From 172f0c055bd9b24ff9b46bf3aa793313aecdb5c7 Mon Sep 17 00:00:00 2001
From: Vaibhav Hemant Dixit <vaibhav.dixit@microsoft.com>
Date: Thu, 20 Apr 2023 22:33:06 +0000
Subject: [PATCH 1/3] LAG keepalive script to reduce lacp session wait during
 warm upgrade

---
 scripts/fast-reboot      |   8 +++
 scripts/lag_keepalive.py | 102 +++++++++++++++++++++++++++++++++++++++
 setup.py                 |   1 +
 3 files changed, 111 insertions(+)
 create mode 100644 scripts/lag_keepalive.py

diff --git a/scripts/fast-reboot b/scripts/fast-reboot
index eea97e792b..99a631046f 100755
--- a/scripts/fast-reboot
+++ b/scripts/fast-reboot
@@ -17,6 +17,7 @@ STRICT=no
 REBOOT_METHOD="/sbin/kexec -e"
 ASSISTANT_IP_LIST=""
 ASSISTANT_SCRIPT="/usr/local/bin/neighbor_advertiser"
+LAG_KEEPALIVE_SCRIPT="/usr/local/bin/lag_keepalive.py"
 WATCHDOG_UTIL="/usr/local/bin/watchdogutil"
 DEVPATH="/usr/share/sonic/device"
 PLATFORM=$(sonic-cfggen -H -v DEVICE_METADATA.localhost.platform)
@@ -682,6 +683,13 @@ fi
 # disable trap-handlers which were set before
 trap '' EXIT HUP INT QUIT TERM KILL ABRT ALRM
 
+# start sending LACPDUs to keep the LAGs refreshed
+# this is a non-blocking call, and the process will die in 300s
+debug "Starting lag_keepalive to send LACPDUs ..."
+timeout 300 python ${LAG_KEEPALIVE_SCRIPT} &
+# give the lag_keepalive script a chance to get ready (30s) and collect one lacpdu before going down (30s)
+sleep 60
+
 if [ -x ${LOG_SSD_HEALTH} ]; then
     debug "Collecting logs to check ssd health before ${REBOOT_TYPE}..."
     ${LOG_SSD_HEALTH}
diff --git a/scripts/lag_keepalive.py b/scripts/lag_keepalive.py
new file mode 100644
index 0000000000..d33cb3580f
--- /dev/null
+++ b/scripts/lag_keepalive.py
@@ -0,0 +1,102 @@
+#!/usr/bin/env python3
+
+from scapy.config import conf
+conf.ipv6_enabled = False
+from scapy.all import sendp, sniff
+from swsssdk import ConfigDBConnector
+import time, threading, traceback
+import syslog
+
+SYSLOG_ID = 'lag_keepalive'
+
+
+def log_info(msg):
+    syslog.openlog(SYSLOG_ID)
+    syslog.syslog(syslog.LOG_INFO, msg)
+    syslog.closelog()
+
+
+def log_error(msg):
+    syslog.openlog(SYSLOG_ID)
+    syslog.syslog(syslog.LOG_ERR, msg)
+    syslog.closelog()
+
+
+def sniff_lacpdu(device_mac, lag_member, lag_member_to_packet):
+    sniffed_packet = sniff(iface=lag_member,
+        filter="ether proto 0x8809 and ether src {}".format(device_mac),
+        count=1, timeout=30)
+    lag_member_to_packet[lag_member] = sniffed_packet
+
+
+def get_lacpdu_per_lag_member():
+    appDB = ConfigDBConnector()
+    appDB.db_connect('APPL_DB')
+    appDB_lag_info = appDB.get_keys('LAG_MEMBER_TABLE')
+    configDB = ConfigDBConnector()
+    configDB.db_connect('CONFIG_DB')
+    device_mac = configDB.get(configDB.CONFIG_DB,  "DEVICE_METADATA|localhost", "mac")
+    hwsku = configDB.get(configDB.CONFIG_DB,  "DEVICE_METADATA|localhost", "hwsku")
+    active_lag_members = list()
+    lag_member_to_packet = dict()
+    sniffer_threads = list()
+    for lag_entry in appDB_lag_info:
+        lag_name = str(lag_entry[0])
+        oper_status = appDB.get(appDB.APPL_DB,"LAG_TABLE:{}".format(lag_name), "oper_status")
+        if oper_status == "up":
+            # only apply the workaround for active lags
+            lag_member = str(lag_entry[1])
+            active_lag_members.append(lag_member)
+            # use threading to capture lacpdus from several lag members simultaneously
+            sniffer_thread = threading.Thread(target=sniff_lacpdu,
+                args=(device_mac, lag_member, lag_member_to_packet))
+            sniffer_thread.start()
+            sniffer_threads.append(sniffer_thread)
+
+    # sniff for lacpdu should finish in <= 30s. sniff timeout is also set to 30s
+    for sniffer in sniffer_threads:
+        sniffer.join(timeout=30)
+    
+    return active_lag_members, lag_member_to_packet
+
+
+def lag_keepalive(lag_member_to_packet):
+    while True:
+        for lag_member, packet in lag_member_to_packet.items():
+            try:
+                sendp(packet, iface=lag_member, verbose=False)
+            except Exception:
+                # log failure and continue to send lacpdu
+                traceback_msg = traceback.format_exc()
+                log_error("Failed to send LACPDU packet from interface {} with error: {}".format(
+                    lag_member, traceback_msg))
+                continue
+        log_info("sent LACPDU packets via {}".format(lag_member_to_packet.keys()))
+        time.sleep(1)
+
+
+def main():
+    while True:
+        try:
+            active_lag_members, lag_member_to_packet = get_lacpdu_per_lag_member()
+            if len(active_lag_members) != len(active_lag_members):
+                log_error("Failed to capture LACPDU packets for some lag members. " +\
+                "Active lag members: {}. LACPDUs captured for: {}".format(
+                    active_lag_members, active_lag_members.keys()))
+
+            log_info("ready to send LACPDU packets via {}".format(lag_member_to_packet.keys()))
+        except Exception:
+            traceback_msg = traceback.format_exc()
+            log_error("Failed to get LAG members and LACPDUs with error: {}".format(
+                traceback_msg))
+            # keep attempting until sniffed packets are ready
+            continue
+        # if no exceptions are thrown, break from loop as LACPDUs are ready to be sent
+        break
+
+    if lag_member_to_packet:
+        # start an infinite loop to keep sending lacpdus from lag member ports
+        lag_keepalive(lag_member_to_packet)
+
+if __name__ == "__main__":
+    main()
diff --git a/setup.py b/setup.py
index a2c851998f..d369e4947c 100644
--- a/setup.py
+++ b/setup.py
@@ -139,6 +139,7 @@
         'scripts/intfutil',
         'scripts/intfstat',
         'scripts/ipintutil',
+        'scripts/lag_keepalive.py',
         'scripts/lldpshow',
         'scripts/log_ssd_health',
         'scripts/mellanox_buffer_migrator.py',

