depth_ msb bits of the hash.
- assert(seg_id < segment_.size());
- SegmentType* target = segment_[seg_id];
+ assert(target_seg_id < segment_.size());
+ SegmentType* target = segment_[target_seg_id];
 
  // Load heap allocated segment data - to avoid TLB miss when accessing the bucket.
  __builtin_prefetch(target, 0, 1);
@@ -731,12 +731,12 @@ auto DashTable<_Key, _Value, Policy>::InsertInternal(U&& key, V&& value, Evictio
 
  if (res) { // success
  ++size_;
- return std::make_pair(iterator{this, seg_id, it.index, it.slot}, true);
+ return std::make_pair(iterator{this, target_seg_id, it.index, it.slot}, true);
  }
 
  /*duplicate insert, insertion failure*/
  if (it.found()) {
- return std::make_pair(iterator{this, seg_id, it.index, it.slot}, false);
+ return std::make_pair(iterator{this, target_seg_id, it.index, it.slot}, false);
  }
 
  // At this point we must split the segment.
@@ -749,12 +749,12 @@ auto DashTable<_Key, _Value, Policy>::InsertInternal(U&& key, V&& value, Evictio
  hotspot.key_hash = key_hash;
 
  for (unsigned j = 0; j < HotspotBuckets::kRegularBuckets; ++j) {
- hotspot.probes.by_type.regular_buckets[j] = bucket_iterator{this, seg_id, bid[j]};
+ hotspot.probes.by_type.regular_buckets[j] = bucket_iterator{this, target_seg_id, bid[j]};
  }
 
  for (unsigned i = 0; i < Policy::kStashBucketNum; ++i) {
  hotspot.probes.by_type.stash_buckets[i] =
- bucket_iterator{this, seg_id, uint8_t(kLogicalBucketNum + i), 0};
+ bucket_iterator{this, target_seg_id, uint8_t(kLogicalBucketNum + i), 0};
  }
  hotspot.num_buckets = HotspotBuckets::kNumBuckets;
 
@@ -770,7 +770,7 @@ auto DashTable<_Key, _Value, Policy>::InsertInternal(U&& key, V&& value, Evictio
  /*unsigned start = (bid[HotspotBuckets::kNumBuckets - 1] + 1) % kLogicalBucketNum;
  for (unsigned i = 0; i < HotspotBuckets::kNumBuckets; ++i) {
  uint8_t id = (start + i) % kLogicalBucketNum;
- buckets.probes.arr[i] = bucket_iterator{this, seg_id, id};
+ buckets.probes.arr[i] = bucket_iterator{this, target_seg_id, id};
  }
  garbage_collected_ += ev.GarbageCollect(buckets, this);
  */
@@ -804,12 +804,12 @@ auto DashTable<_Key, _Value, Policy>::InsertInternal(U&& key, V&& value, Evictio
  if (target->local_depth() == global_depth_) {
  IncreaseDepth(global_depth_ + 1);
 
- seg_id = SegmentId(key_hash);
- assert(seg_id < segment_.size() && segment_[seg_id] == target);
+ target_seg_id = SegmentId(key_hash);
+ assert(target_seg_id < segment_.size() && segment_[target_seg_id] == target);
  }
 
  ev.RecordSplit(target);
- Split(seg_id);
+ Split(target_seg_id);
  }
 
  return std::make_pair(iterator{}, false);

src/core/dash_internal.h

@@ -1220,6 +1220,8 @@ void Segment<Key, Value, Policy>::Split(HFunc&& hfn, Segment* dest_right) {
  auto it = dest_right->InsertUniq(std::forward<Key_t>(Key(bid, slot)),
  std::forward<Value_t>(Value(bid, slot)), hash);
  (void)it;
+ assert(it.index != kNanBid);
+
  if constexpr (USE_VERSION) {
  // Update the version in the destination bucket.
  uint64_t ver = stash.GetVersion();

src/facade/conn_context.h

@@ -17,8 +17,8 @@ class ConnectionContext {
  public:
  ConnectionContext(::io::Sink* stream, Connection* owner);
 
- // We won't have any virtual methods, probably. However, since we allocate derived class,
- // we need to declare a virtual d-tor so we could delete them inside Connection.
+ // We won't have any virtual methods, probably. However, since we allocate a derived class,
+ // we need to declare a virtual d-tor, so we could properly delete it from Connection code.
  virtual ~ConnectionContext() {}
 
  Connection* owner() {
@@ -51,10 +51,6 @@ class ConnectionContext {
  bool authenticated: 1;
  bool force_dispatch: 1; // whether we should route all requests to the dispatch fiber.
 
- virtual void OnClose() {}
-
- virtual std::string GetContextInfo() const { return std::string{}; }
-
  private:
  Connection* owner_;
  Protocol protocol_ = Protocol::REDIS;

src/facade/dragonfly_connection.cc

@@ -300,7 +300,7 @@ string Connection::GetClientInfo() const {
  absl::StrAppend(&res, " age=", now - creation_time_, " idle=", now - last_interaction_);
  absl::StrAppend(&res, " phase=", phase_, " ");
  if (cc_) {
- absl::StrAppend(&res, cc_->GetContextInfo());
+ absl::StrAppend(&res, service_->GetContextInfo(cc_.get()));
  }
 
  return res;
@@ -374,7 +374,7 @@ void Connection::ConnectionFlow(FiberSocketBase* peer) {
  VLOG(1) << "Before dispatch_fb.join()";
  dispatch_fb.join();
  VLOG(1) << "After dispatch_fb.join()";
- cc_->OnClose();
+ service_->OnClose(cc_.get());
 
  stats->read_buf_capacity -= io_buf_.Capacity();
 

src/facade/service_interface.h

@@ -32,6 +32,13 @@ class ServiceInterface {
 
  virtual void ConfigureHttpHandlers(util::HttpListenerBase* base) {
  }
+
+ virtual void OnClose(ConnectionContext* cntx) {
+ }
+
+ virtual std::string GetContextInfo(ConnectionContext* cntx) {
+ return {};
+ }
 };
 
