fix(pacakging): fix cpu/memory metrics when running inside LXC container as systemd service (#14255)

Fixes https://github.com/netdata/netdata/issues/14238

collectors/proc.plugin/plugin_proc.c

@@ -98,6 +98,41 @@ static void proc_main_cleanup(void *ptr)
     worker_unregister();
 }
 
+bool inside_lxc_container = false;
+
+static bool is_lxcfs_proc_mounted() {
+    procfile *ff = NULL;
+
+    if (unlikely(!ff)) {
+        char filename[FILENAME_MAX + 1];
+        snprintfz(filename, FILENAME_MAX, "/proc/self/mounts");
+        ff = procfile_open(filename, " \t", PROCFILE_FLAG_DEFAULT);
+        if (unlikely(!ff))
+            return false;
+    }
+
+    ff = procfile_readall(ff);
+    if (unlikely(!ff))
+        return false;
+
+    unsigned long l, lines = procfile_lines(ff);
+
+    for (l = 0; l < lines; l++) {
+        size_t words = procfile_linewords(ff, l);
+        if (words < 2) {
+            continue;
+        }
+        if (!strcmp(procfile_lineword(ff, l, 0), "lxcfs") && !strncmp(procfile_lineword(ff, l, 1), "/proc", 5)) {
+            procfile_close(ff);
+            return true;   
+        }            
+    }
+
+    procfile_close(ff);
+
+    return false;
+}
+
 void *proc_main(void *ptr)
 {
     worker_register("PROC");
@@ -128,6 +163,8 @@ void *proc_main(void *ptr)
     heartbeat_t hb;
     heartbeat_init(&hb);
 
+    inside_lxc_container = is_lxcfs_proc_mounted();
+
     while (service_running(SERVICE_COLLECTORS)) {
         worker_is_idle();
         usec_t hb_dt = heartbeat_next(&hb, step);

collectors/proc.plugin/plugin_proc.h

@@ -48,6 +48,7 @@ int get_numa_node_count(void);
 
 // metrics that need to be shared among data collectors
 extern unsigned long long zfs_arcstats_shrinkable_cache_size_bytes;
+extern bool inside_lxc_container;
 
 // netdev renames
 void netdev_rename_device_add(

collectors/proc.plugin/proc_meminfo.c

@@ -158,9 +158,11 @@ int do_proc_meminfo(int update_every, usec_t dt) {
     unsigned long long MemCached = Cached + SReclaimable - Shmem;
     unsigned long long MemUsed = MemTotal - MemFree - MemCached - Buffers;
     // The Linux kernel doesn't report ZFS ARC usage as cache memory (the ARC is included in the total used system memory)
-    MemCached += (zfs_arcstats_shrinkable_cache_size_bytes / 1024);
-    MemUsed -= (zfs_arcstats_shrinkable_cache_size_bytes / 1024);
-    MemAvailable += (zfs_arcstats_shrinkable_cache_size_bytes / 1024);
+    if (!inside_lxc_container) {
+        MemCached += (zfs_arcstats_shrinkable_cache_size_bytes / 1024);
+        MemUsed -= (zfs_arcstats_shrinkable_cache_size_bytes / 1024);
+        MemAvailable += (zfs_arcstats_shrinkable_cache_size_bytes / 1024);
+    }
 
     if(do_ram) {
         {

daemon/system-info.sh

@@ -217,6 +217,9 @@ if [ -n "${lscpu}" ] && lscpu > /dev/null 2>&1; then
   LCPU_COUNT="$(echo "${lscpu_output}" | grep "^CPU(s):" | cut -f 2 -d ':' | sed -e 's/^[[:space:]]*//' -e 's/[[:space:]]*$//')"
   CPU_VENDOR="$(echo "${lscpu_output}" | grep "^Vendor ID:" | cut -f 2 -d ':' | sed -e 's/^[[:space:]]*//' -e 's/[[:space:]]*$//')"
   CPU_MODEL="$(echo "${lscpu_output}" | grep "^Model name:" | cut -f 2 -d ':' | sed -e 's/^[[:space:]]*//' -e 's/[[:space:]]*$//')"
+  if grep -q "^lxcfs /proc" /proc/self/mounts 2>/dev/null && count=$(grep -c ^processor /proc/cpuinfo 2>/dev/null); then
+    LCPU_COUNT="$count"
+  fi
   possible_cpu_freq="$(echo "${lscpu_output}" | grep -F "CPU max MHz:" | cut -f 2 -d ':' | sed -e 's/^[[:space:]]*//' -e 's/[[:space:]]*$//' | grep -o '^[0-9]*')"
   if [ -z "$possible_cpu_freq" ]; then
     possible_cpu_freq="$(echo "${lscpu_output}" | grep -F "CPU MHz:" | cut -f 2 -d ':' | sed -e 's/^[[:space:]]*//' -e 's/[[:space:]]*$//' | grep -o '^[0-9]*')"
@@ -437,7 +440,7 @@ CLOUD_INSTANCE_TYPE="unknown"
 CLOUD_INSTANCE_REGION="unknown"
 
 if [ "${VIRTUALIZATION}" != "none" ] && command -v curl > /dev/null 2>&1; then
-  # Returned HTTP status codes: GCP is 200, AWS is 200, DO is 404. 
+  # Returned HTTP status codes: GCP is 200, AWS is 200, DO is 404.
   curl --fail -s -m 1 --noproxy "*" http://169.254.169.254 >/dev/null 2>&1
   ret=$?
   # anything but operation timeout.

system/netdata.service.in

@@ -71,6 +71,10 @@ ProtectControlGroups=on
 ReadWriteDirectories=/run/netdata
 # This is needed to make email-based alert deliver work if Postfix is the email provider on the system.
 ReadWriteDirectories=-/var/spool/postfix/maildrop
+# LXCFS directories (https://github.com/lxc/lxcfs#lxcfs)
+# If we don't set them explicitly, systemd mounts procfs from the host. See https://github.com/netdata/netdata/issues/14238. 
+BindReadOnlyPaths=-/proc/cpuinfo -/proc/diskstats -/proc/loadavg -/proc/meminfo
+BindReadOnlyPaths=-/proc/stat -/proc/swaps -/proc/uptime -/proc/slabinfo
 
 [Install]
 WantedBy=multi-user.target