[core] Main/backup: improved switching rules (#1857)

and refactored the related code
Haivision · May 10, 2021 · 2559797 · 2559797
1 parent 8830943
commit 2559797
Show file tree

Hide file tree

Showing 9 changed files with 1,036 additions and 797 deletions.
diff --git a/srtcore/core.cpp b/srtcore/core.cpp
@@ -248,6 +248,7 @@ void CUDT::construct()
     m_bClosing            = false;
     m_bShutdown           = false;
     m_bBroken             = false;
+    m_bBreakAsUnstable    = false;
     // TODO: m_iBrokenCounter should be still set to some default.
     m_bPeerHealth         = true;
     m_RejectReason        = SRT_REJ_UNKNOWN;
@@ -950,12 +951,13 @@ void CUDT::open()
     m_tsLastRspAckTime                     = currtime;
     m_tsLastSndTime                        = currtime;
 
-    m_iReXmitCount   = 1;
-    m_tsUnstableSince = steady_clock::time_point();
+    m_tsUnstableSince   = steady_clock::time_point();
     m_tsFreshActivation = steady_clock::time_point();
+    m_tsWarySince       = steady_clock::time_point();
+
+    m_iReXmitCount   = 1;
     m_iPktCount      = 0;
     m_iLightACKCount = 1;
-
     m_tsNextSendTime = steady_clock::time_point();
     m_tdSendTimeDiff = microseconds_from(0);
 
@@ -10777,15 +10779,15 @@ bool CUDT::checkExpTimer(const steady_clock::time_point& currtime, int check_rea
         next_exp_time = m_tsLastRspTime + exp_timeout;
     }
 
-    if (currtime <= next_exp_time)
+    if (currtime <= next_exp_time && !m_bBreakAsUnstable)
         return false;
 
     // ms -> us
     const int PEER_IDLE_TMO_US = m_config.iPeerIdleTimeout * 1000;
     // Haven't received any information from the peer, is it dead?!
     // timeout: at least 16 expirations and must be greater than 5 seconds
-    if ((m_iEXPCount > COMM_RESPONSE_MAX_EXP) &&
-        (currtime - m_tsLastRspTime > microseconds_from(PEER_IDLE_TMO_US)))
+    if (m_bBreakAsUnstable || ((m_iEXPCount > COMM_RESPONSE_MAX_EXP) &&
+        (currtime - m_tsLastRspTime > microseconds_from(PEER_IDLE_TMO_US))))
     {
         //
         // Connection is broken.

diff --git a/srtcore/core.h b/srtcore/core.h
@@ -420,6 +420,9 @@ class CUDT
     SRTU_PROPERTY_RR(srt::sync::Condition*, recvDataCond, &m_RecvDataCond);
     SRTU_PROPERTY_RR(srt::sync::Condition*, recvTsbPdCond, &m_RcvTsbPdCond);
 
+    /// @brief  Request a socket to be broken due to too long instability (normally by a group).
+    void breakAsUnstable() { m_bBreakAsUnstable = true; }
+
     void ConnectSignal(ETransmissionEvent tev, EventSlot sl);
     void DisconnectSignal(ETransmissionEvent tev);
 
@@ -726,6 +729,7 @@ class CUDT
     volatile bool m_bClosing;                    // If the UDT entity is closing
     volatile bool m_bShutdown;                   // If the peer side has shutdown the connection
     volatile bool m_bBroken;                     // If the connection has been broken
+    volatile bool m_bBreakAsUnstable;            // A flag indicating that the socket should become broken because it has been unstable for too long.
     volatile bool m_bPeerHealth;                 // If the peer status is normal
     volatile int m_RejectReason;
     bool m_bOpened;                              // If the UDT entity has been opened
@@ -908,7 +912,7 @@ class CUDT
 
     srt::sync::Mutex m_SendLock;                 // used to synchronize "send" call
     srt::sync::Mutex m_RcvLossLock;              // Protects the receiver loss list (access: CRcvQueue::worker, CUDT::tsbpd)
-    srt::sync::Mutex m_StatsLock;                // used to synchronize access to trace statistics
+    mutable srt::sync::Mutex m_StatsLock;        // used to synchronize access to trace statistics
 
     void initSynch();
     void destroySynch();
@@ -1090,8 +1094,9 @@ class CUDT
     static const int PACKETPAIR_MASK = 0xF;
 
 private: // Timers functions
-    time_point m_tsFreshActivation; // time of fresh activation of the link, or 0 if past the activation phase or idle
-    time_point m_tsUnstableSince;   // time since unexpected ACK delay experienced, or 0 if link seems healthy
+    time_point m_tsFreshActivation; // GROUPS: time of fresh activation of the link, or 0 if past the activation phase or idle
+    time_point m_tsUnstableSince;   // GROUPS: time since unexpected ACK delay experienced, or 0 if link seems healthy
+    time_point m_tsWarySince;       // GROUPS: time since an unstable link has first some response
 
     static const int BECAUSE_NO_REASON = 0, // NO BITS
                      BECAUSE_ACK       = 1 << 0,

diff --git a/srtcore/filelist.maf b/srtcore/filelist.maf
@@ -29,6 +29,7 @@ window.cpp
 
 SOURCES - ENABLE_EXPERIMENTAL_BONDING
 group.cpp
+group_backup.cpp
 group_common.cpp
 
 SOURCES - !ENABLE_STDCXX_SYNC
@@ -76,4 +77,5 @@ window.h
 
 PRIVATE HEADERS - ENABLE_EXPERIMENTAL_BONDING
 group.h
+group_backup.h
 group_common.h