From cfee03142a2bdc90fe55f41a8a78467f7429c7eb Mon Sep 17 00:00:00 2001
From: Roy Le <roy.le@viindoo.com>
Date: Thu, 12 Mar 2026 13:10:08 +0700
Subject: [PATCH] [MISC] odoo/service: scale up by one worker when all workers
 are busy, the total number of workers is limited to 1.5x

---
 odoo/service/server.py | 27 +++++++++++++++++++++++++++
 1 file changed, 27 insertions(+)

diff --git a/odoo/service/server.py b/odoo/service/server.py
index b11fe731c9f24..8752f2def8571 100644
--- a/odoo/service/server.py
+++ b/odoo/service/server.py
@@ -4,10 +4,12 @@
 import datetime
 import errno
 import logging
+import multiprocessing
 import os
 import os.path
 import platform
 import random
+import re
 import select
 import signal
 import socket
@@ -67,6 +69,7 @@
 _logger = logging.getLogger(__name__)
 
 SLEEP_INTERVAL = 60     # 1 min
+WORKER_BUSY = multiprocessing.Value('i', 0)
 
 def memory_info(process):
     """
@@ -943,6 +946,26 @@ def process_spawn(self):
         if config['http_enable']:
             while len(self.workers_http) < self.population:
                 self.worker_spawn(WorkerHTTP, self.workers_http)
+            # Scale up by at most one additional worker when all current workers
+            # are busy. The total number of workers is capped at 1.5x the configured
+            # population to avoid uncontrolled worker growth under heavy load.
+            # NOTE: In our current deployment, all instances are restarted daily,
+            # so any extra workers created during traffic spikes will be reset
+            # back to the base population after restart.
+            # TODO:
+            # Implement automatic scale-down for idle workers in the future
+            # (e.g., terminate extra workers after being idle for a certain time)
+            # to avoid keeping unnecessary workers alive.
+            if (WORKER_BUSY.value and len(self.workers_http) == WORKER_BUSY.value):
+                if len(self.workers_http) < self.population + self.population // 2:
+                    worker = self.worker_spawn(WorkerHTTP, self.workers_http)
+                    _logger.info(
+                        f"Scaling up {worker.__class__.__name__} by 1 (busy={WORKER_BUSY.value}, pid={worker.pid})"
+                    )
+                else:
+                    _logger.info(
+                        f"{WorkerHTTP.__name__} scale-up skipped: max workers reached ({self.population + self.population // 2})"
+                    )
             if not self.long_polling_pid:
                 self.long_polling_spawn()
         while len(self.workers_cron) < config['max_cron_threads']:
@@ -1203,7 +1226,11 @@ def process_request(self, client, addr):
         # tolerate broken pipe when the http client closes the socket before
         # receiving the full reply
         try:
+            with WORKER_BUSY.get_lock():
+                WORKER_BUSY.value += 1
             self.server.process_request(client, addr)
+            with WORKER_BUSY.get_lock():
+                WORKER_BUSY.value -= 1
         except IOError as e:
             if e.errno != errno.EPIPE:
                 raise