collectd configuration

collectd/etc/collectd.conf.d/btrfs-data.conf

@@ -1,5 +1,5 @@
-#LoadPlugin exec
-#
-#<Plugin "exec">
-#  Exec nobody "/host/usr/local/bin/btrfs-data"
-#</Plugin>
+LoadPlugin exec
+
+<Plugin "exec">
+  Exec collectd "/host/usr/local/bin/btrfs-data"
+</Plugin>

collectd/etc/collectd.conf.d/cpufreq-data.conf

@@ -0,0 +1,5 @@
+LoadPlugin exec
+
+<Plugin "exec">
+  Exec collectd "/host/usr/local/bin/cpufreq-data"
+</Plugin>

collectd/etc/collectd.conf.d/df.conf

@@ -1,6 +1,5 @@
 <Plugin df>
     MountPoint "/media/docker"
-    MountPoint "/media/scratch"
     FSType "ext4"
     IgnoreSelected false
 </Plugin>

collectd/etc/collectd.conf.d/du-data.conf

@@ -1,6 +1,5 @@
 LoadPlugin exec
 
 <Plugin "exec">
-  Exec nobody "/host/usr/local/bin/du-data"
+  Exec collectd "/host/usr/local/bin/du-data"
 </Plugin>
-

collectd/etc/collectd.conf.d/speedtest-data.conf

@@ -3,4 +3,3 @@ LoadPlugin exec
 <Plugin "exec">
   Exec nobody "/host/usr/local/bin/speedtest-data"
 </Plugin>
-

collectd/usr/local/bin/btrfs-data

@@ -1,18 +1,18 @@
-#!/usr/bin/python
+#!/usr/bin/python3
 
 #
 # Imports
 #
 import sys
 import time
-import commands
+import subprocess
 import argparse
 
 
 #
 # Misc
 #
-#sys.tracebacklimit = 0
+# sys.tracebacklimit = 0
 
 
 #
@@ -28,21 +28,42 @@ size_snapshot_exclusive = 0
 # Methods
 #
 def get_subvol_list(path):
-    command = "btrfs subvolume list -t %s" % (path)
-    status, output = commands.getstatusoutput(command)
+    command = "sudo btrfs subvolume list -t %s" % (path)
+    status, output = subprocess.getstatusoutput(command)
 
-    if status is not 0:
+    if status != 0:
         raise Exception(command)
 
     # Every line contains the following values: subvol_id, gen, toplevel, path
     return output.splitlines()[2:]
 
 
+def get_filesystem_size(path):
+    command = "sudo btrfs filesystem show --raw %s" % (path)
+    status, output = subprocess.getstatusoutput(command)
+
+    if status != 0:
+        # This command fails when running inside Docker container
+        # return maximum size of any filesystem instead
+        command = "sudo btrfs filesystem show --raw"
+        status, output = subprocess.getstatusoutput(command)
+        lines = output.splitlines()
+        lines = [x for x in lines if "devid" in x]
+        sizes = [int(line.split()[3]) for line in lines]
+        return max(sizes)
+
+    # The sizes are on the third line
+    line = output.splitlines()[2]
+
+    # Element 3 and 5 respectively contain total and used sizes
+    return int(line.split()[3])
+
+
 def get_id_root(name, path):
     lines = get_subvol_list(path)
 
     # Filter lines where toplevel == 5
-    subvol_ids = filter(lambda x: int(x.split()[2]) == 5, lines)
+    subvol_ids = [x for x in lines if int(x.split()[2]) == 5]
 
     # Try to retrieve the subvol_id for the root subvolume (if any)
     if len(subvol_ids) == 1:
@@ -51,28 +72,30 @@ def get_id_root(name, path):
     else:
         # The path contains a btrfs filesystem with multiple subvolumes for data
         try:
-            return int(filter(lambda x: x.split()[3] == name, subvol_ids)[0].split()[0])
+            return int(list(filter(lambda x: x.split()[3] == name, subvol_ids))[0].split()[0])
         except IndexError:
             pass
-            
 
