[counters] Keep counters cache in a single directory (sonic-net#2232)

- What I did
To fix sonic-net#9817. Cache all counters in the same place. Created a UserCache helper class to access the cache directory.

- How I did it
Implemented UserCache class. Changed all stats commands to use new class. Adopted fast-reboot script.

- How to verify it
Run on the switch and verify counters stats command and clear commands work correctly. After config reload or cold/fast reboot counters cache is removed.

Signed-off-by: Stepan Blyschak <[email protected]>

Author: stepanblyschak

config/main.py

@@ -1517,11 +1517,6 @@ def reload(db, filename, yes, load_sysinfo, no_service_restart, disable_arp_cach
     if multi_asic.is_multi_asic() and file_format == 'config_db':
         num_cfg_file += num_asic
 
-    # Remove cached PG drop counters data
-    dropstat_dir_prefix = '/tmp/dropstat'
-    command = "rm -rf {}-*".format(dropstat_dir_prefix)
-    clicommon.run_command(command, display_cmd=True)
-
     # If the user give the filename[s], extract the file names.
     if filename is not None:
         cfg_files = filename.split(',')

config/plugins/pbh.py

@@ -6,13 +6,14 @@
 CLI Auto-generation tool HLD - https://github.com/Azure/SONiC/pull/78
 """
 
+import os
 import click
 import json
 import ipaddress
 import re
 import utilities_common.cli as clicommon
 
-from show.plugins.pbh import deserialize_pbh_counters
+from show.plugins.pbh import deserialize_pbh_counters, PBH_COUNTERS_CACHE_FILENAME
 
 GRE_KEY_RE = r"^(0x){1}[a-fA-F0-9]{1,8}/(0x){1}[a-fA-F0-9]{1,8}$"
 
@@ -79,8 +80,6 @@
 PBH_UPDATE = "UPDATE"
 PBH_REMOVE = "REMOVE"
 
-PBH_COUNTERS_LOCATION = "/tmp/.pbh_counters.txt"
-
 #
 # DB interface --------------------------------------------------------------------------------------------------------
 #
@@ -467,11 +466,14 @@ def serialize_pbh_counters(obj):
             obj: counters dict.
     """
 
+    cache = clicommon.UserCache('pbh')
+    counters_cache_file = os.path.join(cache.get_directory(), PBH_COUNTERS_CACHE_FILENAME)
+
     def remap_keys(obj):
         return [{'key': k, 'value': v} for k, v in obj.items()]
 
     try:
-        with open(PBH_COUNTERS_LOCATION, 'w') as f:
+        with open(counters_cache_file, 'w') as f:
             json.dump(remap_keys(obj), f)
     except IOError as err:
         pass

scripts/aclshow

@@ -20,15 +20,13 @@ optional arguments:
 import argparse
 import json
 import os
-from swsscommon.swsscommon import SonicV2Connector, ConfigDBConnector
 import sys
 
+from swsscommon.swsscommon import SonicV2Connector, ConfigDBConnector
+from utilities_common.cli import UserCache
+
 from tabulate import tabulate
 
-### temp file to save counter positions when doing clear counter action.
-### if we could have a SAI command to clear counters will be better, so no need to maintain
-### counters in temp loaction for clear conter action
-COUNTER_POSITION = '/tmp/.counters_acl.p'
 COUNTERS = "COUNTERS"
 ACL_COUNTER_RULE_MAP = "ACL_COUNTER_RULE_MAP"
 
@@ -38,6 +36,9 @@ ACL_HEADER = ["RULE NAME", "TABLE NAME", "PRIO", "PACKETS COUNT", "BYTES COUNT"]
 COUNTER_PACKETS_ATTR = "SAI_ACL_COUNTER_ATTR_PACKETS"
 COUNTER_BYTES_ATTR = "SAI_ACL_COUNTER_ATTR_BYTES"
 
