Added CUDT::breakAsUnstable()

Removed unused sendBackup_CheckRunningLinkStable

New connection error: broken as unstable

srtcore/core.cpp

@@ -82,7 +82,6 @@ using namespace std;
 using namespace srt;
 using namespace srt::sync;
 using namespace srt_logging;
-using namespace srt;
 
 CUDTUnited CUDT::s_UDTUnited;
 
@@ -249,6 +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;
@@ -10780,15 +10780,15 @@ 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.
@@ -10805,7 +10805,7 @@ bool CUDT::checkExpTimer(const steady_clock::time_point& currtime, int check_rea
         m_pSndQueue->m_pSndUList->update(this, CSndUList::DO_RESCHEDULE);
 
         updateBrokenConnection();
-        completeBrokenConnectionDependencies(SRT_ECONNLOST); // LOCKS!
+        completeBrokenConnectionDependencies(m_bBreakAsUnstable ? SRT_ECONNUNSTABLE : SRT_ECONNLOST); // LOCKS!
 
         return true;
     }

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

srtcore/group.cpp

@@ -3091,43 +3091,6 @@ void CUDTGroup::sendBackup_CheckIdleTime(gli_t w_d)
     }
 }
 
-/// TODO: Remove 'weight' parameter? Only needed for logging.
-/// @retval  1 - link is identified as stable
-/// @retval  0 - link state remains unchanged (too early to identify, still in activation phase)
-/// @retval -1 - link is identified as unstable
-static int sendBackup_CheckRunningLinkStable(const CUDT& u, const srt::sync::steady_clock::time_point& currtime, uint16_t weight ATR_UNUSED)
-{
-    const uint32_t latency_us = u.peerLatency_us();
-    const int32_t min_stability_us     = 60000; // Minimum Link Stability Timeout: 60ms.
-    const int64_t initial_stabtout_us  = max<int64_t>(min_stability_us, latency_us);
-    const int64_t activation_period_us = initial_stabtout_us + 5 * CUDT::COMM_SYN_INTERVAL_US;
-
-    // RTT and RTTVar values are still being refined during activation period,
-    // therefore the dymanic timeout should not be used in activation phase.
-    const bool is_activation_phase = !is_zero(u.freshActivationStart())
-        && (count_microseconds(currtime - u.freshActivationStart()) <= activation_period_us);
-
-    const int64_t stability_tout_us = is_activation_phase
-        ? initial_stabtout_us // activation phase
-        : min<int64_t>(max<int64_t>(min_stability_us, 2 * u.SRTT() + 4 * u.RTTVar()), latency_us);
-
-    const steady_clock::time_point last_rsp = max(u.freshActivationStart(), u.lastRspTime());
-    const steady_clock::duration td_response = currtime - last_rsp;
-    if (count_microseconds(td_response) > stability_tout_us)
-    {
-#if SRT_DEBUG_BONDING_STATES
-        s_stab_trace.trace(u, currtime, activation_period_us, stability_tout_us, is_activation_phase ? "ACTIVATION-UNSTABLE" : "UNSTABLE", weight);
-#endif
-        return -1;
-    }
-
-    // u.lastRspTime() > currtime is alwats true due to the very first check above in this function
-#if SRT_DEBUG_BONDING_STATES
-    s_stab_trace.trace(u, currtime, activation_period_us, stability_tout_us, is_activation_phase ? "ACTIVATION" : "STABLE", weight);
-#endif
-    return is_activation_phase ? 0 : 1;
-}
-
 // [[using locked(this->m_GroupLock)]]
 CUDTGroup::BackupMemberState CUDTGroup::sendBackup_QualifyActiveState(const gli_t d, const time_point currtime)
 {
@@ -3551,12 +3514,15 @@ void CUDTGroup::sendBackup_CheckUnstableSockets(SendBackupCtx& w_sendBackupCtx,
         if (unstable_for_ms < sock.peerIdleTimeout_ms())
             continue;
 
+        // Requesting this socket to be broken with the next CUDT::checkExpTimer() call.
+        sock.breakAsUnstable();
+
         LOGC(gslog.Warn, log << "grp/send*: Socket @" << member->socketID << " is unstable for " << unstable_for_ms 
-            << "ms - qualifying as broken.");
+            << "ms - requesting breakage.");
 
-        w_sendBackupCtx.updateMemberState(member->pSocketData, BKUPST_BROKEN);
-        if (member->pSocketData->ps)
-            sendBackup_AssignBackupState(member->pSocketData->ps->core(), BKUPST_BROKEN, currtime);
+        //w_sendBackupCtx.updateMemberState(member->pSocketData, BKUPST_BROKEN);
+        //if (member->pSocketData->ps)
+        //    sendBackup_AssignBackupState(member->pSocketData->ps->core(), BKUPST_BROKEN, currtime);
     }
 }
 

srtcore/srt.h

@@ -449,6 +449,7 @@ enum CodeMinor
     // MJ_CONNECTION
     MN_CONNLOST        =  1,
     MN_NOCONN          =  2,
+    MN_UNSTABLE        =  3,
     // MJ_SYSTEMRES
     MN_THREAD          =  1,
     MN_MEMORY          =  2,
@@ -501,6 +502,7 @@ typedef enum SRT_ERRNO
     SRT_ECONNFAIL       = MJ(CONNECTION),
     SRT_ECONNLOST       = MN(CONNECTION, CONNLOST),
     SRT_ENOCONN         = MN(CONNECTION, NOCONN),
+    SRT_ECONNUNSTABLE   = MN(CONNECTION, UNSTABLE),
 
     SRT_ERESOURCE       = MJ(SYSTEMRES),
     SRT_ETHREAD         = MN(SYSTEMRES, THREAD),

srtcore/strerror_defs.cpp

@@ -39,6 +39,7 @@ const char* strerror_msgs_connection [] = {
     "", // MN_NONE = 0
     "Connection was broken", // MN_CONNLOST = 1
     "Connection does not exist", // MN_NOCONN = 2
+    "Connection broken as unstable", // MN_UNSTABLE = 3
     ""
 };