SERVER-67704 SERVER-67705 Fix processor package counting for ARM

10 files changed, 145 insertions, 106 deletions

diff --git a/src/mongo/db/commands/generic_servers.cpp b/src/mongo/db/commands/generic_servers.cpp
index af1ffc8d110..46276bb285b 100644
--- a/src/mongo/db/commands/generic_servers.cpp
+++ b/src/mongo/db/commands/generic_servers.cpp
@@ -158,8 +158,11 @@ HostInfoReply HostInfoCmd::Invocation::typedRun(OperationContext*) {
     system.setMemSizeMB(static_cast<long>(p.getSystemMemSizeMB()));
     system.setMemLimitMB(static_cast<long>(p.getMemSizeMB()));
     system.setNumCores(static_cast<int>(p.getNumAvailableCores()));
+    system.setNumPhysicalCores(static_cast<int>(p.getNumPhysicalCores()));
+    system.setNumCpuSockets(static_cast<int>(p.getNumCpuSockets()));
     system.setCpuArch(p.getArch());
     system.setNumaEnabled(p.hasNumaEnabled());
+    system.setNumNumaNodes(static_cast<int>(p.getNumNumaNodes()));
 
     HostInfoOsReply os;
     os.setType(p.getOsType());
diff --git a/src/mongo/db/commands/generic_servers.idl b/src/mongo/db/commands/generic_servers.idl
index 4e6f08ab71f..33b54dc4651 100644
--- a/src/mongo/db/commands/generic_servers.idl
+++ b/src/mongo/db/commands/generic_servers.idl
@@ -65,8 +65,11 @@ structs:
             memSizeMB: long
             memLimitMB: long
             numCores: int
+            numPhysicalCores: int
+            numCpuSockets: int
             cpuArch: string
             numaEnabled: bool
+            numNumaNodes: int
 
     hostInfoOsReply:
         description: "hostInfo.os reply fields"
diff --git a/src/mongo/util/processinfo.h b/src/mongo/util/processinfo.h
index 3046f809e4f..509a9d5ccf1 100644
--- a/src/mongo/util/processinfo.h
+++ b/src/mongo/util/processinfo.h
@@ -113,6 +113,13 @@ public:
     }
 
     /**
+     * Get the number of CPU sockets
+     */
+    static unsigned getNumCpuSockets() {
+        return sysInfo().numCpuSockets;
+    }
+
+    /**
      * Get the number of cores available. Make a best effort to get the cores for this process.
      * If that information is not available, get the total number of CPUs.
      */
@@ -142,6 +149,16 @@ public:
     }
 
     /**
+     * Get the number of NUMA nodes if NUMA is enabled, or 1 otherwise.
+     */
+    static unsigned long getNumNumaNodes() {
+        if (sysInfo().hasNuma) {
+            return sysInfo().numNumaNodes;
+        }
+        return 1;
+    }
+
+    /**
      * Determine if we need to workaround slow msync performance on Illumos/Solaris
      */
     static bool preferMsyncOverFSync() {
@@ -180,9 +197,11 @@ private:
         unsigned long long memLimit;
         unsigned numCores;
         unsigned numPhysicalCores;
+        unsigned numCpuSockets;
         unsigned long long pageSize;
         std::string cpuArch;
         bool hasNuma;
+        unsigned numNumaNodes;
         BSONObj _extraStats;
 
         // On non-Solaris (ie, Linux, Darwin, *BSD) kernels, prefer msync.
@@ -198,8 +217,10 @@ private:
               memLimit(0),
               numCores(0),
               numPhysicalCores(0),
+              numCpuSockets(0),
               pageSize(0),
               hasNuma(false),
+              numNumaNodes(0),
               preferMsyncOverFSync(true) {
             // populate SystemInfo during construction
             collectSystemInfo();
@@ -230,8 +251,6 @@ private:
 
     ProcessId _pid;
 
