Feature/hugepages (#2497)

* enabled checks for memory management

* added MADV_NOHUGEPAGE

* added documentation about NUMA and file systems

3rdParty/jemalloc/arangodb-jemalloc.c

@@ -143,6 +143,12 @@ void* adb_mmap(void* addr, size_t length, int prot, int flags) {
         __atomic_add_fetch(&adb_total_size, (uint64_t)length, __ATOMIC_SEQ_CST);
       }
 
+#ifdef MADV_NOHUGEPAGE
+      if (ret != MAP_FAILED) {
+        madvise(ret, length, MADV_NOHUGEPAGE);
+      }
+#endif
+
       return ret;
     }
   }

CHANGELOG

@@ -1,6 +1,8 @@
 devel
 -----
 
+* added checks for memory managment and warn (i. e. if hugepages are enabled)
+
 * set default SSL cipher suite string to "HIGH:!EXPORT:!aNULL@STRENGTH"
 
 * fixed issue #2469: Authentication = true does not protect foxx-routes

Documentation/Books/Manual/Administration/Configuration/OperatingSystem.md

@@ -1,9 +1,23 @@
 Operating System Configuration
 ==============================
 
+File Systems
+------------
+
+(LINUX)
+
+We recommend **not** to use BTRFS on linux, it's known to not work
+well in conjunction with ArangoDB.  We experienced that arangodb
+facing latency issues on accessing its database files on BTRFS
+partitions.  In conjunction with BTRFS and AUFS we also saw data loss
+on restart.
+
+
 Virtual Memory Page Sizes
 --------------------------
 
+(LINUX)
+
 By default, ArangoDB uses Jemalloc as the memory allocator. Jemalloc does a good
 job of reducing virtual memory fragmentation, especially for long-running
 processes. Unfortunately, some OS configurations can interfere with Jemalloc's
@@ -14,18 +28,64 @@ high memory use. Therefore, we recommend disabling these features when using
 Jemalloc with ArangoDB. Please consult your operating system's documentation for
 how to do this.
 
+Execute
+
+    sudo bash -c "echo madvise >/sys/kernel/mm/transparent_hugepage/enabled"
+    sudo bash -c "echo madvise >/sys/kernel/mm/transparent_hugepage/defrag"
+
+before executing `arangod`.
 
 Swap Space
 ----------
 
+(LINUX)
+
 It is recommended to assign swap space for a server that is running arangod.
 Configuring swap space can prevent the operating system's OOM killer from
 killing ArangoDB too eagerly on Linux.
 
+### Over-Commit Memory
+
+Execute
+
+    sudo bash -c "echo 0 >/proc/sys/vm/overcommit_memory"
+
+before executing `arangod`.
+
+From [www.kernel.org](https://www.kernel.org/doc/Documentation/sysctl/vm.txt):
+
+- When this flag is 0, the kernel attempts to estimate the amount
+  of free memory left when userspace requests more memory.
+
+- When this flag is 1, the kernel pretends there is always enough
+  memory until it actually runs out.
+
+- When this flag is 2, the kernel uses a "never overcommit"
+  policy that attempts to prevent any overcommit of memory.
+
+### Zone Reclaim
+
+Execute
+
+    sudo bash -c "echo 0 >/proc/sys/vm/overcommit_memory"
+
+before executing `arangod`.
+
+From [www.kernel.org](https://www.kernel.org/doc/Documentation/sysctl/vm.txt):
+
+This is value ORed together of
+
+- 1 = Zone reclaim on
+- 2 = Zone reclaim writes dirty pages out
+- 4 = Zone reclaim swaps pages
 
 NUMA
 ----
 
+Multi-prozessor systems often have non-uniform Access Memory (NUMA). ArangoDB
+should be started with interleave on such system. This can be archived using
+
+    numactl --interleave=all arangod ...
 
