[slack-19.0] backport 18967: connpool: fix connection leak during idle connection reopen (#18967)

go/flags/endtoend/vtcombo.txt

@@ -297,18 +297,21 @@ Flags:
       --queryserver-config-pool-conn-max-lifetime duration               query server connection max lifetime, vttablet manages various mysql connection pools. This config means if a connection has lived at least this long, it connection will be removed from pool upon the next time it is returned to the pool.
       --queryserver-config-pool-size int                                 query server read pool size, connection pool is used by regular queries (non streaming, not in a transaction) (default 16)
       --queryserver-config-query-cache-memory int                        query server query cache size in bytes, maximum amount of memory to be used for caching. vttablet analyzes every incoming query and generate a query plan, these plans are being cached in a lru cache. This config controls the capacity of the lru cache. (default 33554432)
+      --queryserver-config-query-pool-max-idle-count int                 query server query pool - maximum number of idle connections to retain in the pool. Use this to balance between faster response times during traffic bursts and resource efficiency during low-traffic periods.
       --queryserver-config-query-pool-timeout duration                   query server query pool timeout, it is how long vttablet waits for a connection from the query pool. If set to 0 (default) then the overall query timeout is used instead.
       --queryserver-config-query-timeout duration                        query server query timeout, this is the query timeout in vttablet side. If a query takes more than this timeout, it will be killed. (default 30s)
       --queryserver-config-schema-change-signal                          query server schema signal, will signal connected vtgates that schema has changed whenever this is detected. VTGates will need to have -schema_change_signal enabled for this to work (default true)
       --queryserver-config-schema-reload-time duration                   query server schema reload time, how often vttablet reloads schemas from underlying MySQL instance. vttablet keeps table schemas in its own memory and periodically refreshes it from MySQL. This config controls the reload time. (default 30m0s)
       --queryserver-config-stream-buffer-size int                        query server stream buffer size, the maximum number of bytes sent from vttablet for each stream call. It's recommended to keep this value in sync with vtgate's stream_buffer_size. (default 32768)
+      --queryserver-config-stream-pool-max-idle-count int                query server stream pool - maximum number of idle connections to retain in the pool. Use this to balance between faster response times during traffic bursts and resource efficiency during low-traffic periods.
       --queryserver-config-stream-pool-size int                          query server stream connection pool size, stream pool is used by stream queries: queries that return results to client in a streaming fashion (default 200)
       --queryserver-config-stream-pool-timeout duration                  query server stream pool timeout, it is how long vttablet waits for a connection from the stream pool. If set to 0 (default) then there is no timeout.
       --queryserver-config-strict-table-acl                              only allow queries that pass table acl checks
       --queryserver-config-terse-errors                                  prevent bind vars from escaping in client error messages
       --queryserver-config-transaction-cap int                           query server transaction cap is the maximum number of transactions allowed to happen at any given point of a time for a single vttablet. E.g. by setting transaction cap to 100, there are at most 100 transactions will be processed by a vttablet and the 101th transaction will be blocked (and fail if it cannot get connection within specified timeout) (default 20)
       --queryserver-config-transaction-timeout duration                  query server transaction timeout, a transaction will be killed if it takes longer than this value (default 30s)
       --queryserver-config-truncate-error-len int                        truncate errors sent to client if they are longer than this value (0 means do not truncate)
+      --queryserver-config-txpool-max-idle-count int                     query server transaction pool - maximum number of idle connections to retain in the pool. Use this to balance between faster response times during traffic bursts and resource efficiency during low-traffic periods.
       --queryserver-config-txpool-timeout duration                       query server transaction pool timeout, it is how long vttablet waits if tx pool is full (default 1s)
       --queryserver-config-warn-result-size int                          query server result size warning threshold, warn if number of rows returned from vttablet for non-streaming queries exceeds this
       --queryserver-enable-settings-pool                                 Enable pooling of connections with modified system settings (default true)