     # Volume not found, root is probably the btrfs default (5)
     return 5
-            
+
+
 def get_id_subvolumes(path, subvol_id):
     lines = get_subvol_list(path)
-    lines = filter(lambda x: int(x.split()[2]) == subvol_id, lines)
-    return list(map(lambda x: int(x.split()[0]), lines))
+    lines = [x for x in lines if int(x.split()[2]) == subvol_id]
+    return list([int(x.split()[0]) for x in lines])
 
 
 def get_disk_usage(name, path):
     id_root = get_id_root(name, path)
     id_subvolumes = get_id_subvolumes(path, id_root)
+    size_filesystem = get_filesystem_size(path)
 
-    command = "btrfs qgroup show --raw %s" % (path)
-    status, output = commands.getstatusoutput(command)
+    # Get disk usage from quota
+    command = "sudo btrfs qgroup show --raw %s" % (path)
+    status, output = subprocess.getstatusoutput(command)
 
-    if status is not 0:
+    if status != 0:
         raise Exception(command)
 
     lines = output.splitlines()[2:]
@@ -94,16 +117,28 @@ def get_disk_usage(name, path):
     #   -> data that is not (yet) incorporated in a snapshot
     size_data_exclusive = 0
 
-
     # Data exclusively available in snapshots
     #   -> data that was removed from volume
     size_snapshot_exclusive = 0
 
     for line in lines:
         split = line.split()
-        subvol_id = int(split[0].split("/")[1])
-        size_total = float(split[1])
-        size_exclusive = float(split[2])
+        subvol_id = 0
+        size_total = 0
+        size_exclusive = 0
+        try:
+            subvol_id = int(split[0].split("/")[1])
+            size_total = float(split[1])
+            size_exclusive = float(split[2])
+        except IndexError:
+            # ignore "WARNING: Quota disabled"
+            pass
+
+        # size_exclusive is incorrect when snapshot is
+        # removed and qgroups are not updated yet,
+        # ignore the value when it seems unrealistic
+        if size_exclusive > size_filesystem:
+            size_exclusive = 0
 
         if subvol_id == id_root:
             size_data_total = size_total
@@ -112,27 +147,57 @@ def get_disk_usage(name, path):
             size_snapshot_total += size_total
             size_snapshot_exclusive += size_exclusive
 
+
 def rescan_quota(path):
-    command = "btrfs quota rescan %s" % (path)
-    status, output = commands.getstatusoutput(command)
-    if status is not 0:
+    command = "sudo btrfs quota rescan %s" % (path)
+    status, output = subprocess.getstatusoutput(command)
+    if status != 0:
         Exception(command)
 
+
 def print_human_readable(name):
     global size_data_total
     global size_data_exclusive
     global size_snapshot_exclusive
-    size_data_total = size_data_total / (1024*1e6)
-    size_data_exclusive = size_data_exclusive / (1024*1e6)
-    size_snapshot_exclusive = size_snapshot_exclusive / (1024*1e6)
-    print "%10s: %6.1f Gb, %6.1f Gb, %6.1f Gb" % (name, size_data_total, size_data_exclusive, size_snapshot_exclusive)
+    size_data_total = size_data_total / (1024 * 1e6)
+    size_data_exclusive = size_data_exclusive / (1024 * 1e6)
+    size_snapshot_exclusive = size_snapshot_exclusive / (1024 * 1e6)
+    print(
+        "%10s: %6.1f Gb, %6.1f Gb, %6.1f Gb"
+        % (name, size_data_total, size_data_exclusive, size_snapshot_exclusive)
+    )
+
 
 def print_rrd(name):
     timestamp = int(time.time())
-    print("PUTVAL {}/exec-btrfs_{}/gauge-data_total {}:{:.1f}".format(hostname, name, timestamp, size_data_total))
-    print("PUTVAL {}/exec-btrfs_{}/gauge-data_exclusive {}:{:.1f}".format(hostname, name, timestamp, size_data_exclusive))
-    print("PUTVAL {}/exec-btrfs_{}/gauge-snapshot_total {}:{:.1f}".format(hostname, name, timestamp, size_snapshot_total))
-    print("PUTVAL {}/exec-btrfs_{}/gauge-snapshot_exclusive {}:{:.1f}".format(hostname, name, timestamp, size_snapshot_exclusive))
+    print(
+        (
+            "PUTVAL {}/exec-btrfs_{}/gauge-data_total {}:{:.1f}".format(
+                hostname, name, timestamp, size_data_total
+            )
+        )
+    )
+    print(
+        (
+            "PUTVAL {}/exec-btrfs_{}/gauge-data_exclusive {}:{:.1f}".format(
+                hostname, name, timestamp, size_data_exclusive
+            )
+        )
+    )
+    print(
+        (
+            "PUTVAL {}/exec-btrfs_{}/gauge-snapshot_total {}:{:.1f}".format(
+                hostname, name, timestamp, size_snapshot_total
+            )
+        )
+    )
+    print(
+        (
+            "PUTVAL {}/exec-btrfs_{}/gauge-snapshot_exclusive {}:{:.1f}".format(
+                hostname, name, timestamp, size_snapshot_exclusive
+            )
+        )
+    )
 
