Add daemon which periodically pushes process and docker stats to State DB (#3525)

files/build_templates/sonic_debian_extension.j2

@@ -253,6 +253,11 @@ sudo cp $IMAGE_CONFIGS/caclmgrd/caclmgrd.service  $FILESYSTEM_ROOT/etc/systemd/s
 echo "caclmgrd.service" | sudo tee -a $GENERATED_SERVICE_FILE
 sudo cp $IMAGE_CONFIGS/caclmgrd/caclmgrd $FILESYSTEM_ROOT/usr/bin/
 
+# Copy process/docker cpu/memory utilization data export daemon
+sudo cp $IMAGE_CONFIGS/procdockerstatsd/procdockerstatsd.service  $FILESYSTEM_ROOT/etc/systemd/system/
+echo "procdockerstatsd.service" | sudo tee -a $GENERATED_SERVICE_FILE
+sudo cp $IMAGE_CONFIGS/procdockerstatsd/procdockerstatsd $FILESYSTEM_ROOT/usr/bin/
+
 # Copy process-reboot-cause service files
 sudo cp $IMAGE_CONFIGS/process-reboot-cause/process-reboot-cause.service  $FILESYSTEM_ROOT/etc/systemd/system/
 echo "process-reboot-cause.service" | sudo tee -a $GENERATED_SERVICE_FILE

