Improve recovery on Hibernate error and cleanup

rdkcentral · May 15, 2024 · 4bab06d · 4bab06d
1 parent e9026f3
commit 4bab06d
Show file tree

Hide file tree

Showing 3 changed files with 13 additions and 34 deletions.
diff --git a/daemon/lib/source/DobbyHibernate.cpp b/daemon/lib/source/DobbyHibernate.cpp
@@ -43,7 +43,8 @@ typedef enum {
 typedef enum {
     MEMCR_OK = 0,
     MEMCR_ERROR = -1,
-    MEMCR_INVALID_PID = -2
+    MEMCR_INVALID_PID = -2,
+    MEMCR_SOCKET_READ_ERROR = -3
 } ServerResponseCode;
 
 typedef struct {
@@ -142,7 +143,6 @@ static bool SendRcvCmd(const ServerRequest* cmd, ServerResponse* resp, uint32_t
     AI_LOG_FN_ENTRY();
     int cd;
     int ret;
-    struct sockaddr_in addr = { 0 };
     resp->respCode = MEMCR_ERROR;
 
     cd = Connect(serverLocator, timeoutMs);
@@ -154,15 +154,16 @@ static bool SendRcvCmd(const ServerRequest* cmd, ServerResponse* resp, uint32_t
 
     ret = write(cd, cmd, sizeof(ServerRequest));
     if (ret != sizeof(ServerRequest)) {
-        AI_LOG_ERROR("Socket write failed: ret %d", ret);
+        AI_LOG_ERROR("Socket write failed: ret %d, %m", ret);
         close(cd);
         AI_LOG_FN_EXIT();
         return false;
     }
 
     ret = read(cd, resp, sizeof(ServerResponse));
     if (ret != sizeof(ServerResponse)) {
-        AI_LOG_ERROR("Socket read failed: ret %d", ret);
+        AI_LOG_ERROR("Socket read failed: ret %d, %m", ret);
+        resp->respCode = MEMCR_SOCKET_READ_ERROR;
         close(cd);
         AI_LOG_FN_EXIT();
         return false;
@@ -188,6 +189,10 @@ DobbyHibernate::Error DobbyHibernate::HibernateProcess(const pid_t pid, const ui
         AI_LOG_INFO("Hibernate process PID %d success", pid);
         AI_LOG_FN_EXIT();
         return DobbyHibernate::Error::ErrorNone;
+    } else if (resp.respCode == MEMCR_SOCKET_READ_ERROR) {
+        AI_LOG_WARN("Error Hibernate timeout process PID %d ret %d", pid, resp.respCode);
+        AI_LOG_FN_EXIT();
+        return DobbyHibernate::Error::ErrorTimeout;
     } else {
         AI_LOG_WARN("Error Hibernate process PID %d ret %d", pid, resp.respCode);
         AI_LOG_FN_EXIT();

diff --git a/daemon/lib/source/DobbyHibernate.h b/daemon/lib/source/DobbyHibernate.h
@@ -32,7 +32,8 @@ class DobbyHibernate
     enum Error
     {
         ErrorNone = 0,
-        ErrorGeneral = 1
+        ErrorGeneral = 1,
+        ErrorTimeout = 2
     };
 
     enum CompressionAlg

diff --git a/daemon/lib/source/DobbyManager.cpp b/daemon/lib/source/DobbyManager.cpp
@@ -1566,35 +1566,7 @@ bool DobbyManager::hibernateContainer(int32_t cd, const std::string& options)
     std::thread hibernateThread =
         std::thread([=]()
         {
-            //TODO: --delay support is temporary and should be removed
-            int delayMs = 0;
-            size_t delayMsPos = options.find("--delay=");
-            if (delayMsPos != std::string::npos)
-            {
-                delayMs = std::stoi(&options[delayMsPos + std::string("--delay=").length()], nullptr, 10);
-            }
             DobbyHibernate::Error ret = DobbyHibernate::Error::ErrorNone;
-
-            int delayChunkMs = 100;
-            while (delayMs > 0)
-            {
-                int sleepTime = delayMs > delayChunkMs? delayChunkMs : delayMs;
-                delayMs -= sleepTime;
-                std::this_thread::sleep_for(std::chrono::milliseconds(sleepTime));
-                {
-                    std::lock_guard<std::mutex> locker(mLock);
-                    if (mContainers.find(id) == mContainers.end() ||
-                        mContainers[id]->descriptor != cd ||
-                        mContainers[id]->state != DobbyContainer::State::Hibernating)
-                    {
-                        AI_LOG_WARN("Hibernation of: %s with descriptor %d aborted", id.c_str(), cd);
-                        AI_LOG_FN_EXIT();
-                        return;
-                    }
-                }
-            }
-            //TODO: --delay support end
-
             // create a stats object for the container to get list of PIDs
             std::unique_lock<std::mutex> locker(mLock);
             DobbyStats stats(it->first, mEnvironment, mUtilities);
@@ -1619,7 +1591,8 @@ bool DobbyManager::hibernateContainer(int32_t cd, const std::string& options)
                 if (ret != DobbyHibernate::Error::ErrorNone)
                 {
                     AI_LOG_WARN("Error hibernating pid: '%d'", pid);
-                    // revert previous Hibernations and break
+                    // try to revert current and previous Hibernations and break
+                    DobbyHibernate::WakeupProcess(pid);
                     while (pidIt != jsonPids.begin())
                     {
                         --pidIt;