+USER_CACHE = UserCache()
+COUNTERS_CACHE_DIR = USER_CACHE.get_directory()
+COUNTERS_CACHE = os.path.join(COUNTERS_CACHE_DIR, 'aclstat')
 
 class AclStat(object):
     """
@@ -78,9 +79,9 @@ class AclStat(object):
                 res[e['key'][0], e['key'][1]] = e['value']
             return res
 
-        if os.path.isfile(COUNTER_POSITION):
+        if os.path.isfile(COUNTERS_CACHE):
             try:
-                with open(COUNTER_POSITION) as fp:
+                with open(COUNTERS_CACHE) as fp:
                     self.saved_acl_counters = remap_keys(json.load(fp))
             except Exception:
                 pass
@@ -207,7 +208,7 @@ class AclStat(object):
         def remap_keys(dict):
             return [{'key': k, 'value': v} for k, v in dict.items()]
 
-        with open(COUNTER_POSITION, 'w') as fp:
+        with open(COUNTERS_CACHE, 'w') as fp:
             json.dump(remap_keys(self.acl_counters), fp)
 
 def main():

scripts/dropstat

@@ -35,6 +35,7 @@ except KeyError:
     pass
 
 from swsscommon.swsscommon import SonicV2Connector, ConfigDBConnector
+from utilities_common.cli import UserCache
 
 
 # COUNTERS_DB Tables
@@ -80,8 +81,7 @@ std_switch_description_header = ['DEVICE']
 
 
 def get_dropstat_dir():
-    dropstat_dir_prefix = '/tmp/dropstat'
-    return "{}-{}/".format(dropstat_dir_prefix, os.getuid())
+    return UserCache().get_directory()
 
 
 class DropStat(object):
@@ -411,18 +411,7 @@ Examples:
     group = args.group
     counter_type = args.type
 
-    dropstat_dir = get_dropstat_dir()
-
-    # Create the directory to hold clear results
-    if not os.path.exists(dropstat_dir):
-        try:
-            os.makedirs(dropstat_dir)
-        except IOError as e:
-            print(e)
-            sys.exit(e.errno)
-
     dcstat = DropStat()
-
     if command == 'clear':
         dcstat.clear_drop_counts()
     elif command == 'show':

scripts/fast-reboot

@@ -462,21 +462,14 @@ function unload_kernel()
 }
 
 function save_counters_folder() {
-    debug "Saving counters folder before warmboot..."
+    if [[ "$REBOOT_TYPE" = "warm-reboot" ]]; then
+        debug "Saving counters folder before warmboot..."
 
-    counters_folder="/host/counters"
-    if [[ ! -d $counters_folder ]]; then
-        mkdir $counters_folder
-    fi
-    if [[ "$REBOOT_TYPE" = "warm-reboot" || "$REBOOT_TYPE" = "fastfast-reboot" ]]; then
-        modules=("portstat-0" "dropstat" "pfcstat-0" "queuestat-0" "intfstat-0")
-        for module in ${modules[@]}
-        do
-            statfile="/tmp/$module"
-            if [[ -d $statfile ]]; then
-                cp -rf $statfile $counters_folder
-            fi
-        done
+        counters_folder="/host/counters"
+        if [[ ! -d $counters_folder ]]; then
+            mkdir $counters_folder
+        fi
+        cp -rf /tmp/cache $counters_folder
     fi
 }
 

scripts/flow_counters_stat

@@ -27,6 +27,7 @@ import utilities_common.multi_asic as multi_asic_util
 from flow_counter_util.route import build_route_pattern, extract_route_pattern, exit_if_route_flow_counter_not_support, DEFAULT_VRF, COUNTERS_ROUTE_TO_PATTERN_MAP
 from utilities_common import constants
 from utilities_common.netstat import format_number_with_comma, table_as_json, ns_diff, format_prate
+from utilities_common.cli import UserCache
 