-    static bool checkNumaEnabled();
-
     static const SystemInfo& sysInfo() {
         static ProcessInfo::SystemInfo systemInfo;
         return systemInfo;
diff --git a/src/mongo/util/processinfo_freebsd.cpp b/src/mongo/util/processinfo_freebsd.cpp
index b57e9ff6c1a..66709cbfe57 100644
--- a/src/mongo/util/processinfo_freebsd.cpp
+++ b/src/mongo/util/processinfo_freebsd.cpp
@@ -94,10 +94,6 @@ int getSysctlByNameWithDefault<std::string>(const char* sysctlName,
     return 0;
 }
 
-bool ProcessInfo::checkNumaEnabled() {
-    return false;
-}
-
 int ProcessInfo::getVirtualMemorySize() {
     kvm_t* kd = NULL;
     int cnt = 0;
@@ -166,7 +162,7 @@ void ProcessInfo::SystemInfo::collectSystemInfo() {
 
     pageSize = static_cast<unsigned long long>(sysconf(_SC_PAGESIZE));
 
-    hasNuma = checkNumaEnabled();
+    hasNuma = false;
 }
 
 void ProcessInfo::getExtraInfo(BSONObjBuilder& info) {}
diff --git a/src/mongo/util/processinfo_linux.cpp b/src/mongo/util/processinfo_linux.cpp
index c0d4454ad64..6f762f8c580 100644
--- a/src/mongo/util/processinfo_linux.cpp
+++ b/src/mongo/util/processinfo_linux.cpp
@@ -427,6 +427,24 @@ public:
     }
 
     /**
+     * count the number of processor packages
+     */
+    static int getNumCpuSockets() {
+        std::set<std::string> socketIds;
+
+        CpuInfoParser cpuInfoParser{
+            {
+                {"physical id", [&](const std::string& value) { socketIds.insert(value); }},
+            },
+            []() {}};
+        cpuInfoParser.run();
+
+        // On ARM64, the "physical id" field is unpopulated, causing there to be 0 sockets found. In
+        // this case, we default to 1.
+        return std::max(socketIds.size(), 1ul);
+    }
+
+    /**
      * Get some details about the CPU
      */
     static void getCpuInfo(int& procCount, std::string& freq, std::string& features) {
@@ -674,6 +692,44 @@ void ProcessInfo::getExtraInfo(BSONObjBuilder& info) {
 }
 
 /**
+ * If the process is running with (cc)NUMA enabled, return the number of NUMA nodes. Else, return 0.
+ */
+unsigned long countNumaNodes() {
+    bool hasMultipleNodes = false;
+    bool hasNumaMaps = false;
+
+    try {
+        hasMultipleNodes = boost::filesystem::exists("/sys/devices/system/node/node1");
+        hasNumaMaps = boost::filesystem::exists("/proc/self/numa_maps");
+
+        if (hasMultipleNodes && hasNumaMaps) {
+            // proc is populated with numa entries
+
+            // read the second column of first line to determine numa state
+            // ('default' = enabled, 'interleave' = disabled).  Logic from version.cpp's warnings.
+            std::string line =
+                LinuxSysHelper::readLineFromFile("/proc/self/numa_maps").append(" \0");
+            size_t pos = line.find(' ');
+            if (pos != std::string::npos &&
+                line.substr(pos + 1, 10).find("interleave") == std::string::npos) {
+                // interleave not found, count NUMA nodes by finding the highest numbered node file
+                unsigned long i = 2;
+                while (boost::filesystem::exists(
+                    std::string(str::stream() << "/sys/devices/system/node/node" << i++)))
+                    ;
+                return i;
+            }
+        }
+    } catch (boost::filesystem::filesystem_error& e) {
+        LOGV2(23340,
+              "WARNING: Cannot detect if NUMA interleaving is enabled. Failed to probe",
+              "path"_attr = e.path1().string(),
+              "reason"_attr = e.code().message());
+    }
+    return 0;
+}
+
+/**
  * Save a BSON obj representing the host system's details
  */
 void ProcessInfo::SystemInfo::collectSystemInfo() {
@@ -682,10 +738,12 @@ void ProcessInfo::SystemInfo::collectSystemInfo() {
     std::string cpuFreq, cpuFeatures;
     int cpuCount;
     int physicalCores;
+    int cpuSockets;
 