 
 #
@@ -143,19 +208,18 @@ interval = 10
 volumes = list()
 
 # 275 GB SSD
-volumes.append(["@",       "/host/root/"])
-volumes.append(["@home",   "/host/root/home"])
-volumes.append(["opt",     "/host/root/opt"])
+volumes.append(["home",   "/host/root/home"])
 
 # 2x 4TB HDD
 volumes.append(["data",    "/host/root/media/data"])
-
+volumes.append(["backup",    "/host/root/media/backup"])
+volumes.append(["seafile",    "/host/root/media/seafile"])
 
 #
 # Command line arguments
 #
-parser = argparse.ArgumentParser(description='Get BTRFS disk usage')
-parser.add_argument('-s', action='store_true', help='print in human readable format')
+parser = argparse.ArgumentParser(description="Get BTRFS disk usage")
+parser.add_argument("-s", action="store_true", help="print in human readable format")
 args = parser.parse_args()
 human_readable = args.s
 
@@ -163,7 +227,7 @@ human_readable = args.s
 #
 # Main
 #
-if (human_readable):
+if human_readable:
     for (name, path) in volumes:
         get_disk_usage(name, path)
         print_human_readable(name)
@@ -172,9 +236,9 @@ else:
     while True:
         for (name, path) in volumes:
             get_disk_usage(name, path)
-        
+
             print_rrd(name)
 
             sys.stdout.flush()
         time.sleep(interval)
-        #rescan_quota(path)
+        # rescan_quota(path)

collectd/usr/local/bin/cpufreq-data

@@ -0,0 +1,57 @@
+#!/usr/bin/python3
+import argparse
+import time
+import sys
+import os
+
+hostname = "sepia"
+measurement_interval = 5
+
+
+def get_cpu_frequencies():
+    frequencies = []
+    try:
+        cpu_dirs = [
+            d
+            for d in os.listdir("/sys/devices/system/cpu/")
+            if d.startswith("cpu") and d[3:].isdigit()
+        ]
+        for cpu_dir in cpu_dirs:
+            with open(
+                f"/sys/devices/system/cpu/{cpu_dir}/cpufreq/scaling_cur_freq", "r"
+            ) as f:
+                frequency = int(f.read().strip()) / 1000  # Convert Hz to MHz
+                frequencies.append((int(cpu_dir[3:]), frequency))
+    except Exception as e:
+        print("Error:", e)
+    return frequencies
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Query CPU frequencies.")
+    parser.add_argument(
+        "-s",
+        "--human-readable",
+        action="store_true",
+        help="Print frequencies in human-readable format",
+    )
+    args = parser.parse_args()
+
+    if args.human_readable:
+        frequencies = get_cpu_frequencies()
+        for cpu, frequency in frequencies:
+            print(f"CPU{cpu} Frequency: {frequency:.2f} MHz")
+    else:
+        while True:
+            frequencies = get_cpu_frequencies()
+            timestamp = int(time.time())
+            for cpu, frequency in frequencies:
+                print(
+                    f"PUTVAL {hostname}/cpu-frequency/gauge-cpu{cpu} {timestamp}:{frequency:.0f}"
+                )
+            sys.stdout.flush()
+            time.sleep(measurement_interval)
+
+
+if __name__ == "__main__":
+    main()

collectd/usr/local/bin/du-data

@@ -1,25 +1,77 @@
-#!/bin/bash
-COLLECTION=sepia
-INTERVAL=90
+#!/usr/bin/python3
 