 # Flow counter meta data, new type of flow counters can extend this dictinary to reuse existing logic
 flow_counter_meta = {
@@ -57,9 +58,10 @@ class FlowCounterStats(object):
         meta_data = flow_counter_meta[args.type]
         self.name_map = meta_data['name_map']
         self.headers = meta_data['headers']
-        self.data_file = os.path.join('/tmp/{}-stats-{}'.format(args.type, os.getuid()))
-        if self.args.delete and os.path.exists(self.data_file):
-            os.remove(self.data_file)
+        self.cache = UserCache()
+        self.data_file = os.path.join(self.cache.get_directory(), "flow-counter-stats")
+        if self.args.delete:
+            self.cache.remove()
         self.data = {}
 
     def show(self):

scripts/intfstat

@@ -28,6 +28,7 @@ from collections import namedtuple, OrderedDict
 from natsort import natsorted
 from tabulate import tabulate
 from utilities_common.netstat import ns_diff, table_as_json, STATUS_NA, format_brate, format_prate
+from utilities_common.cli import UserCache
 from swsscommon.swsscommon import SonicV2Connector
 
 nstat_fields = (
@@ -274,63 +275,34 @@ def main():
     delete_saved_stats = args.delete
     delete_all_stats = args.delete_all
     use_json = args.json
-    tag_name = args.tag if args.tag else ""
-    uid = str(os.getuid())
+    tag_name = args.tag
     wait_time_in_seconds = args.period
     interface_name = args.interface if args.interface else ""
 
-    # fancy filename with dashes: uid-tag / uid etc
-    filename_components = [uid, tag_name]
+    cnstat_file = "intfstat"
 
-    cnstat_file = "-".join(filter(None, filename_components))
+    cache = UserCache(tag=tag_name)
 
-    cnstat_dir = "/tmp/intfstat-" + uid
+    cache_general = UserCache()
+    cnstat_dir = cache.get_directory()
+    cnstat_general_dir = cache_general.get_directory()
+
+    cnstat_fqn_general_file = cnstat_general_dir + "/" + cnstat_file
     cnstat_fqn_file = cnstat_dir + "/" + cnstat_file
 
     if delete_all_stats:
-        # There is nothing to delete
-        if not os.path.isdir(cnstat_dir):
-            sys.exit(0)
-
-        for file in os.listdir(cnstat_dir):
-            os.remove(cnstat_dir + "/" + file)
-
-        try:
-            os.rmdir(cnstat_dir)
-            sys.exit(0)
-        except IOError as e:
-            print(e.errno, e)
-            sys.exit(e)
+        cache.remove_all()
 
     if delete_saved_stats:
-        try:
-            os.remove(cnstat_fqn_file)
-        except IOError as e:
-            if e.errno != ENOENT:
-                print(e.errno, e)
-                sys.exit(1)
-        finally:
-            if os.listdir(cnstat_dir) == []:
-                os.rmdir(cnstat_dir)
-            sys.exit(0)
+        cache.remove()
 
     intfstat = Intfstat()
     cnstat_dict, ratestat_dict = intfstat.get_cnstat(rif=interface_name)
 
-    # At this point, either we'll create a file or open an existing one.
-    if not os.path.exists(cnstat_dir):
-        try:
-            os.makedirs(cnstat_dir)
-        except IOError as e:
-            print(e.errno, e)
-            sys.exit(1)
-
     if save_fresh_stats:
         try:
             # Add the information also to the general file - i.e. without the tag name
-            if tag_name != '' and tag_name in cnstat_fqn_file.split('/')[-1]:
-                gen_index = cnstat_fqn_file.rfind('/')
-                cnstat_fqn_general_file = cnstat_fqn_file[:gen_index] + cnstat_fqn_file[gen_index:].split('-')[0]
+            if tag_name is not None:
                 if os.path.isfile(cnstat_fqn_general_file):
                     try:
                         general_data = pickle.load(open(cnstat_fqn_general_file, 'rb'))
@@ -354,9 +326,6 @@ def main():
             sys.exit(0)
 
