ReadClient should delay re-subscribe attempts based on delay in BUSY responses.

bzbarsky-apple · bzbarsky-apple · commit 97892203ab65 · 2024-04-01T17:15:21.000-04:00
The ifdefs are to reduce the codesize costs for situations where the BUSY
handling is not needed (e.g. if the application does not use ReadClient at all
and does not care about handling the timeout from BUSY from
OperationalSessionSetup).
diff --git a/src/app/BUILD.gn b/src/app/BUILD.gn
@@ -77,6 +77,7 @@ buildconfig_header("app_buildconfig") {
     "TIME_SYNC_ENABLE_TSC_FEATURE=${time_sync_enable_tsc_feature}",
     "NON_SPEC_COMPLIANT_OTA_ACTION_DELAY_FLOOR=${non_spec_compliant_ota_action_delay_floor}",
     "CHIP_DEVICE_CONFIG_DYNAMIC_SERVER=${chip_build_controller_dynamic_server}",
+    "CHIP_CONFIG_ENABLE_BUSY_HANDLING_FOR_OPERATIONAL_SESSION_SETUP=${chip_enable_busy_handling_for_operational_session_setup}",
   ]
 
   visibility = [ ":app_config" ]
diff --git a/src/app/OperationalSessionSetup.cpp b/src/app/OperationalSessionSetup.cpp
@@ -351,6 +351,12 @@ void OperationalSessionSetup::DequeueConnectionCallbacks(CHIP_ERROR error, Sessi
     auto * exchangeMgr                            = mInitParams.exchangeMgr;
     Optional<SessionHandle> optionalSessionHandle = mSecureSession.Get();
     ScopedNodeId peerId                           = mPeerId;
+    System::Clock::Milliseconds16 requestedBusyDelay =
+#if CHIP_CONFIG_ENABLE_BUSY_HANDLING_FOR_OPERATIONAL_SESSION_SETUP
+        mRequestedBusyDelay;
+#else
+        System::Clock::kZero;
+#endif // CHIP_CONFIG_ENABLE_BUSY_HANDLING_FOR_OPERATIONAL_SESSION_SETUP
 
     if (releaseBehavior == ReleaseBehavior::Release)
     {
@@ -361,14 +367,15 @@ void OperationalSessionSetup::DequeueConnectionCallbacks(CHIP_ERROR error, Sessi
     // DO NOT touch any members of this object after this point.  It's dead.
 
     NotifyConnectionCallbacks(failureReady, setupFailureReady, successReady, error, stage, peerId, performingAddressUpdate,
-                              exchangeMgr, optionalSessionHandle);
+                              exchangeMgr, optionalSessionHandle, requestedBusyDelay);
 }
 
 void OperationalSessionSetup::NotifyConnectionCallbacks(Cancelable & failureReady, Cancelable & setupFailureReady,
                                                         Cancelable & successReady, CHIP_ERROR error,
                                                         SessionEstablishmentStage stage, const ScopedNodeId & peerId,
                                                         bool performingAddressUpdate, Messaging::ExchangeManager * exchangeMgr,
-                                                        const Optional<SessionHandle> & optionalSessionHandle)
+                                                        const Optional<SessionHandle> & optionalSessionHandle,
+                                                        System::Clock::Milliseconds16 requestedBusyDelay)
 {
     //
     // If we encountered no error, go ahead and call all success callbacks. Otherwise,
@@ -401,6 +408,12 @@ void OperationalSessionSetup::NotifyConnectionCallbacks(Cancelable & failureRead
         {
             // Initialize the ConnnectionFailureInfo object
             ConnnectionFailureInfo failureInfo(peerId, error, stage);
+#if CHIP_CONFIG_ENABLE_BUSY_HANDLING_FOR_OPERATIONAL_SESSION_SETUP
+            if (error == CHIP_ERROR_BUSY)
+            {
+                failureInfo.requestedBusyDelay.Emplace(requestedBusyDelay);
+            }
+#endif // CHIP_CONFIG_ENABLE_BUSY_HANDLING_FOR_OPERATIONAL_SESSION_SETUP
             cb->mCall(cb->mContext, failureInfo);
         }
     }
@@ -482,9 +495,9 @@ void OperationalSessionSetup::OnSessionEstablishmentError(CHIP_ERROR error, Sess
 
