Merge pull request #585 from mronstro/RONDB-789

RONDB-789: Automatic memory management when running in a container
logicalclocks · Dec 3, 2024 · a4acab9 · a4acab9
2 parents dd13385 + b279d61
commit a4acab9
Show file tree

Hide file tree

Showing 4 changed files with 27 additions and 1 deletion.
diff --git a/storage/ndb/include/portlib/NdbHW.hpp b/storage/ndb/include/portlib/NdbHW.hpp
@@ -73,6 +73,7 @@ struct ndb_hwinfo
    * Number of Processors, Cores and Sockets
    * MHz of CPU and Model name of CPU.
    */
+  Uint32 is_running_in_container;
   Uint32 cpu_cnt_max;
   Uint32 cpu_cnt;
   Uint32 total_cpu_capacity;

diff --git a/storage/ndb/src/common/portlib/NdbHW.cpp b/storage/ndb/src/common/portlib/NdbHW.cpp
@@ -1523,6 +1523,7 @@ static int Ndb_ReloadHWInfo(struct ndb_hwinfo *hwinfo)
   hwinfo->num_shared_l3_caches = l3_cache_id;
   hwinfo->cpu_model_name[0] = 0;
   hwinfo->hw_memory_size = mem_status.ullTotalPhys;
+  hwinfo->is_running_in_container = 0;
   if (hwinfo->cpu_cnt_max < hwinfo->cpu_cnt)
   {
     hwinfo->cpu_cnt = hwinfo->cpu_cnt_max;
@@ -1661,6 +1662,7 @@ static int Ndb_ReloadHWInfo(struct ndb_hwinfo * hwinfo)
   hwinfo->cpu_model_name[size_var] = 0; // Null terminate
   hwinfo->is_cpuinfo_available = 0;
   hwinfo->is_cpudata_available = 0;
+  hwinfo->is_running_in_container = 0;
   return 0;
 
 error_exit:
@@ -1755,6 +1757,7 @@ static int Ndb_ReloadHWInfo(struct ndb_hwinfo * hwinfo)
   hwinfo->cpu_model_name[0] = 0; // Null terminate
   hwinfo->is_cpuinfo_available = 0;
   hwinfo->is_cpudata_available = 0;
+  hwinfo->is_running_in_container = 0;
   return 0;
 
 error_exit:
@@ -2147,6 +2150,20 @@ static int
 get_meminfo(struct ndb_hwinfo *hwinfo)
 {
   char buf[1024];
+  FILE * cgroup_meminfo = fopen("/sys/fs/cgroup/memory.max", "r");
+  if (cgroup_meminfo != nullptr) {
+    hwinfo->is_running_in_container = 1;
+    FileGuard g(cgroup_meminfo); // close at end...
+    int ret_code = fgets(buf, sizeof(buf), cgroup_meminfo))
+    Uint64 memory_size = 0;
+    if (sscanf(buf, "%llu", &memory_size) == 1) {
+      hwinfo->hw_memory_size = memory_size;
+      return 0;
+    }
+    perror("failed to read /sys/fs/cgroup/memory.max");
+    return -1;
+  }
+  hwinfo->is_running_in_container = 0;
   FILE * meminfo = fopen("/proc/meminfo", "r");
   if (meminfo == nullptr)
   {
@@ -2719,6 +2736,7 @@ static int Ndb_ReloadHWInfo(struct ndb_hwinfo * hwinfo)
   hwinfo->cpu_cnt_max = ncpu;
   hwinfo->cpu_cnt = ncpu;
   check_cpu_online(hwinfo);
+  hwinfo->is_running_in_container = 0;
   return 0;
 }
 

diff --git a/storage/ndb/src/kernel/vm/Configuration.cpp b/storage/ndb/src/kernel/vm/Configuration.cpp
@@ -358,9 +358,11 @@ Configuration::get_total_memory(const ndb_mgm_configuration_iterator *p,
     struct ndb_hwinfo *hwinfo = Ndb_GetHWInfo(false);
     return hwinfo->hw_memory_size;
     total_memory_set = false;
+    globalData.theUseContainerMemoryFlag = hwinfo->is_running_in_container;
   }
   else
   {
+    globalData.theUseContainerMemoryFlag = false;
     total_memory_set = true;
     return total_memory_size;
   }
@@ -1053,7 +1055,11 @@ Configuration::compute_os_overhead(
   ndb_mgm_get_int64_parameter(p, CFG_DB_OS_CPU_OVERHEAD, &os_cpu_overhead);
   if (os_static_overhead == 0)
   {
-    os_static_overhead = Uint64(1400) * MBYTE64;
+    if (globalData.theUseContainerMemoryFlag) {
+      os_static_overhead = Uint64(200) * MBYTE64;
+    } else {
+      os_static_overhead = Uint64(1400) * MBYTE64;
+    }
   }
   if (os_cpu_overhead == 0)
   {

diff --git a/storage/ndb/src/kernel/vm/GlobalData.hpp b/storage/ndb/src/kernel/vm/GlobalData.hpp
@@ -132,6 +132,7 @@ struct GlobalData {
 
   bool       theGracefulShutdownFlag;
   bool       theUseOnlyIPv4Flag;
+  bool       theUseContainerMemoryFlag;
 
 
   NdbMutex   *theIO_lag_mutex;