go/flags/endtoend/vttablet.txt

@@ -287,18 +287,21 @@ Flags:
       --queryserver-config-pool-conn-max-lifetime duration               query server connection max lifetime, vttablet manages various mysql connection pools. This config means if a connection has lived at least this long, it connection will be removed from pool upon the next time it is returned to the pool.
       --queryserver-config-pool-size int                                 query server read pool size, connection pool is used by regular queries (non streaming, not in a transaction) (default 16)
       --queryserver-config-query-cache-memory int                        query server query cache size in bytes, maximum amount of memory to be used for caching. vttablet analyzes every incoming query and generate a query plan, these plans are being cached in a lru cache. This config controls the capacity of the lru cache. (default 33554432)
+      --queryserver-config-query-pool-max-idle-count int                 query server query pool - maximum number of idle connections to retain in the pool. Use this to balance between faster response times during traffic bursts and resource efficiency during low-traffic periods.
       --queryserver-config-query-pool-timeout duration                   query server query pool timeout, it is how long vttablet waits for a connection from the query pool. If set to 0 (default) then the overall query timeout is used instead.
       --queryserver-config-query-timeout duration                        query server query timeout, this is the query timeout in vttablet side. If a query takes more than this timeout, it will be killed. (default 30s)
       --queryserver-config-schema-change-signal                          query server schema signal, will signal connected vtgates that schema has changed whenever this is detected. VTGates will need to have -schema_change_signal enabled for this to work (default true)
       --queryserver-config-schema-reload-time duration                   query server schema reload time, how often vttablet reloads schemas from underlying MySQL instance. vttablet keeps table schemas in its own memory and periodically refreshes it from MySQL. This config controls the reload time. (default 30m0s)
       --queryserver-config-stream-buffer-size int                        query server stream buffer size, the maximum number of bytes sent from vttablet for each stream call. It's recommended to keep this value in sync with vtgate's stream_buffer_size. (default 32768)
+      --queryserver-config-stream-pool-max-idle-count int                query server stream pool - maximum number of idle connections to retain in the pool. Use this to balance between faster response times during traffic bursts and resource efficiency during low-traffic periods.
       --queryserver-config-stream-pool-size int                          query server stream connection pool size, stream pool is used by stream queries: queries that return results to client in a streaming fashion (default 200)
       --queryserver-config-stream-pool-timeout duration                  query server stream pool timeout, it is how long vttablet waits for a connection from the stream pool. If set to 0 (default) then there is no timeout.
       --queryserver-config-strict-table-acl                              only allow queries that pass table acl checks
       --queryserver-config-terse-errors                                  prevent bind vars from escaping in client error messages
       --queryserver-config-transaction-cap int                           query server transaction cap is the maximum number of transactions allowed to happen at any given point of a time for a single vttablet. E.g. by setting transaction cap to 100, there are at most 100 transactions will be processed by a vttablet and the 101th transaction will be blocked (and fail if it cannot get connection within specified timeout) (default 20)
       --queryserver-config-transaction-timeout duration                  query server transaction timeout, a transaction will be killed if it takes longer than this value (default 30s)
       --queryserver-config-truncate-error-len int                        truncate errors sent to client if they are longer than this value (0 means do not truncate)
+      --queryserver-config-txpool-max-idle-count int                     query server transaction pool - maximum number of idle connections to retain in the pool. Use this to balance between faster response times during traffic bursts and resource efficiency during low-traffic periods.
       --queryserver-config-txpool-timeout duration                       query server transaction pool timeout, it is how long vttablet waits if tx pool is full (default 1s)
       --queryserver-config-warn-result-size int                          query server result size warning threshold, warn if number of rows returned from vttablet for non-streaming queries exceeds this
       --queryserver-enable-settings-pool                                 Enable pooling of connections with modified system settings (default true)

go/pools/smartconnpool/pool.go

@@ -91,6 +91,7 @@ type RefreshCheck func() (bool, error)
 
 type Config[C Connection] struct {
 	Capacity        int64
+	MaxIdleCount    int64
 	IdleTimeout     time.Duration
 	MaxLifetime     time.Duration
 	RefreshInterval time.Duration
@@ -122,6 +123,8 @@ type ConnPool[C Connection] struct {
 	active atomic.Int64
 	// capacity is the maximum number of connections that this pool can open
 	capacity atomic.Int64
+	// maxIdleCount is the maximum idle connections in the pool
+	idleCount atomic.Int64
 
 	// workers is a waitgroup for all the currently running worker goroutines
 	workers    sync.WaitGroup
@@ -136,6 +139,8 @@ type ConnPool[C Connection] struct {
 		// maxCapacity is the maximum value to which capacity can be set; when the pool
 		// is re-opened, it defaults to this capacity
 		maxCapacity int64
+		// maxIdleCount is the maximum idle connections in the pool
+		maxIdleCount int64
 		// maxLifetime is the maximum time a connection can be open
 		maxLifetime atomic.Int64
 		// idleTimeout is the maximum time a connection can remain idle
@@ -155,6 +160,7 @@ type ConnPool[C Connection] struct {
 func NewPool[C Connection](config *Config[C]) *ConnPool[C] {
 	pool := &ConnPool[C]{}
 	pool.config.maxCapacity = config.Capacity
+	pool.config.maxIdleCount = config.MaxIdleCount
 	pool.config.maxLifetime.Store(config.MaxLifetime.Nanoseconds())
 	pool.config.idleTimeout.Store(config.IdleTimeout.Nanoseconds())
 	pool.config.refreshInterval.Store(config.RefreshInterval.Nanoseconds())
@@ -194,6 +200,7 @@ func (pool *ConnPool[C]) open() {
 	}
 
 	pool.capacity.Store(pool.config.maxCapacity)
+	pool.setIdleCount()
 
 	// The expire worker takes care of removing from the waiter list any clients whose
 	// context has been cancelled.
@@ -324,6 +331,16 @@ func (pool *ConnPool[C]) MaxCapacity() int64 {
 	return pool.config.maxCapacity
 }
 