 void OperationalSessionSetup::OnResponderBusy(System::Clock::Milliseconds16 requestedDelay)
 {
-#if CHIP_DEVICE_CONFIG_ENABLE_AUTOMATIC_CASE_RETRIES
+#if CHIP_DEVICE_CONFIG_ENABLE_AUTOMATIC_CASE_RETRIES || CHIP_CONFIG_ENABLE_BUSY_HANDLING_FOR_OPERATIONAL_SESSION_SETUP
     // Store the requested delay, so that we can use it for scheduling our
-    // retry.
+    // retry or communicate it to our API consumer.
     mRequestedBusyDelay = requestedDelay;
 #endif
 }
diff --git a/src/app/OperationalSessionSetup.h b/src/app/OperationalSessionSetup.h
@@ -26,6 +26,7 @@
 
 #pragma once
 
+#include <app/AppConfig.h>
 #include <app/CASEClient.h>
 #include <app/CASEClientPool.h>
 #include <app/DeviceProxy.h>
@@ -161,6 +162,13 @@ class DLL_EXPORT OperationalSessionSetup : public SessionEstablishmentDelegate,
         CHIP_ERROR error;
         SessionEstablishmentStage sessionStage;
 
+        // When the response was BUSY, error will be CHIP_ERROR_BUSY and
+        // requestedBusyDelay will be set, if handling of BUSY responses is
+        // enabled.
+#if CHIP_CONFIG_ENABLE_BUSY_HANDLING_FOR_OPERATIONAL_SESSION_SETUP
+        Optional<System::Clock::Milliseconds16> requestedBusyDelay;
+#endif // CHIP_CONFIG_ENABLE_BUSY_HANDLING_FOR_OPERATIONAL_SESSION_SETUP
+
         ConnnectionFailureInfo(const ScopedNodeId & peer, CHIP_ERROR err, SessionEstablishmentStage stage) :
             peerId(peer), error(err), sessionStage(stage)
         {}
@@ -306,6 +314,10 @@ class DLL_EXPORT OperationalSessionSetup : public SessionEstablishmentDelegate,
 
     bool mPerformingAddressUpdate = false;
 
+#if CHIP_DEVICE_CONFIG_ENABLE_AUTOMATIC_CASE_RETRIES || CHIP_CONFIG_ENABLE_BUSY_HANDLING_FOR_OPERATIONAL_SESSION_SETUP
+    System::Clock::Milliseconds16 mRequestedBusyDelay = System::Clock::kZero;
+#endif // CHIP_DEVICE_CONFIG_ENABLE_AUTOMATIC_CASE_RETRIES || CHIP_CONFIG_ENABLE_BUSY_HANDLING_FOR_OPERATIONAL_SESSION_SETUP
+
 #if CHIP_DEVICE_CONFIG_ENABLE_AUTOMATIC_CASE_RETRIES
     // When we TryNextResult on the resolver, it will synchronously call back
     // into our OnNodeAddressResolved when it succeeds.  We need to track
@@ -320,8 +332,6 @@ class DLL_EXPORT OperationalSessionSetup : public SessionEstablishmentDelegate,
 
     uint8_t mResolveAttemptsAllowed = 0;
 
-    System::Clock::Milliseconds16 mRequestedBusyDelay = System::Clock::kZero;
-
     Callback::CallbackDeque mConnectionRetry;
 #endif // CHIP_DEVICE_CONFIG_ENABLE_AUTOMATIC_CASE_RETRIES
 