-DIRS=$(cat <<LIST
-/host/root/media/data/Inverter
-/host/root/media/data/Monique
-/host/root/media/data/Music
-/host/root/media/data/Peter
-/host/root/media/data/Photographs
-/host/root/media/data/Raw
-/host/root/media/data/Sanne
-/host/root/media/data/Wii
-LIST
-)
+#
+# Imports
+#
+import sys
+import time
+import subprocess
+import argparse
 
-while :; do
-    SECONDS=0    
-    for DIR in $DIRS; do
-        SIZE=$(du -cs $DIR | tail -1 | awk '{print $1}')
-        NAME=$(echo $DIR | sed 's/.//' | tr / - )
-        echo "PUTVAL $COLLECTION/exec-du-$NAME/gauge-size interval=$INTERVAL N:$SIZE"
-    done
-    sleep $((INTERVAL-$SECONDS))
-done
+
+#
+# Methods
+#
+def get_disk_usage(path, human_readable):
+    """disk usage in human readable format (e.g. '2,1GB')"""
+    arguments = "-sh" if human_readable else "-s"
+    command = "du %s %s" % (arguments, path)
+    status, output = subprocess.getstatusoutput(command)
+
+    if status != 0:
+        raise Exception(command)
+
+    disk_usage = output.split()[0]
+    if not human_readable:
+        # du reports in units of 1024 bytes, convert to plain number of bytes
+        disk_usage = int(disk_usage) * 1024
+    return disk_usage
+
+
+#
+# Directories to scan
+#
+hostname = "sepia"
+interval = 10
+directories = list()
+
+directories.append(["inverter", "/host/root/media/data/Inverter"])
+directories.append(["monique", "/host/root/media/data/Monique"])
+directories.append(["music", "/host/root/media/data/Music"])
+directories.append(["peter", "/host/root/media/data/Peter"])
+directories.append(["photographs", "/host/root/media/data/Photographs"])
+directories.append(["sanne", "/host/root/media/data/Sanne"])
+directories.append(["wii", "/host/root/media/data/Wii"])
+
+#
+# Command line arguments
+#
+parser = argparse.ArgumentParser(description="Get DU disk usage")
+parser.add_argument("-s", action="store_true", help="print in human readable format")
+args = parser.parse_args()
+human_readable = args.s
+
+
+#
+# Main
+#
+if human_readable:
+    for (name, path) in directories:
+        disk_usage = get_disk_usage(path, human_readable)
+        print(("%s: %s" % (name, disk_usage)))
+else:
+    # RRD mode
+    while True:
+        for (name, path) in directories:
+            disk_usage = get_disk_usage(path, human_readable)
+            timestamp = int(time.time())
+            size = float(disk_usage)
+            print(
+                (
+                    "PUTVAL {}/exec-du_{}/gauge-size {}:{:.1f}".format(
+                        hostname, name, timestamp, size
+                    )
+                )
+            )
+            sys.stdout.flush()
+        time.sleep(interval)

collectd/usr/local/bin/power-data

@@ -1,4 +1,4 @@
-#!/usr/bin/python
+#!/usr/bin/python3
 
 #
 # Imports
@@ -6,61 +6,58 @@
 import sys
 import time
 import argparse
-import pylikwid
+import pmt
 
 #
 # Configuration
 #
 hostname = "sepia"
-cpuid = 0
-pinfo = pylikwid.getpowerinfo()
-domainid = pinfo.get("domains").get("PKG").get("ID")
 measurement_duration = 5
 measurement_interval = 15
-dinfo = pinfo.get("domains")
-domain_names = dinfo.keys()
-domain_ids = [domain['ID'] for domain in dinfo.values()]
+pm = pmt.create("rapl")
 
 #
 # Command line arguments
 #
-parser = argparse.ArgumentParser(description='Get CPU power consumption')
-parser.add_argument('-s', action='store_true', help='print in human readable format')
+parser = argparse.ArgumentParser(description="Get CPU power consumption")
+parser.add_argument("-s", action="store_true", help="print in human readable format")
 args = parser.parse_args()
 human_readable = args.s
 