     std::string verSig = LinuxSysHelper::readLineFromFile("/proc/version_signature");
     LinuxSysHelper::getCpuInfo(cpuCount, cpuFreq, cpuFeatures);
     LinuxSysHelper::getNumPhysicalCores(physicalCores);
+    cpuSockets = LinuxSysHelper::getNumCpuSockets();
     LinuxSysHelper::getLinuxDistro(distroName, distroVersion);
 
     if (uname(&unameData) == -1) {
@@ -702,9 +760,12 @@ void ProcessInfo::SystemInfo::collectSystemInfo() {
     memLimit = LinuxSysHelper::getMemorySizeLimit();
     addrSize = sizeof(void*) * CHAR_BIT;
     numCores = cpuCount;
+    numPhysicalCores = physicalCores;
+    numCpuSockets = cpuSockets;
     pageSize = static_cast<unsigned long long>(sysconf(_SC_PAGESIZE));
     cpuArch = unameData.machine;
-    hasNuma = checkNumaEnabled();
+    numNumaNodes = countNumaNodes();
+    hasNuma = numNumaNodes;
 
     BSONObjBuilder bExtra;
     bExtra.append("versionString", LinuxSysHelper::readLineFromFile("/proc/version"));
@@ -729,44 +790,10 @@ void ProcessInfo::SystemInfo::collectSystemInfo() {
     bExtra.append("pageSize", static_cast<long long>(pageSize));
     bExtra.append("numPages", static_cast<int>(sysconf(_SC_PHYS_PAGES)));
     bExtra.append("maxOpenFiles", static_cast<int>(sysconf(_SC_OPEN_MAX)));
-    bExtra.append("physicalCores", physicalCores);
 
     appendMountInfo(bExtra);
 
     _extraStats = bExtra.obj();
 }
 
-/**
- * Determine if the process is running with (cc)NUMA
- */
-bool ProcessInfo::checkNumaEnabled() {
-    bool hasMultipleNodes = false;
-    bool hasNumaMaps = false;
-
-    try {
-        hasMultipleNodes = boost::filesystem::exists("/sys/devices/system/node/node1");
-        hasNumaMaps = boost::filesystem::exists("/proc/self/numa_maps");
-    } catch (boost::filesystem::filesystem_error& e) {
-        LOGV2(23340,
-              "WARNING: Cannot detect if NUMA interleaving is enabled. Failed to probe",
-              "path"_attr = e.path1().string(),
-              "reason"_attr = e.code().message());
-        return false;
-    }
-
-    if (hasMultipleNodes && hasNumaMaps) {
-        // proc is populated with numa entries
-
-        // read the second column of first line to determine numa state
-        // ('default' = enabled, 'interleave' = disabled).  Logic from version.cpp's warnings.
-        std::string line = LinuxSysHelper::readLineFromFile("/proc/self/numa_maps").append(" \0");
-        size_t pos = line.find(' ');
-        if (pos != std::string::npos &&
-            line.substr(pos + 1, 10).find("interleave") == std::string::npos)
-            // interleave not found;
-            return true;
-    }
-    return false;
-}
-
 }  // namespace mongo
diff --git a/src/mongo/util/processinfo_openbsd.cpp b/src/mongo/util/processinfo_openbsd.cpp
index 09ef930501a..ab7773ab4b1 100644
--- a/src/mongo/util/processinfo_openbsd.cpp
+++ b/src/mongo/util/processinfo_openbsd.cpp
@@ -98,10 +98,6 @@ int getSysctlByIDWithDefault<std::string>(const int* sysctlID,
     return 0;
 }
 
-bool ProcessInfo::checkNumaEnabled() {
-    return false;
-}
-
 int ProcessInfo::getVirtualMemorySize() {
     kvm_t* kd = NULL;
     int cnt = 0;
@@ -183,7 +179,7 @@ void ProcessInfo::SystemInfo::collectSystemInfo() {
 
     pageSize = static_cast<unsigned long long>(sysconf(_SC_PAGESIZE));
 