 } // namespace facade

src/redis/redis_aux.h

@@ -27,7 +27,7 @@
 #define MAXMEMORY_NO_EVICTION (7<<8)
 
 
-#define CONFIG_RUN_ID_SIZE 40
+#define CONFIG_RUN_ID_SIZE 40U
 
 #define EVPOOL_CACHED_SDS_SIZE 255
 #define EVPOOL_SIZE 16

src/server/CMakeLists.txt

@@ -2,7 +2,7 @@ add_executable(dragonfly dfly_main.cc)
 cxx_link(dragonfly base dragonfly_lib)
 
 add_library(dfly_transaction db_slice.cc engine_shard_set.cc blocking_controller.cc common.cc
- io_mgr.cc journal/journal.cc journal/journal_shard.cc table.cc
+ io_mgr.cc journal/journal.cc journal/journal_slice.cc table.cc
  tiered_storage.cc transaction.cc)
 cxx_link(dfly_transaction uring_fiber_lib dfly_core strings_lib)
 

src/server/common.h

@@ -5,6 +5,7 @@
 #pragma once
 
 #include <absl/strings/ascii.h>
+#include <absl/strings/str_cat.h>
 #include <absl/types/span.h>
 
 #include <string_view>
@@ -131,4 +132,28 @@ extern unsigned kernel_version;
 
 const char* GlobalStateName(GlobalState gs);
 
+template <typename RandGen> std::string GetRandomHex(RandGen& gen, size_t len) {
+ static_assert(std::is_same<uint64_t, decltype(gen())>::value);
+ std::string res(len, '\0');
+ size_t indx = 0;
+
+ for (size_t i = 0; i < len / 16; ++i) { // 2 chars per byte
+ absl::AlphaNum an(absl::Hex(gen(), absl::kZeroPad16));
+
+ for (unsigned j = 0; j < 16; ++j) {
+ res[indx++] = an.Piece()[j];
+ }
+ }
+
+ if (indx < res.size()) {
+ absl::AlphaNum an(absl::Hex(gen(), absl::kZeroPad16));
+
+ for (unsigned j = 0; indx < res.size(); indx++, j++) {
+ res[indx] = an.Piece()[j];
+ }
+ }
+
+ return res;
+}
+
 } // namespace dfly

src/server/conn_context.cc

@@ -211,45 +211,6 @@ void ConnectionContext::SendSubscriptionChangedResponse(string_view action,
  (*this)->SendLong(count);
 }
 
-void ConnectionContext::OnClose() {
- if (!conn_state.exec_info.watched_keys.empty()) {
- shard_set->RunBriefInParallel([this](EngineShard* shard) {
- return shard->db_slice().UnregisterConnectionWatches(&conn_state.exec_info);
- });
- }
-
- if (!conn_state.subscribe_info)
- return;
-
- if (!conn_state.subscribe_info->channels.empty()) {
- auto token = conn_state.subscribe_info->borrow_token;
- UnsubscribeAll(false);
- // Check that all borrowers finished processing
- token.Wait();
- }
-
- if (conn_state.subscribe_info) {
- DCHECK(!conn_state.subscribe_info->patterns.empty());
- auto token = conn_state.subscribe_info->borrow_token;
- PUnsubscribeAll(false);
- // Check that all borrowers finished processing
- token.Wait();
- DCHECK(!conn_state.subscribe_info);
- }
-}
-
-string ConnectionContext::GetContextInfo() const {
- char buf[16] = {0};
- unsigned index = 0;
- if (async_dispatch)
- buf[index++] = 'a';
-
- if (conn_closing)
- buf[index++] = 't';
-
- return index ? absl::StrCat("flags:", buf) : string();
-}
-
 void ConnectionState::ExecInfo::Clear() {
  state = EXEC_INACTIVE;
  body.clear();

src/server/conn_context.h

@@ -83,8 +83,11 @@ struct ConnectionState {
  // For set op - it's the flag value we are storing along with the value.
  // For get op - we use it as a mask of MCGetMask values.
  uint32_t memcache_flag = 0;
- // If it's a replication client - then it holds positive sync session id.
- uint32_t sync_session_id = 0;
+
+ // If this server is master, and this connection is from a secondary replica,
+ // then it holds positive sync session id.
+ uint32_t repl_session_id = 0;
+ uint32_t repl_threadid = kuint32max;
 
  ExecInfo exec_info;
  std::optional<ScriptInfo> script_info;
@@ -97,8 +100,6 @@ class ConnectionContext : public facade::ConnectionContext {
  : facade::ConnectionContext(stream, owner) {
  }
 
- void OnClose() override;
-
  struct DebugInfo {
  uint32_t shards_count = 0;
  TxClock clock = 0;
@@ -123,8 +124,6 @@ class ConnectionContext : public facade::ConnectionContext {
 
  bool is_replicating = false;
 
- std::string GetContextInfo() const override;
-
  private:
  void SendSubscriptionChangedResponse(std::string_view action,
  std::optional<std::string_view> topic,