-# 
-# Methods
-# 
-def get_power():
-    start = list()
-    end = list()
-    power = list()
-    for domain_id in domain_ids:
-        e_start = pylikwid.startpower(cpuid, domain_id)
-        start.append(e_start)
-    time.sleep(measurement_duration)
-    for domain_id in domain_ids:
-        e_stop = pylikwid.stoppower(cpuid, domain_id)
-        end.append(e_stop)
-    for events in zip(start, end, domain_ids):
-        joules = pylikwid.getpower(events[0], events[1], events[2])
-        power.append(joules / measurement_duration)
 
-    return dict(zip(domain_names, power))
+#
+# Methods
+#
+def get_power():
+    time.sleep(measurement_duration)
+    measurements = dict()
+    state = pm.read()
+    for i in range(state.nr_measurements()):
+        name = state.name(i)
+        watts = state.watts(i)
+        measurements[name] = watts
+    return measurements
+
 
 def print_rrd(measurements):
     timestamp = int(time.time())
-    for measurement in measurements.items():
+    for measurement in list(measurements.items()):
         name = measurement[0].lower()
         power = measurement[1]
-        print("PUTVAL {}/exec-power/gauge-{} {}:{:.1f}".format(hostname, name, timestamp, power))
+        print(
+            (
+                "PUTVAL {}/exec-power/gauge-{} {}:{:.1f}".format(
+                    hostname, name, timestamp, power
+                )
+            )
+        )
+
 
 #
 # Main
 #
-if (human_readable):
-    print get_power()
+if human_readable:
+    print(get_power())
 else:
     while True:
         power = get_power()

collectd/usr/local/bin/speedtest-data

@@ -1,12 +1,15 @@
 #!/bin/bash
-SPEEDTEST=/sbin/speedtest-cli
+SPEEDTEST=/usr/bin/speedtest-cli
 COLLECTION=sepia
 INTERVAL=900
 
 while :; do
     SECONDS=0 
     RESULT=($($SPEEDTEST | grep Mbit | cut -d' ' -f 2))
-    echo "PUTVAL $COLLECTION/exec-speedtest/gauge-download interval=$INTERVAL N:${RESULT[0]}"
-    echo "PUTVAL $COLLECTION/exec-speedtest/gauge-upload interval=$INTERVAL N:${RESULT[1]}"
+    TIMESTAMP=$(date +%s)
+    #echo "PUTVAL $COLLECTION/exec-speedtest/gauge-download interval=$INTERVAL N:${RESULT[0]}"
+    #echo "PUTVAL $COLLECTION/exec-speedtest/gauge-upload interval=$INTERVAL N:${RESULT[1]}"
+    echo "PUTVAL $COLLECTION/exec-speedtest/gauge-download ${TIMESTAMP}:${RESULT[0]}"
+    echo "PUTVAL $COLLECTION/exec-speedtest/gauge-upload ${TIMESTAMP}:${RESULT[1]}"
     sleep $((INTERVAL-$SECONDS))
 done

docker-compose.collectd.yaml

@@ -1,15 +1,19 @@
 services:
-    collectd:
-        container_name: collectd
-        image: collectd:latest
-        privileged: true
-        restart: unless-stopped
-        volumes:
-        - /opt/collectd/etc:/etc/collectd
-        - /opt/collectd/var:/var/lib/collectd
-        - /opt/collectd/usr:/host/usr
-        - /root/scripts/speedtest-cli:/sbin/speedtest-cli
-        - /:/host/root
-        - /media:/host/media
-        - /var/lib/docker:/media/docker
-        - /dev/mapper:/dev/mapper
+  collectd:
+    build:
+      context: /opt/collectd/docker
+      dockerfile: Dockerfile
+    container_name: collectd
+    image: collectd:bookworm
+    privileged: true
+    restart: unless-stopped
+    volumes:
+    - /opt/collectd/etc:/etc/collectd
+    - /opt/collectd/var:/var/lib/collectd
+    - /opt/collectd/usr:/host/usr
+    - /:/host/root
+    - /media:/host/media
+    - /media/jupiter/borg:/host/media/borg
+    - /media/jupiter/rsnapshot:/host/media/rsnapshot
+    - /var/lib/docker:/media/docker
+    - /dev/mapper:/dev/mapper

docker-compose.yaml

@@ -12,7 +12,6 @@ include:
 
 # Storage
 - docker-compose.seafile.yaml
-
 - docker-compose.collectd.yaml
 
 # Sensors