     if wait_time_in_seconds == 0:
-        gen_index = cnstat_fqn_file.rfind('/')
-        cnstat_fqn_general_file = cnstat_fqn_file[:gen_index] + cnstat_fqn_file[gen_index:].split('-')[0]
-
         if os.path.isfile(cnstat_fqn_file) or (os.path.isfile(cnstat_fqn_general_file)):
             try:
                 cnstat_cached_dict = {}

scripts/pfcstat

@@ -18,9 +18,6 @@ from natsort import natsorted
 from tabulate import tabulate
 
 from sonic_py_common.multi_asic import get_external_ports
-from utilities_common.netstat import ns_diff, STATUS_NA, format_number_with_comma
-from utilities_common import multi_asic as multi_asic_util
-from utilities_common import constants
 
 # mock the redis for unit test purposes #
 try:
@@ -37,6 +34,12 @@ try:
 except KeyError:
     pass
 
+from utilities_common.netstat import ns_diff, STATUS_NA, format_number_with_comma
+from utilities_common import multi_asic as multi_asic_util
+from utilities_common import constants
+from utilities_common.cli import UserCache
+
+
 PStats = namedtuple("PStats", "pfc0, pfc1, pfc2, pfc3, pfc4, pfc5, pfc6, pfc7")
 header_Rx = ['Port Rx', 'PFC0', 'PFC1', 'PFC2', 'PFC3', 'PFC4', 'PFC5', 'PFC6', 'PFC7']
 
@@ -224,10 +227,10 @@ Examples:
     save_fresh_stats = args.clear
     delete_all_stats = args.delete
 
-    uid = str(os.getuid())
-    cnstat_file = uid
+    cache = UserCache()
+    cnstat_file = 'pfcstat'
 
-    cnstat_dir = os.path.join(os.sep, "tmp", "pfcstat-{}".format(uid))
+    cnstat_dir = cache.get_directory()
     cnstat_fqn_file_rx = os.path.join(cnstat_dir, "{}rx".format(cnstat_file))
     cnstat_fqn_file_tx = os.path.join(cnstat_dir, "{}tx".format(cnstat_file))
 
@@ -239,15 +242,7 @@ Examples:
     pfcstat = Pfcstat(args.namespace, args.show)
 
     if delete_all_stats:
-        for file in os.listdir(cnstat_dir):
-            os.remove(os.path.join(cnstat_dir, file))
-
-        try:
-            os.rmdir(cnstat_dir)
-            sys.exit(0)
-        except IOError as e:
-            print(e.errno, e)
-            sys.exit(e)
+        cache.remove()
 
     """
         Get the counters of pfc rx counter
@@ -259,14 +254,6 @@ Examples:
     """
     cnstat_dict_tx = deepcopy(pfcstat.get_cnstat(False))
 
-    # At this point, either we'll create a file or open an existing one.
-    if not os.path.exists(cnstat_dir):
-        try:
-            os.makedirs(cnstat_dir)
-        except IOError as e:
-            print(e.errno, e)
-            sys.exit(1)
-
     if save_fresh_stats:
         try:
             pickle.dump(cnstat_dict_rx, open(cnstat_fqn_file_rx, 'wb'))

scripts/pg-drop

@@ -26,6 +26,7 @@ try:
 except KeyError:
     pass
 
+from utilities_common.cli import UserCache
 from swsscommon.swsscommon import ConfigDBConnector, SonicV2Connector
 
 STATUS_NA = 'N/A'
@@ -38,8 +39,7 @@ COUNTERS_PG_PORT_MAP = "COUNTERS_PG_PORT_MAP"
 COUNTERS_PG_INDEX_MAP = "COUNTERS_PG_INDEX_MAP"
 
 def get_dropstat_dir():
-    dropstat_dir_prefix = '/tmp/dropstat'
-    return "{}-{}/".format(dropstat_dir_prefix, os.getuid())
+    return UserCache().get_directory()
 
 class PgDropStat(object):