-    hasNuma = checkNumaEnabled();
+    hasNuma = false;
 }
 
 void ProcessInfo::getExtraInfo(BSONObjBuilder& info) {}
diff --git a/src/mongo/util/processinfo_osx.cpp b/src/mongo/util/processinfo_osx.cpp
index f3eb932cd86..662b26dcfd3 100644
--- a/src/mongo/util/processinfo_osx.cpp
+++ b/src/mongo/util/processinfo_osx.cpp
@@ -184,9 +184,11 @@ void ProcessInfo::SystemInfo::collectSystemInfo() {
     memSize = getSysctlByName<NumberVal>("hw.memsize");
     memLimit = memSize;
     numCores = getSysctlByName<NumberVal>("hw.ncpu");  // includes hyperthreading cores
+    numPhysicalCores = getSysctlByName<NumberVal>("machdep.cpu.core_count");
+    numCpuSockets = getSysctlByName<NumberVal>("hw.packages");
     pageSize = static_cast<unsigned long long>(sysconf(_SC_PAGESIZE));
     cpuArch = getSysctlByName<std::string>("hw.machine");
-    hasNuma = checkNumaEnabled();
+    hasNuma = false;
 
     BSONObjBuilder bExtra;
     bExtra.append("versionString", getSysctlByName<std::string>("kern.version"));
@@ -196,16 +198,10 @@ void ProcessInfo::SystemInfo::collectSystemInfo() {
         "nfsAsync",
         static_cast<int>(getSysctlByName<NumberVal>("vfs.generic.nfs.client.allow_async")));
     bExtra.append("model", getSysctlByName<std::string>("hw.model"));
-    bExtra.append("physicalCores",
-                  static_cast<int>(getSysctlByName<NumberVal>("machdep.cpu.core_count")));
     bExtra.append("cpuString", getSysctlByName<std::string>("machdep.cpu.brand_string"));
     bExtra.append("pageSize", static_cast<int>(getSysctlByName<NumberVal>("hw.pagesize")));
     bExtra.append("scheduler", getSysctlByName<std::string>("kern.sched"));
     _extraStats = bExtra.obj();
 }
 
-bool ProcessInfo::checkNumaEnabled() {
-    return false;
-}
-
 }  // namespace mongo
diff --git a/src/mongo/util/processinfo_solaris.cpp b/src/mongo/util/processinfo_solaris.cpp
index de64cf46998..ba66d31d709 100644
--- a/src/mongo/util/processinfo_solaris.cpp
+++ b/src/mongo/util/processinfo_solaris.cpp
@@ -134,6 +134,33 @@ void ProcessInfo::getExtraInfo(BSONObjBuilder& info) {
     info.appendNumber("page_faults", static_cast<long long>(p.prusage.pr_majf));
 }
 
+bool checkNumaEnabled() {
+    lgrp_cookie_t cookie = lgrp_init(LGRP_VIEW_OS);
+
+    if (cookie == LGRP_COOKIE_NONE) {
+        auto ec = lastSystemError();
+        LOGV2_WARNING(23362,
+                      "lgrp_init failed: {errnoWithDescription}",
+                      "errnoWithDescription"_attr = errorMessage(ec));
+        return false;
+    }
+
+    ON_BLOCK_EXIT([&] { lgrp_fini(cookie); });
+
+    int groups = lgrp_nlgrps(cookie);
+
+    if (groups == -1) {
+        auto ec = lastSystemError();
+        LOGV2_WARNING(23363,
+                      "lgrp_nlgrps failed: {errnoWithDescription}",
+                      "errnoWithDescription"_attr = errorMessage(ec));
+        return false;
+    }
+
+    // NUMA machines have more then 1 locality group
+    return groups > 1;
+}
+
 /**
  * Save a BSON obj representing the host system's details
  */
@@ -203,31 +230,4 @@ void ProcessInfo::SystemInfo::collectSystemInfo() {
     _extraStats = bExtra.obj();
 }
 
