T2-Route-Conv: Process Crash Optimization (sonic-net#21587)

Fixes issue: sonic-net#21586

Author: deepak-singhal0408

dockers/docker-fpm-frr/base_image_files/TSA

@@ -43,6 +43,11 @@ if [ -z "$STARTED_BY_TSA_TSB_SERVICE" ]; then
      [[ $(/bin/systemctl show $service --property SubState  --value) == "running" ]]; then
       echo "Stopping $service before configuring TSA"
       systemctl stop $service
+      if sonic-db-cli STATE_DB HDEL "ALL_SERVICE_STATUS|tsa_tsb_service" "running" >/dev/null; then
+        echo "Successfully removed TSA-TSB service flag."
+      else
+        echo "Failed to remove TSA-TSB service flag!" >&2
+      fi
   fi
 fi
 

dockers/docker-fpm-frr/base_image_files/TSB

@@ -42,6 +42,11 @@ if [ -z "$STARTED_BY_TSA_TSB_SERVICE" ]; then
      [[ $(/bin/systemctl show $service --property SubState  --value) == "running" ]]; then
       echo "Stopping $service before configuring TSB"
       systemctl stop $service
+      if sonic-db-cli STATE_DB HDEL "ALL_SERVICE_STATUS|tsa_tsb_service" "running" >/dev/null; then
+        echo "Successfully removed TSA-TSB service flag."
+      else
+        echo "Failed to remove TSA-TSB service flag!" >&2
+      fi
   fi
 fi
 

files/scripts/startup_tsa_tsb.py

@@ -69,19 +69,42 @@ def config_tsa():
     if tsa_ena == True:
         logger.log_info("Configuring TSA")
         subprocess.check_output(['TSA']).strip()
+        logger.log_info("Setting TSA-TSB service field in STATE_DB")
+        subprocess.check_output([
+            'sonic-db-cli', 'STATE_DB', 'HSET', 'ALL_SERVICE_STATUS|tsa_tsb_service', 'running', 'OK'
+        ]).strip()
     else:
-        if num_asics > 1:
-            logger.log_info("Either TSA is already configured or switch sub_role is not Frontend - not configuring TSA")
+        #check if tsa_tsb service is already running, restart the timer
+        try:
+            startup_tsa_tsb_service_status = subprocess.check_output([
+                'sonic-db-cli', 'STATE_DB', 'HGET', 'ALL_SERVICE_STATUS|tsa_tsb_service', 'running'
+            ]).strip().decode('utf-8')  # Convert bytes to string
+        except subprocess.CalledProcessError:
+            startup_tsa_tsb_service_status = None  # Default if the field is missing
+
+        if startup_tsa_tsb_service_status == 'OK':
+            logger.log_info("TSA-TSB service is already running, just restart the timer")
+            return True
         else:
-            logger.log_info("Either TSA is already configured - not configuring TSA")
+            if num_asics > 1:
+                logger.log_info("Either TSA is already configured or switch sub_role is not Frontend - not configuring TSA")
+            else:
+                logger.log_info("Either TSA is already configured - not configuring TSA")
     return tsa_ena
 
 def config_tsb():
     logger.log_info("Configuring TSB")
     subprocess.check_output(['TSB']).strip()
+
+    logger.log_info("Removing the TSA-TSB service field from STATE_DB")
+    subprocess.check_output([
+        'sonic-db-cli', 'STATE_DB', 'HDEL', 'ALL_SERVICE_STATUS|tsa_tsb_service', 'running'
+    ]).strip()
+
     tsb_issued = True
     return
 
+
 def start_tsb_timer(interval):
     global timer
     logger.log_info("Starting timer with interval {} seconds to configure TSB".format(interval))

files/scripts/swss.sh

@@ -8,7 +8,7 @@ DEBUGLOG="/tmp/swss-syncd-debug$DEV.log"
 LOCKFILE="/tmp/swss-syncd-lock$DEV"
 NAMESPACE_PREFIX="asic"
 ETC_SONIC_PATH="/etc/sonic/"
-
+TSA_TSB_SERVICE="startup_tsa_tsb.service"
 
 . /usr/local/bin/asic_status.sh
 
@@ -109,9 +109,9 @@ function clean_up_tables()
 
 # This function cleans up the chassis db table entries created ONLY by this asic
 # This is used to do the clean up operation when the line card / asic reboots
-# When the asic/lc is RE-booting, the chassis db server is supposed to be running 
-# in the supervisor.  So the clean up is done when only the chassis db connectable. 
-# Otherwise no need to do the clean up since both the supervisor and line card may be 
+# When the asic/lc is RE-booting, the chassis db server is supposed to be running
+# in the supervisor.  So the clean up is done when only the chassis db connectable.
+# Otherwise no need to do the clean up since both the supervisor and line card may be
 # rebooting (the whole chassis scenario)
 # The clean up operation is required to delete only those entries created by
 # the asic that is rebooted. Entries from the following tables are deleted in the order
@@ -212,7 +212,7 @@ function clean_up_chassis_db_tables()
     debug "Chassis db clean up for ${SERVICE}$DEV. Number of SYSTEM_LAG_MEMBER_TABLE entries deleted: $num_lag_mem"
 
     # Wait for some time before deleting system lag so that the all the memebers of the
-    # system lag will be cleared. 
+    # system lag will be cleared.
     # This delay is needed only if some system lag members were deleted
 
     if [[ $num_lag_mem > 0 ]]; then
@@ -258,6 +258,12 @@ start_peer_and_dependent_services() {
     check_warm_boot
 
     if [[ x"$WARM_BOOT" != x"true" ]]; then
+        SERVICES_CONF="/usr/share/sonic/device/$PLATFORM/services.conf"
+        if [[ -f $SERVICES_CONF ]] && grep -q "^startup_tsa_tsb.service$" $SERVICES_CONF; then
+            echo "${SERVICE}$DEV: starting TSA-TSB service"
+            /bin/systemctl restart $TSA_TSB_SERVICE
+        fi
+
         for peer in ${PEER}; do
             if [[ ! -z $DEV ]]; then
                 /bin/systemctl start ${peer}@$DEV