diff --git a/partitioning/zabbix_partitioning.conf b/partitioning/zabbix_partitioning.conf
index 02899c8..2d211da 100644
--- a/partitioning/zabbix_partitioning.conf
+++ b/partitioning/zabbix_partitioning.conf
@@ -40,6 +40,14 @@ logging: syslog
 # premake: Number of partitions to create in advance
 premake: 10
 
+# initial_partitioning_start: Strategy for the first partition during initialization (--init).
+# Options:
+#   db_min:    (Default) Queries SELECT MIN(clock) to ensure ALL data is covered. Slow on huge tables consistently.
+#   retention: Starts partitioning from (Now - Retention Period). 
+#              Creates a 'p_archive' partition for all data older than retention.
+#              Much faster as it skips the MIN(clock) query. (Recommended for large DBs)
+initial_partitioning_start: db_min
+
 # replicate_sql: False - Disable binary logging. Partitioning changes are NOT replicated to slaves (use for independent maintenance).
 # replicate_sql: True - Enable binary logging. Partitioning changes ARE replicated to slaves (use for consistent cluster schema).
 replicate_sql: False
\ No newline at end of file
diff --git a/partitioning/zabbix_partitioning.py b/partitioning/zabbix_partitioning.py
index e4da393..a1a8f76 100644
--- a/partitioning/zabbix_partitioning.py
+++ b/partitioning/zabbix_partitioning.py
@@ -371,7 +371,7 @@ class ZabbixPartitioner:
         for name in to_drop:
             self.execute_query(f"ALTER TABLE `{table}` DROP PARTITION {name}")
 
-    def initialize_partitioning(self, table: str, period: str, premake: int):
+    def initialize_partitioning(self, table: str, period: str, premake: int, retention_str: str):
         """Initial partitioning for a table (convert regular table to partitioned)."""
         self.logger.info(f"Initializing partitioning for {table}")
         
@@ -384,35 +384,63 @@ class ZabbixPartitioner:
              self.logger.info(f"Table {table} is already partitioned.")
              return
 
-        # Check for data
-        min_clock = self.get_table_min_clock(table)
-        
-        if not min_clock:
-            # Empty table. Start from NOW
-            start_dt = self.truncate_date(datetime.now(), period)
+        init_strategy = self.config.get('initial_partitioning_start', 'db_min')
+        start_dt = None
+        p_archive_ts = None
+
+        if init_strategy == 'retention':
+            self.logger.info(f"Strategy 'retention': Calculating start date from retention ({retention_str})")
+            retention_date = self.get_lookback_date(retention_str)
+            # Start granular partitions from the retention date
+            start_dt = self.truncate_date(retention_date, period)
+            # Create a catch-all for anything older
+            p_archive_ts = int(start_dt.timestamp())
         else:
-             # Table has data. 
-             # For a safe migration, we usually create a catch-all for old data (p_old) or just start partitions covering existing data.
-             # This script's strategy: Create partitions starting from min_clock.
-             start_dt = self.truncate_date(min_clock, period)
+            # Default 'db_min' strategy
+            self.logger.info("Strategy 'db_min': Querying table for minimum clock (may be slow)")
+            min_clock = self.get_table_min_clock(table)
+            
+            if not min_clock:
+                # Empty table. Start from NOW
+                start_dt = self.truncate_date(datetime.now(), period)
+            else:
+                 # Table has data. 
+                 start_dt = self.truncate_date(min_clock, period)
         
         # Build list of partitions from start_dt up to NOW + premake
         target_dt = self.get_next_date(self.truncate_date(datetime.now(), period), period, premake)
         
         curr = start_dt
         partitions_def = {}
+        
+        # If we have an archive partition, add it first
+        if p_archive_ts:
+             partitions_def['p_archive'] = str(p_archive_ts)
+
         while curr < target_dt:
             name = self.get_partition_name(curr, period)
             desc = self.get_partition_description(curr, period)
             partitions_def[name] = desc
             curr = self.get_next_date(curr, period, 1)
             
+        # Re-doing the loop to be cleaner on types
         parts_sql = []
-        for name, timestamp_expr in sorted(partitions_def.items()):
-            parts_sql.append(PARTITION_TEMPLATE % (name, timestamp_expr))
+        
+        # 1. Archive Partition
+        if p_archive_ts:
+             parts_sql.append(f"PARTITION p_archive VALUES LESS THAN ({p_archive_ts}) ENGINE = InnoDB")
+        
+        # 2. Granular Partitions
+        # We need to iterate again from start_dt
+        curr = start_dt
+        while curr < target_dt:
+            name = self.get_partition_name(curr, period)
+            desc_date_str = self.get_partition_description(curr, period) # Returns "YYYY-MM-DD HH:MM:SS"
+            parts_sql.append(PARTITION_TEMPLATE % (name, desc_date_str))
+            curr = self.get_next_date(curr, period, 1)
             
         query = f"ALTER TABLE `{table}` PARTITION BY RANGE (`clock`) (\n" + ",\n".join(parts_sql) + "\n)"
-        self.logger.info(f"Applying initial partitioning to {table} ({len(partitions_def)} partitions)")
+        self.logger.info(f"Applying initial partitioning to {table} ({len(parts_sql)} partitions)")
         self.execute_query(query)
 
     def run(self, mode: str):
@@ -437,7 +465,7 @@ class ZabbixPartitioner:
                      retention = item[table]
                      
                      if mode == 'init':
-                         self.initialize_partitioning(table, period, premake)
+                         self.initialize_partitioning(table, period, premake, retention)
                      else:
                          # Maintenance mode (Add new, remove old)
                          self.create_future_partitions(table, period, premake)