-bool ProcessInfo::checkNumaEnabled() {
-    lgrp_cookie_t cookie = lgrp_init(LGRP_VIEW_OS);
-
-    if (cookie == LGRP_COOKIE_NONE) {
-        auto ec = lastSystemError();
-        LOGV2_WARNING(23362,
-                      "lgrp_init failed: {errnoWithDescription}",
-                      "errnoWithDescription"_attr = errorMessage(ec));
-        return false;
-    }
-
-    ON_BLOCK_EXIT([&] { lgrp_fini(cookie); });
-
-    int groups = lgrp_nlgrps(cookie);
-
-    if (groups == -1) {
-        auto ec = lastSystemError();
-        LOGV2_WARNING(23363,
-                      "lgrp_nlgrps failed: {errnoWithDescription}",
-                      "errnoWithDescription"_attr = errorMessage(ec));
-        return false;
-    }
-
-    // NUMA machines have more then 1 locality group
-    return groups > 1;
-}
-
 }  // namespace mongo
diff --git a/src/mongo/util/processinfo_unknown.cpp b/src/mongo/util/processinfo_unknown.cpp
index 1011c4dbfef..c57516eaeac 100644
--- a/src/mongo/util/processinfo_unknown.cpp
+++ b/src/mongo/util/processinfo_unknown.cpp
@@ -51,10 +51,6 @@ int ProcessInfo::getResidentSize() {
     return -1;
 }
 
-bool ProcessInfo::checkNumaEnabled() {
-    return false;
-}
-
 void ProcessInfo::SystemInfo::collectSystemInfo() {}
 
 void ProcessInfo::getExtraInfo(BSONObjBuilder& info) {}
diff --git a/src/mongo/util/processinfo_windows.cpp b/src/mongo/util/processinfo_windows.cpp
index 3f4e5ba74f1..d90d0898b6c 100644
--- a/src/mongo/util/processinfo_windows.cpp
+++ b/src/mongo/util/processinfo_windows.cpp
@@ -93,13 +93,27 @@ LpiRecords getLogicalProcessorInformationRecords() {
     return lpiRecords;
 }
 
-int getPhysicalCores() {
-    int processorCoreCount = 0;
+struct ParsedProcessorInfo {
+    int physicalCoreCount;
+    int numaNodeCount;
+    int processorPackageCount;
+};
+
+ParsedProcessorInfo getProcessorInfo() {
+    ParsedProcessorInfo ppi{0, 0, 0};
     for (auto&& lpi : getLogicalProcessorInformationRecords()) {
-        if (lpi.Relationship == RelationProcessorCore)
-            processorCoreCount++;
+        switch (lpi.Relationship) {
+            case RelationProcessorCore:
+                ppi.physicalCoreCount++;
+                break;
+            case RelationNumaNode:
+                ppi.numaNodeCount++;
+                break;
+            case RelationProcessorPackage:
+                ppi.processorPackageCount++;
+        }
     }
-    return processorCoreCount;
+    return ppi;
 }
 
 }  // namespace
@@ -247,10 +261,13 @@ void ProcessInfo::SystemInfo::collectSystemInfo() {
     GetNativeSystemInfo(&ntsysinfo);
     addrSize = (ntsysinfo.wProcessorArchitecture == PROCESSOR_ARCHITECTURE_AMD64 ? 64 : 32);
     numCores = ntsysinfo.dwNumberOfProcessors;
-    numPhysicalCores = getPhysicalCores();
+    auto ppi = getProcessorInfo();
+    numPhysicalCores = ppi.physicalCoreCount;
+    numCpuSockets = ppi.processorPackageCount;
+    hasNuma = ppi.numaNodeCount > 1;
+    numNumaNodes = ppi.numaNodeCount;
     pageSize = static_cast<unsigned long long>(ntsysinfo.dwPageSize);
     bExtra.append("pageSize", static_cast<long long>(pageSize));
-    bExtra.append("physicalCores", static_cast<int>(numPhysicalCores));
 
     // get memory info
     mse.dwLength = sizeof(mse);
@@ -354,21 +371,7 @@ void ProcessInfo::SystemInfo::collectSystemInfo() {
 
     osType = "Windows";
     osVersion = verstr.str();
-    hasNuma = checkNumaEnabled();
     _extraStats = bExtra.obj();
 }
 
-
-bool ProcessInfo::checkNumaEnabled() {
-    DWORD numaNodeCount = 0;
-    for (auto&& lpi : getLogicalProcessorInformationRecords()) {
-        if (lpi.Relationship == RelationNumaNode)
-            // Non-NUMA systems report a single record of this type.
-            ++numaNodeCount;
-    }
-
-    // For non-NUMA machines, the count is 1
-    return numaNodeCount > 1;
-}
-
 }  // namespace mongo