 Environment Variables
 ---------------------
@@ -34,3 +94,18 @@ It is recommended to set the environment variable `GLIBCXX_FORCE_NEW` to 1 on
 systems that use glibc++ in order to disable the memory pooling built into
 glibc++. That memory pooling is unnecessary because Jemalloc will already do
 memory pooling.
+
+Execute
+
+    export GLIBCXX_FORCE_NEW=1
+
+
+before starting `arangod`.
+
+32bit
+-----
+
+While it is possible to compile ArangoDB on 32bit system, this is not a
+recommended environment. 64bit systems can address a significantly bigger
+memory region.
+

Installation/ARM/rc.arangodb

@@ -37,6 +37,8 @@ start () {
     ( cd /var/run/arangodb && chown -R arangodb:arangodb . ) || exit 1
     ( cd $PIDDIR && chown arangodb:arangodb . ) || exit 1
 
+    export GLIBCXX_FORCE_NEW=1
+
     if [ "$1" = "--database.auto-upgrade" ]; then
       $DAEMON -c $CONF --uid arangodb --gid arangodb --server.rest-server false $@
       RETVAL=$?

Installation/debian/arangodb.init

@@ -37,6 +37,8 @@ start () {
     ulimit -H -n 131072 || true
     ulimit -S -n 131072 || true
 
+    export GLIBCXX_FORCE_NEW=1
+
     case "$1" in
         "--upgrade")
             shift

Installation/rpm/rc.arangod.Centos

@@ -35,6 +35,8 @@ start() {
   ulimit -H -n 131072 || true
   ulimit -S -n 131072 || true
 
+  export GLIBCXX_FORCE_NEW=1
+
   case "$1" in
       "--upgrade")
           shift

Installation/rpm/rc.arangod.OpenSuSE

@@ -55,6 +55,8 @@ start () {
     ulimit -H -n 131072 || true
     ulimit -S -n 131072 || true
 
+    export GLIBCXX_FORCE_NEW=1
+
     case "$1" in
         "--upgrade")
             shift

Installation/rpm/rc.arangod.OpenSuSE_13

@@ -55,6 +55,8 @@ start () {
     ulimit -H -n 131072 || true
     ulimit -S -n 131072 || true
 
+    export GLIBCXX_FORCE_NEW=1
+
     case "$1" in
         "--upgrade")
             shift

lib/ApplicationFeatures/EnvironmentFeature.cpp

@@ -38,29 +38,35 @@ EnvironmentFeature::EnvironmentFeature(
 }
 
 void EnvironmentFeature::prepare() {
-#if 0
   if (sizeof(void*) == 4) {
     // 32 bit build
-    LOG_TOPIC(WARN, arangodb::Logger::FIXME) << "this is a 32 bit build of ArangoDB. "
+    LOG_TOPIC(WARN, arangodb::Logger::MEMORY)
-               << "it is recommended to run a 64 bit build instead because it can "
+        << "this is a 32 bit build of ArangoDB. "
-               << "address significantly bigger regions of memory";
+        << "it is recommended to run a 64 bit build instead because it can "
+        << "address significantly bigger regions of memory";
   }
 
 #ifdef __linux__
 
 #ifdef __GLIBC__
   char const* v = getenv("GLIBCXX_FORCE_NEW");
+
   if (v == nullptr) {
     // environment variable not set
-    LOG_TOPIC(DEBUG, arangodb::Logger::FIXME) << "environment variable GLIBCXX_FORCE_NEW' is not set. "
+    LOG_TOPIC(WARN, arangodb::Logger::MEMORY)
-               << "it is recommended to set it to some value to avoid memory pooling in glibc++";
+        << "environment variable GLIBCXX_FORCE_NEW' is not set. "
+        << "it is recommended to set it to some value to avoid memory pooling "
+           "in glibc++";
+    LOG_TOPIC(WARN, arangodb::Logger::MEMORY)
+        << "execute 'export GLIBCXX_FORCE_NEW=1'";
   }
 #endif
 