files/image_config/procdockerstatsd/procdockerstatsd

@@ -0,0 +1,214 @@
+# !/usr/bin/env python
+'''
+procdockerstatsd
+Daemon which periodically gathers process and docker statistics and pushes the data to STATE_DB
+'''
+
+import os
+import re
+import subprocess
+import sys
+import syslog
+import time
+from datetime import datetime
+
+import swsssdk
+
+VERSION = '1.0'
+
+SYSLOG_IDENTIFIER = "procdockerstatsd"
+
+REDIS_HOSTIP = "127.0.0.1"
+
+# ========================== Syslog wrappers ==========================
+def log_info(msg, also_print_to_console=False):
+    syslog.openlog(SYSLOG_IDENTIFIER)
+    syslog.syslog(syslog.LOG_INFO, msg)
+    syslog.closelog()
+
+def log_warning(msg, also_print_to_console=False):
+    syslog.openlog(SYSLOG_IDENTIFIER)
+    syslog.syslog(syslog.LOG_WARNING, msg)
+    syslog.closelog()
+
+def log_error(msg, also_print_to_console=False):
+    syslog.openlog(SYSLOG_IDENTIFIER)
+    syslog.syslog(syslog.LOG_ERR, msg)
+    syslog.closelog()
+
+# ========================== ProcessDocker class ==========================
+class ProcDockerStats:
+
+    def __init__(self):
+        self.state_db = swsssdk.SonicV2Connector(host=REDIS_HOSTIP)
+        self.state_db.connect("STATE_DB") 
+    
+    def run_command(self, cmd):
+        proc = subprocess.Popen(cmd, shell=True, stdout=subprocess.PIPE)
+        (stdout, stderr) = proc.communicate()
+        if proc.returncode != 0:
+            log_error("Error running command '{}'".format(cmd))
+            return None
+        else:
+            return stdout
+
+    def format_docker_cmd_output(self, cmdout):
+        lines = re.split("\n", cmdout)
+        keys = re.split("   +", lines[0])
+        docker_data = dict()
+        docker_data_list = []
+        for item in lines[1:]:
+            values1 = re.split("   +", item)
+            docker_data =  dict(zip(keys, values1))
+            docker_data_list.append(docker_data)
+        formatted_dict = self.create_docker_dict(docker_data_list)
+        return formatted_dict
+
+    def format_process_cmd_output(self, cmdout):
+        lines = re.split("\n", cmdout)
+        keys = re.split(" +", lines[0])
+        keylist = list(filter(None, keys))
+        process_data = dict()
+        process_data_list = []
+        for item in lines[1:]:
+            values1 = re.split(" +", str(item))
+            # To remove extra space before UID
+            val = list(filter(None, values1))
+            # Merging extra columns created due to space in cmd ouput
+            val[8:] = [''.join(val[8:])] 
+            process_data = dict(zip(keylist, val))
+            process_data_list.append(process_data)
+        return process_data_list
+
+    def convert_to_bytes(self, value):
+        unit_value = re.search('[a-zA-Z]+', value)
+        value_to_convert = float(filter(str.isdigit, value))
+        unit = unit_value.group(0)
+        UNITS_B = 'B'
+        UNITS_KB = 'KB'
+        UNITS_MB = 'MB'
+        UNITS_MiB = 'MiB'
+        UNITS_GiB = 'GiB'
+        if unit.lower() == UNITS_B.lower():
+            return int(round(value_to_convert))
+        elif unit.lower() == UNITS_KB.lower():
+            value_converted = value_to_convert * 1000
+            return int(round(value_converted))
+        elif unit.lower() == UNITS_MB.lower():
+            value_converted = value_to_convert * 1000 * 1000
+            return int(round(value_converted))
+        elif unit.lower() == UNITS_MiB.lower():
+            value_converted = value_to_convert * 1024 * 1024
+            return int(round(value_converted))
+        elif unit.lower() == UNITS_GiB.lower():
+            value_converted = value_to_convert * 1024 * 1024 * 1024
+            return int(round(value_converted))
+
+    def create_docker_dict(self, dict_list):
+        dockerdict = {}
+        for row in dict_list[0:]:
+            cid = row.get('CONTAINER ID')
+            if cid:
+                key = 'DOCKER_STATS|' + str(cid) 
+                dockerdict[key] = {}
+                dockerdict[key]['NAME'] = row.get('NAME')
+                
+                splitcol = row.get('CPU %')
+                cpu = re.split("%", str(splitcol))
+                dockerdict[key]['CPU%'] = str(cpu[0])
+
+                splitcol = row.get('MEM USAGE / LIMIT')
+                memuse = re.split(" / ", str(splitcol))
+                # converting MiB and GiB to bytes
+                dockerdict[key]['MEM_BYTES'] = str(self.convert_to_bytes(memuse[0]))
+                dockerdict[key]['MEM_LIMIT_BYTES'] = str(self.convert_to_bytes(memuse[1]))
+
+                splitcol = row.get('MEM %')
+                mem = re.split("%", str(splitcol))
+                dockerdict[key]['MEM%'] = str(mem[0])
+
+                splitcol = row.get('NET I/O')
+                netio = re.split(" / ", str(splitcol))
+                dockerdict[key]['NET_IN_BYTES'] = str(self.convert_to_bytes(netio[0]))
+                dockerdict[key]['NET_OUT_BYTES'] = str(self.convert_to_bytes(netio[1]))
+                
+                splitcol = row.get('BLOCK I/O')
+                blockio = re.split(" / ", str(splitcol))
+                dockerdict[key]['BLOCK_IN_BYTES'] = str(self.convert_to_bytes(blockio[0]))
+                dockerdict[key]['BLOCK_OUT_BYTES'] = str(self.convert_to_bytes(blockio[1]))
+
+                dockerdict[key]['PIDS'] = row.get('PIDS')
+        return dockerdict
+
+    def update_dockerstats_command(self):
+        cmd = "docker stats --no-stream -a"
+        data = self.run_command(cmd)
+        if not data:
+            log_error("'{}' returned null output".format(cmd))
+            return False
+        dockerdata = self.format_docker_cmd_output(data)
+        if not dockerdata:
+            log_error("formatting for docker output failed")
+            return False
+        # wipe out all data from state_db before updating
+        self.state_db.delete_all_by_pattern('STATE_DB', 'DOCKER_STATS|*')
+        for k1,v1 in dockerdata.iteritems():
+            for k2,v2 in v1.iteritems():
+                self.update_state_db(k1, k2, v2)
+        return True
+
+    def update_processstats_command(self):
+        data = self.run_command("ps -eo uid,pid,ppid,%mem,%cpu,stime,tty,time,cmd --sort -%cpu | head -1024")
+        processdata = self.format_process_cmd_output(data)
+        value = ""
+        # wipe out all data before updating with new values
+        self.state_db.delete_all_by_pattern('STATE_DB', 'PROCESS_STATS|*')
+        for row in processdata[0:]:
+            cid = row.get('PID')
+            if cid:
+                value = 'PROCESS_STATS|' + str(cid)
+                uid = row.get('UID')
+                self.update_state_db(value, 'UID', uid)
+                ppid = row.get('PPID')
+                self.update_state_db(value, 'PPID', ppid)
+                cpu = row.get('%CPU')
+                self.update_state_db(value, '%CPU', str(cpu))
+                mem = row.get('%MEM')
+                self.update_state_db(value, '%MEM', str(mem))
+                stime = row.get('STIME')
+                self.update_state_db(value, 'STIME', stime)
+                tty = row.get('TT')
+                self.update_state_db(value, 'TT', tty)
+                time = row.get('TIME')
+                self.update_state_db(value, 'TIME', time)
+                cmd = row.get('CMD')
+                self.update_state_db(value, 'CMD', cmd)
+
+    def update_state_db(self, key1, key2, value2):
+        self.state_db.set('STATE_DB', key1, key2, value2)
+
+    def run(self):
+        self.update_dockerstats_command()
+        datetimeobj = datetime.now()
+        # Adding key to store latest update time.
+        self.update_state_db('DOCKER_STATS|LastUpdateTime', 'lastupdate', datetimeobj)
+        self.update_processstats_command()
+        self.update_state_db('PROCESS_STATS|LastUpdateTime', 'lastupdate', datetimeobj)
+
+# main start
+def main():
+    log_info("process-docker stats daemon starting up..")
+    if not os.getuid() == 0:
+        log_error("Must be root to run process-docker daemon")
+        print "Error: Must be root to run process-docker daemon"
+        sys.exit(1)
+    pd = ProcDockerStats()
+    # Data need to be updated every 2 mins. hence adding delay of 120 seconds
+    while True:
+        pd.run()
+        time.sleep(120)
+    log_info("process-docker stats daemon exited")
+
+if __name__ == '__main__':
+    main()
+

files/image_config/procdockerstatsd/procdockerstatsd.service

@@ -0,0 +1,13 @@
+[Unit]
+Description=Process and docker CPU/memory utilization data export daemon
+Requires=database.service updategraph.service
+After=database.service updategraph.service
+
+[Service]
+Type=simple
+ExecStart=/usr/bin/procdockerstatsd
+Restart=Always
+
+[Install]
+WantedBy=multi-user.target
+