@@ -385,7 +395,12 @@ class DLL_EXPORT OperationalSessionSetup : public SessionEstablishmentDelegate,
                                           Callback::Cancelable & successReady, CHIP_ERROR error, SessionEstablishmentStage stage,
                                           const ScopedNodeId & peerId, bool performingAddressUpdate,
                                           Messaging::ExchangeManager * exchangeMgr,
-                                          const Optional<SessionHandle> & optionalSessionHandle);
+                                          const Optional<SessionHandle> & optionalSessionHandle,
+                                          // requestedBusyDelay will be 0 if not
+                                          // CHIP_CONFIG_ENABLE_BUSY_HANDLING_FOR_OPERATIONAL_SESSION_SETUP,
+                                          // and only has a meaningful value
+                                          // when the error is CHIP_ERROR_BUSY.
+                                          System::Clock::Milliseconds16 requestedBusyDelay);
 
     /**
      * Triggers a DNSSD lookup to find a usable peer address.
diff --git a/src/app/ReadClient.cpp b/src/app/ReadClient.cpp
@@ -135,6 +135,11 @@ uint32_t ReadClient::ComputeTimeTillNextSubscription()
         waitTimeInMsec    = minWaitTimeInMsec + (Crypto::GetRandU32() % (maxWaitTimeInMsec - minWaitTimeInMsec));
     }
 
+    if (mMinimalResubscribeDelay.count() > waitTimeInMsec)
+    {
+        waitTimeInMsec = mMinimalResubscribeDelay.count();
+    }
+
     return waitTimeInMsec;
 }
 
@@ -1058,6 +1063,10 @@ CHIP_ERROR ReadClient::ProcessSubscribeResponse(System::PacketBufferHandle && aP
 
 CHIP_ERROR ReadClient::SendAutoResubscribeRequest(ReadPrepareParams && aReadPrepareParams)
 {
+    // Make sure we don't use minimal resubscribe delays from previous attempts
+    // for this one.
+    mMinimalResubscribeDelay = System::Clock::kZero;
+
     mReadPrepareParams = std::move(aReadPrepareParams);
     CHIP_ERROR err     = SendSubscribeRequest(mReadPrepareParams);
     if (err != CHIP_NO_ERROR)
@@ -1239,14 +1248,28 @@ void ReadClient::HandleDeviceConnected(void * context, Messaging::ExchangeManage
     }
 }
 
-void ReadClient::HandleDeviceConnectionFailure(void * context, const ScopedNodeId & peerId, CHIP_ERROR err)
+void ReadClient::HandleDeviceConnectionFailure(void * context, const OperationalSessionSetup::ConnnectionFailureInfo & failureInfo)
 {
     ReadClient * const _this = static_cast<ReadClient *>(context);
     VerifyOrDie(_this != nullptr);
 
-    ChipLogError(DataManagement, "Failed to establish CASE for re-subscription with error '%" CHIP_ERROR_FORMAT "'", err.Format());
+    ChipLogError(DataManagement, "Failed to establish CASE for re-subscription with error '%" CHIP_ERROR_FORMAT "'",
+                 failureInfo.error.Format());
+
+#if CHIP_CONFIG_ENABLE_BUSY_HANDLING_FOR_OPERATIONAL_SESSION_SETUP
+#if CHIP_DETAIL_LOGGING
+    if (failureInfo.requestedBusyDelay.HasValue())
+    {
+        ChipLogDetail(DataManagement, "Will delay resubscription by %u ms due to BUSY response",
+                      failureInfo.requestedBusyDelay.Value().count());
+    }
+#endif // CHIP_DETAIL_LOGGING
+    _this->mMinimalResubscribeDelay = failureInfo.requestedBusyDelay.ValueOr(System::Clock::kZero);
+#else
+    _this->mMinimalResubscribeDelay = System::Clock::kZero;
+#endif // CHIP_CONFIG_ENABLE_BUSY_HANDLING_FOR_OPERATIONAL_SESSION_SETUP
 