   try {
-    std::string value = basics::FileUtils::slurp("/proc/sys/vm/overcommit_memory");
+    std::string value =
+        basics::FileUtils::slurp("/proc/sys/vm/overcommit_memory");
     uint64_t v = basics::StringUtils::uint64(value);
-    if (v == 2) {
+    if (v != 0 && v != 1) {
       // from https://www.kernel.org/doc/Documentation/sysctl/vm.txt:
       //
       //   When this flag is 0, the kernel attempts to estimate the amount
@@ -69,15 +75,19 @@ void EnvironmentFeature::prepare() {
       //   memory until it actually runs out.
       //   When this flag is 2, the kernel uses a "never overcommit"
       //   policy that attempts to prevent any overcommit of memory.
-      LOG_TOPIC(WARN, Logger::FIXME) << "/proc/sys/vm/overcommit_memory is set to '" 
+      LOG_TOPIC(WARN, Logger::MEMORY)
-                 << v << "'. it is recommended to set it to a value of 0 or 1";
+          << "/proc/sys/vm/overcommit_memory is set to '" << v
+          << "'. It is recommended to set it to a value of 0 or 1";
+      LOG_TOPIC(WARN, Logger::MEMORY) << "execute 'sudo bash -c \"echo 0 > "
+                                         "/proc/sys/vm/overcommit_memory\"'";
     }
   } catch (...) {
     // file not found or value not convertible into integer
   }
 
   try {
-    std::string value = basics::FileUtils::slurp("/proc/sys/vm/zone_reclaim_mode");
+    std::string value =
+        basics::FileUtils::slurp("/proc/sys/vm/zone_reclaim_mode");
     uint64_t v = basics::StringUtils::uint64(value);
     if (v != 0) {
       // from https://www.kernel.org/doc/Documentation/sysctl/vm.txt:
@@ -85,31 +95,52 @@ void EnvironmentFeature::prepare() {
       //    This is value ORed together of
       //    1 = Zone reclaim on
       //    2 = Zone reclaim writes dirty pages out
-      //    4 =  Zone reclaim swaps pages
+      //    4 = Zone reclaim swaps pages
       //
       // https://www.poempelfox.de/blog/2010/03/19/
-      LOG_TOPIC(WARN, Logger::FIXME) << "/proc/sys/vm/zone_reclaim_mode is set to '" 
+      LOG_TOPIC(WARN, Logger::PERFORMANCE)
-                 << v << "'. it is recommended to set it to a value of 0";
+          << "/proc/sys/vm/zone_reclaim_mode is set to '" << v
+          << "'. It is recommended to set it to a value of 0";
+      LOG_TOPIC(WARN, Logger::PERFORMANCE)
+          << "execute 'sudo bash -c \"echo 0 > "
+             "/proc/sys/vm/zone_reclaim_mode\"'";
     }
   } catch (...) {
     // file not found or value not convertible into integer
   }
 
-  try {
+  bool showHuge = false;
-    std::string value = basics::FileUtils::slurp("/sys/kernel/mm/transparent_hugepage/enabled");
+  std::vector<std::string> paths = {
-    size_t start = value.find('[');
+      "/sys/kernel/mm/transparent_hugepage/enabled",
-    size_t end = value.find(']');
+      "/sys/kernel/mm/transparent_hugepage/defrag"};
-    if (start != std::string::npos && end != std::string::npos && start < end && end - start >= 4) {
+
-      value = value.substr(start + 1, end - start - 1);
+  for (auto file : paths) {
-      if (value == "always") {
+    try {
-        LOG_TOPIC(WARN, Logger::FIXME) << "/sys/kernel/mm/transparent_hugepage/enabled is set to '" 
+      std::string value = basics::FileUtils::slurp(file);
-                   << value << "'. it is recommended to set it to a value of 'never' or 'madvise'";
+      size_t start = value.find('[');
+      size_t end = value.find(']');
+
+      if (start != std::string::npos && end != std::string::npos &&
+          start < end && end - start >= 4) {
+        value = value.substr(start + 1, end - start - 1);
+        if (value == "always") {
+          LOG_TOPIC(WARN, Logger::MEMORY)
+              << file << " is set to '" << value
+              << "'. It is recommended to set it to a value of 'never' "
+                 "or 'madvise'";
+          showHuge = true;
+        }
       }
+    } catch (...) {
+      // file not found
+    }
+  }
+
+  if (showHuge) {
+    for (auto file : paths) {
+      LOG_TOPIC(WARN, Logger::MEMORY)
+          << "execute 'sudo bash -c \"echo madvise > " << file << "\"'";
     }
-  } catch (...) {
-    // file not found
   }
 #endif
-
-#endif
 }