+func (pool *ConnPool[C]) setIdleCount() {
+	capacity := pool.Capacity()
+	maxIdleCount := pool.config.maxIdleCount
+	if maxIdleCount == 0 || maxIdleCount > capacity {
+		pool.idleCount.Store(capacity)
+	} else {
+		pool.idleCount.Store(maxIdleCount)
+	}
+}
+
 // InUse returns the number of connections that the pool has lent out to clients and that
 // haven't been returned yet.
 func (pool *ConnPool[C]) InUse() int64 {
@@ -349,6 +366,10 @@ func (pool *ConnPool[C]) SetIdleTimeout(duration time.Duration) {
 	pool.config.idleTimeout.Store(duration.Nanoseconds())
 }
 
+func (pool *ConnPool[D]) IdleCount() int64 {
+	return pool.idleCount.Load()
+}
+
 func (pool *ConnPool[D]) RefreshInterval() time.Duration {
 	return time.Duration(pool.config.refreshInterval.Load())
 }
@@ -416,6 +437,9 @@ func (pool *ConnPool[C]) tryReturnConn(conn *Pooled[C]) bool {
 		return true
 	}
 
+	if pool.closeOnIdleLimitReached(conn) {
+		return false
+	}
 	connSetting := conn.Conn.Setting()
 	if connSetting == nil {
 		pool.clean.Push(conn)
@@ -436,9 +460,13 @@ func (pool *ConnPool[C]) pop(stack *connStack[C]) *Pooled[C] {
 	// to expire this connection (even if it's still visible to them), so it's
 	// safe to return it
 	for conn, ok := stack.Pop(); ok; conn, ok = stack.Pop() {
-		if conn.timeUsed.borrow() {
-			return conn
+		if !conn.timeUsed.borrow() {
+			// Ignore the connection that couldn't be borrowed;
+			// it's being closed by the idle worker and replaced by a new connection.
+			continue
 		}
+
+		return conn
 	}
 	return nil
 }
@@ -457,6 +485,23 @@ func (pool *ConnPool[C]) tryReturnAnyConn() bool {
 	return false
 }
 
+// closeOnIdleLimitReached closes a connection if the number of idle connections (active - inuse) in the pool
+// exceeds the idleCount limit. It returns true if the connection is closed, false otherwise.
+func (pool *ConnPool[C]) closeOnIdleLimitReached(conn *Pooled[C]) bool {
+	for {
+		open := pool.active.Load()
+		idle := open - pool.borrowed.Load()
+		if idle <= pool.idleCount.Load() {
+			return false
+		}
+		if pool.active.CompareAndSwap(open, open-1) {
+			pool.Metrics.idleClosed.Add(1)
+			conn.Close()
+			return true
+		}
+	}
+}
+
 func (pool *ConnPool[D]) extendedMaxLifetime() time.Duration {
 	maxLifetime := pool.config.maxLifetime.Load()
 	if maxLifetime == 0 {
@@ -695,6 +740,9 @@ func (pool *ConnPool[C]) setCapacity(ctx context.Context, newcap int64) error {
 	if oldcap == newcap {
 		return nil
 	}
+	// update the idle count to match the new capacity if necessary
+	// wait for connections to be returned to the pool if we're reducing the capacity.
+	defer pool.setIdleCount()
 
 	const delay = 10 * time.Millisecond
 
@@ -744,11 +792,23 @@ func (pool *ConnPool[C]) closeIdleResources(now time.Time) {
 		for conn := s.Peek(); conn != nil; conn = conn.next.Load() {
 			if conn.timeUsed.expired(mono, timeout) {
 				pool.Metrics.idleClosed.Add(1)
+
 				conn.Close()
+				pool.closedConn()
+
 				// Using context.Background() is fine since MySQL connection already enforces
-				// a connect timeout via the `db_connect_timeout_ms` config param.
-				if err := pool.connReopen(context.Background(), conn, mono); err != nil {
-					pool.closedConn()
+				// a connect timeout via the `db-connect-timeout-ms` config param.
+				c, err := pool.getNew(context.Background())
+				if err != nil {
+					// If we couldn't open a new connection, just continue
+					continue
+				}
+
+				// opening a new connection might have raced with other goroutines,
+				// so it's possible that we got back `nil` here
+				if c != nil {
+					// Return the new connection to the pool
+					pool.tryReturnConn(c)
 				}
 			}
 		}
@@ -798,6 +858,9 @@ func (pool *ConnPool[C]) RegisterStats(stats *servenv.Exporter, name string) {
 		// the smartconnpool doesn't have a maximum capacity
 		return pool.Capacity()
 	})
+	stats.NewGaugeFunc(name+"IdleAllowed", "Tablet server conn pool idle allowed limit", func() int64 {
+		return pool.IdleCount()
+	})
 	stats.NewCounterFunc(name+"WaitCount", "Tablet server conn pool wait count", func() int64 {
 		return pool.Metrics.WaitCount()
 	})