-    _this->Close(err);
+    _this->Close(failureInfo.error);
 }
 
 void ReadClient::OnResubscribeTimerCallback(System::Layer * /* If this starts being used, fix callers that pass nullptr */,
diff --git a/src/app/ReadClient.h b/src/app/ReadClient.h
@@ -612,7 +612,7 @@ class ReadClient : public Messaging::ExchangeDelegate
 
     static void HandleDeviceConnected(void * context, Messaging::ExchangeManager & exchangeMgr,
                                       const SessionHandle & sessionHandle);
-    static void HandleDeviceConnectionFailure(void * context, const ScopedNodeId & peerId, CHIP_ERROR error);
+    static void HandleDeviceConnectionFailure(void * context, const OperationalSessionSetup::ConnnectionFailureInfo & failureInfo);
 
     CHIP_ERROR GetMinEventNumber(const ReadPrepareParams & aReadPrepareParams, Optional<EventNumber> & aEventMin);
 
@@ -642,8 +642,12 @@ class ReadClient : public Messaging::ExchangeDelegate
     bool mForceCaseOnNextResub      = true;
     bool mIsResubscriptionScheduled = false;
 
+    // mMinimalResubscribeDelay is used to store the delay returned with a BUSY
+    // response to a Sigma1 message.
+    System::Clock::Milliseconds16 mMinimalResubscribeDelay = System::Clock::kZero;
+
     chip::Callback::Callback<OnDeviceConnected> mOnConnectedCallback;
-    chip::Callback::Callback<OnDeviceConnectionFailure> mOnConnectionFailureCallback;
+    chip::Callback::Callback<OperationalSessionSetup::OnSetupFailure> mOnConnectionFailureCallback;
 
     ReadClient * mpNext                 = nullptr;
     InteractionModelEngine * mpImEngine = nullptr;
diff --git a/src/app/common_flags.gni b/src/app/common_flags.gni
@@ -17,4 +17,8 @@ declare_args() {
   chip_app_use_echo = false
   chip_enable_read_client = true
   chip_build_controller_dynamic_server = false
+
+  # Flag that controls whether the time-to-wait from BUSY responses is
+  # communicated to OperationalSessionSetup API consumers.
+  chip_enable_busy_handling_for_operational_session_setup = true
 }

Original file line number	Diff line number	Diff line change
`@@ -77,6 +77,7 @@ buildconfig_header("app_buildconfig") {`
`77`	`77`	`"TIME_SYNC_ENABLE_TSC_FEATURE=${time_sync_enable_tsc_feature}",`
`78`	`78`	`"NON_SPEC_COMPLIANT_OTA_ACTION_DELAY_FLOOR=${non_spec_compliant_ota_action_delay_floor}",`
`79`	`79`	`"CHIP_DEVICE_CONFIG_DYNAMIC_SERVER=${chip_build_controller_dynamic_server}",`
	`80`	`+ "CHIP_CONFIG_ENABLE_BUSY_HANDLING_FOR_OPERATIONAL_SESSION_SETUP=${chip_enable_busy_handling_for_operational_session_setup}",`
`80`	`81`	`]`
`81`	`82`
`82`	`83`	`visibility = [ ":app_config" ]`
Original file line number	Diff line number	Diff line change
`@@ -17,4 +17,8 @@ declare_args() {`
`17`	`17`	`chip_app_use_echo = false`
`18`	`18`	`chip_enable_read_client = true`
`19`	`19`	`chip_build_controller_dynamic_server = false`
	`20`	`+`
	`21`	`+ # Flag that controls whether the time-to-wait from BUSY responses is`
	`22`	`+ # communicated to OperationalSessionSetup API consumers.`
	`23`	`+ chip_enable_busy_handling_for_operational_session_setup = true`
`20`	`24`	`}`