From 882cb1225731c99bf66cf85b4e04445787214339 Mon Sep 17 00:00:00 2001
From: Vaibhav Hemant Dixit <vaibhav.dixit@microsoft.com>
Date: Wed, 26 Apr 2023 23:09:42 +0000
Subject: [PATCH 2/3] Fix the check for missed lags and corresponding error

---
 scripts/lag_keepalive.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/scripts/lag_keepalive.py b/scripts/lag_keepalive.py
index d33cb3580f..80575d6e6a 100644
--- a/scripts/lag_keepalive.py
+++ b/scripts/lag_keepalive.py
@@ -79,10 +79,10 @@ def main():
     while True:
         try:
             active_lag_members, lag_member_to_packet = get_lacpdu_per_lag_member()
-            if len(active_lag_members) != len(active_lag_members):
+            if len(active_lag_members) != len(lag_member_to_packet.keys()):
                 log_error("Failed to capture LACPDU packets for some lag members. " +\
                 "Active lag members: {}. LACPDUs captured for: {}".format(
-                    active_lag_members, active_lag_members.keys()))
+                    active_lag_members, lag_member_to_packet.keys()))
 
             log_info("ready to send LACPDU packets via {}".format(lag_member_to_packet.keys()))
         except Exception:

From eee54155779765b9ed7a778a0aa5058cc30d2d88 Mon Sep 17 00:00:00 2001
From: Vaibhav Hemant Dixit <vaibhav.dixit@microsoft.com>
Date: Thu, 4 May 2023 05:55:56 +0000
Subject: [PATCH 3/3] Fix incorrect import

---
 scripts/lag_keepalive.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/scripts/lag_keepalive.py b/scripts/lag_keepalive.py
index 80575d6e6a..04ac01a074 100644
--- a/scripts/lag_keepalive.py
+++ b/scripts/lag_keepalive.py
@@ -3,7 +3,7 @@
 from scapy.config import conf
 conf.ipv6_enabled = False
 from scapy.all import sendp, sniff
-from swsssdk import ConfigDBConnector
+from swsscommon.swsscommon import ConfigDBConnector
 import time, threading, traceback
 import syslog