From dd952c3c52378083f62ce78281c29512fe007ef2 Mon Sep 17 00:00:00 2001
From: Roman Gershman <roman@dragonflydb.io>
Date: Sun, 19 Feb 2023 21:20:14 +0200
Subject: [PATCH] feat(server): add debug information about running
 transactions (#820)

Specifically, provide a total length over all shards of pending txs.
Also, provide number of "free" transactions pending in the queue -
those that could progress immediately but do not because
they are not first in the queue.

Signed-off-by: Roman Gershman <roman@dragonflydb.io>
---
 src/server/debugcmd.cc | 61 ++++++++++++++++++++++++++++++++++++++++++
 src/server/debugcmd.h  |  1 +
 2 files changed, 62 insertions(+)

diff --git a/src/server/debugcmd.cc b/src/server/debugcmd.cc
index 6b6d18bcd..50ede4a08 100644
--- a/src/server/debugcmd.cc
+++ b/src/server/debugcmd.cc
@@ -148,6 +148,10 @@ void DebugCmd::Run(CmdArgList args) {
     return Inspect(key);
   }
 
+  if (subcmd == "TRANSACTION") {
+    return TxAnalysis();
+  }
+
   string reply = UnknownSubCmd(subcmd, "DEBUG");
   return (*cntx_)->SendError(reply, kSyntaxErrType);
 }
@@ -430,4 +434,61 @@ void DebugCmd::Watched() {
   (*cntx_)->SendStringArr(watched_keys);
 }
 
+void DebugCmd::TxAnalysis() {
+  atomic_uint32_t queue_len{0}, free_cnt{0}, armed_cnt{0};
+
+  using SvLockTable = absl::flat_hash_map<string_view, IntentLock>;
+  vector<SvLockTable> lock_table_arr(shard_set->size());
+
+  auto cb = [&](EngineShard* shard) {
+    ShardId sid = shard->shard_id();
+
+    TxQueue* queue = shard->txq();
+
+    if (queue->Empty())
+      return;
+
+    auto cur = queue->Head();
+    do {
+      auto value = queue->At(cur);
+      Transaction* trx = std::get<Transaction*>(value);
+      queue_len.fetch_add(1, std::memory_order_relaxed);
+
+      if (trx->IsArmedInShard(sid)) {
+        armed_cnt.fetch_add(1, std::memory_order_relaxed);
+
+        IntentLock::Mode mode = trx->Mode();
+
+        // We consider keys from the currently assigned command inside the transaction.
+        // Meaning that for multi-tx it does not take into account all the keys.
+        KeyLockArgs lock_args = trx->GetLockArgs(sid);
+        auto& lock_table = lock_table_arr[sid];
+
+        // We count locks ourselves and do not rely on the lock table inside dbslice.
+        // The reason for this - to account for ordering information.
+        // For example, consider T1, T2 both residing in the queue and both lock 'x' exclusively.
+        // DbSlice::CheckLock returns false for both transactions, but T1 in practice owns the lock.
+        bool can_take = true;
+        for (size_t i = 0; i < lock_args.args.size(); i += lock_args.key_step) {
+          string_view s = lock_args.args[i];
+          bool was_ack = lock_table[s].Acquire(mode);
+          if (!was_ack) {
+            can_take = false;
+          }
+        }
+
+        if (can_take) {
+          free_cnt.fetch_add(1, std::memory_order_relaxed);
+        }
+      }
+      cur = queue->Next(cur);
+    } while (cur != queue->Head());
+  };
+
+  shard_set->RunBriefInParallel(cb);
+
+  (*cntx_)->SendSimpleString(absl::StrCat("queue_len:", queue_len.load(),
+                                          "armed: ", armed_cnt.load(), " free:", free_cnt.load()));
+}
+
 }  // namespace dfly
diff --git a/src/server/debugcmd.h b/src/server/debugcmd.h
index 1ca385630..e8348338b 100644
--- a/src/server/debugcmd.h
+++ b/src/server/debugcmd.h
@@ -26,6 +26,7 @@ class DebugCmd {
   void Load(std::string_view filename);
   void Inspect(std::string_view key);
   void Watched();
+  void TxAnalysis();
 
   ServerFamily& sf_;
   ConnectionContext* cntx_;