lib/Basics/MutexLocker.h

@@ -99,7 +99,7 @@ class MutexLocker {
 
 #ifdef TRI_SHOW_LOCK_TIME
     if (_time > TRI_SHOW_LOCK_THRESHOLD) {
-      LOG_TOPIC(WARN, arangodb::Logger::PERFORMANCE) << "MutexLocker " << _file << ":" << _line << " took " << _time << " s";
+      LOG_TOPIC(INFO, arangodb::Logger::PERFORMANCE) << "MutexLocker " << _file << ":" << _line << " took " << _time << " s";
     }
 #endif
   }

lib/Basics/ReadLocker.h

@@ -101,7 +101,7 @@ class ReadLocker {
 
 #ifdef TRI_SHOW_LOCK_TIME
     if (_time > TRI_SHOW_LOCK_THRESHOLD) {
-      LOG_TOPIC(WARN, arangodb::Logger::PERFORMANCE) << "ReadLocker " << _file << ":" << _line << " took " << _time << " s";
+      LOG_TOPIC(INFO, arangodb::Logger::PERFORMANCE) << "ReadLocker " << _file << ":" << _line << " took " << _time << " s";
     }
 #endif
   }

lib/Basics/WriteLocker.h

@@ -102,7 +102,7 @@ class WriteLocker {
 
 #ifdef TRI_SHOW_LOCK_TIME
     if (_time > TRI_SHOW_LOCK_THRESHOLD) {
-      LOG_TOPIC(WARN, arangodb::Logger::PERFORMANCE) << "WriteLocker " << _file << ":" << _line << " took " << _time << " s";
+      LOG_TOPIC(INFO, arangodb::Logger::PERFORMANCE) << "WriteLocker " << _file << ":" << _line << " took " << _time << " s";
     }
 #endif
   }

lib/Logger/LogTopic.cpp

@@ -52,9 +52,9 @@ LogTopic Logger::ENGINES("engines", LogLevel::INFO);
 LogTopic Logger::FIXME("general", LogLevel::INFO);
 LogTopic Logger::GRAPHS("graphs", LogLevel::INFO);
 LogTopic Logger::HEARTBEAT("heartbeat", LogLevel::INFO);
-LogTopic Logger::MEMORY("memory", LogLevel::FATAL);  // suppress
+LogTopic Logger::MEMORY("memory", LogLevel::WARN);
 LogTopic Logger::MMAP("mmap");
-LogTopic Logger::PERFORMANCE("performance", LogLevel::FATAL);  // suppress
+LogTopic Logger::PERFORMANCE("performance", LogLevel::WARN);
 LogTopic Logger::PREGEL("pregel", LogLevel::INFO);
 LogTopic Logger::QUERIES("queries", LogLevel::INFO);
 LogTopic Logger::REPLICATION("replication", LogLevel::INFO);

lib/Rest/Version.cpp

@@ -347,10 +347,7 @@ std::string Version::getVerboseVersionString() {
           << " with ASAN"
 #endif
           << ", using "
-#ifdef TRI_HAVE_TCMALLOC
+#ifdef ARANGODB_HAVE_JEMALLOC
-          << "tcmalloc, "
-#endif
-#ifdef TRI_HAVE_JEMALLOC
           << "jemalloc, "
 #endif
           << "VPack " << getVPackVersion() << ", "