Небольшие изменения

.env.example

@@ -21,6 +21,8 @@ ENABLE_TIMESCALE=false
 DRY_RUN=false
 READ_LIMIT=0
 CHUNK_SIZE=5000
+INCREMENTAL_CHUNK_SIZE=5000
+FULL_LOAD_CHUNK_SIZE=5000
 WRITE_CHUNK_SIZE=5000
 CREATE_FOREIGN_KEYS=true
 QUEUE_POLL_SECONDS=1

app/config.py

@@ -100,6 +100,8 @@ class Config:
     # Настройки миграции
     CHUNK_SIZE = int(os.getenv('CHUNK_SIZE', '5000'))
     WRITE_CHUNK_SIZE = int(os.getenv('WRITE_CHUNK_SIZE', str(CHUNK_SIZE)))
+    INCREMENTAL_CHUNK_SIZE = int(os.getenv('INCREMENTAL_CHUNK_SIZE', str(CHUNK_SIZE * 2)))
+    FULL_LOAD_CHUNK_SIZE = int(os.getenv('FULL_LOAD_CHUNK_SIZE', str(CHUNK_SIZE * 4)))
     BATCH_SIZE = 10  # Через сколько чанков выводить прогресс
     REPLICATOR_SCHEMA = os.getenv('REPLICATOR_SCHEMA', 'replicator')
     STATE_TABLE = 'migration_state'

app/logging_utils.py

@@ -37,15 +37,16 @@ class MigrationLogger:
         
     def setup_logging(self):
         """Настройка системы логирования"""
-        logging.basicConfig(
+        fmt = logging.Formatter('%(asctime)s - %(levelname)s - %(message)s')
-            level=self.config.LOG_LEVEL,
+        self.logger = logging.getLogger(f"migration.{self.timestamp}")
-            format='%(asctime)s - %(levelname)s - %(message)s',
+        self.logger.setLevel(self.config.LOG_LEVEL)
-            handlers=[
+        self.logger.propagate = False
-                logging.FileHandler(self.log_file, encoding='utf-8'),
+        file_handler = logging.FileHandler(self.log_file, encoding='utf-8')
-                logging.StreamHandler()  # Вывод в консоль
+        file_handler.setFormatter(fmt)
-            ]
+        stream_handler = logging.StreamHandler()
-        )
+        stream_handler.setFormatter(fmt)
-        self.logger = logging.getLogger(__name__)
+        self.logger.addHandler(file_handler)
+        self.logger.addHandler(stream_handler)
         
     def log_info(self, message: str):
         """Логирование информационного сообщения"""

app/migrator.py

@@ -1,12 +1,15 @@
 import os
 import re
 import time
+import csv
+import io
 from datetime import datetime
 from typing import Any, Dict, List, Optional, Tuple
 
 import pandas as pd
 from sqlalchemy import create_engine, inspect, text
 from sqlalchemy.exc import DBAPIError, OperationalError
+from sqlalchemy.pool import NullPool
 from sqlalchemy.sql import sqltypes
 
 from .config import Config, TableMigrationConfig
@@ -50,12 +53,8 @@ class DatabaseMigrator:
             connect_args={
                 'charset': self.config.MSSQL_CHARSET,
                 'login_timeout': self.config.MSSQL_CONNECT_TIMEOUT,
-                'timeout': self.config.MSSQL_CONNECT_TIMEOUT,
             },
-            pool_pre_ping=True,
+            poolclass=NullPool,
-            pool_recycle=self.config.MSSQL_POOL_RECYCLE,
-            pool_size=self.config.MSSQL_POOL_SIZE,
-            max_overflow=self.config.MSSQL_MAX_OVERFLOW,
         )
 
     def reconnect_mssql_engine(self):
@@ -72,10 +71,10 @@ class DatabaseMigrator:
 
     def is_retryable_mssql_error(self, exception: Exception) -> bool:
         """Определение временной ошибки MSSQL/pymssql, при которой есть смысл повторить таблицу."""
-        if isinstance(exception, (OperationalError, DBAPIError)):
         message = str(exception).lower()
         retry_markers = (
             'dbprocess is dead',
+            'unexpected eof',
             'adaptive server connection failed',
             'server closed the connection unexpectedly',
             'connection reset',
@@ -85,14 +84,32 @@ class DatabaseMigrator:
             'closed connection',
             'not enabled',
             '08s01',
+            '20002',
+            '20017',
         )
-            return any(marker in message for marker in retry_markers)
+
-        return False
+        if any(marker in message for marker in retry_markers):
+            return True
+
+        return isinstance(exception, (OperationalError, DBAPIError))
 
     def migrate_table_once(self, table_config: TableMigrationConfig, force_full: bool = False) -> bool:
         """Один проход миграции таблицы без retry-обертки."""
         if force_full:
             self.logger.log_info(f"Force full reload для таблицы {table_config.source_table}")
+            is_initial_force_full = (
+                table_config.mode == 'incremental'
+                and table_config.initial_load_mode == 'full_then_incremental'
+                and not self.table_exists(table_config.pg_table)
+                and self.get_last_watermark(table_config.pg_table).get('last_x_datetime') is None
+            )
+            if is_initial_force_full:
+                self.logger.log_info(
+                    f"Первый force_full для {table_config.source_table}: "
+                    "загрузка выполняется без SQLAlchemy"
+                )
+                success = self.migrate_full_table_without_sqlalchemy(table_config)
+            else:
                 success = self.migrate_full_table(table_config)
             if success and table_config.mode == 'incremental' and table_config.life_table:
                 upper_bound = self.get_incremental_upper_bound(table_config)
@@ -453,7 +470,7 @@ class DatabaseMigrator:
             sql,
             self.src_engine,
             params=params,
-            chunksize=self.config.CHUNK_SIZE,
+            chunksize=self.config.INCREMENTAL_CHUNK_SIZE,
         )
 
     def read_full_chunks(
@@ -464,9 +481,9 @@ class DatabaseMigrator:
         """Чтение полной таблицы чанками с опциональным лимитом для проверки."""
         if read_limit:
             sql = text(f"SELECT TOP {int(read_limit)} * FROM {self.quote_mssql_identifier(table_name)}")
-            return pd.read_sql_query(sql, self.src_engine, chunksize=self.config.CHUNK_SIZE)
+            return pd.read_sql_query(sql, self.src_engine, chunksize=self.config.FULL_LOAD_CHUNK_SIZE)
 
-        return pd.read_sql_table(table_name, self.src_engine, chunksize=self.config.CHUNK_SIZE)
+        return pd.read_sql_table(table_name, self.src_engine, chunksize=self.config.FULL_LOAD_CHUNK_SIZE)
 
     def write_dataframe_batch(
         self,
@@ -488,6 +505,80 @@ class DatabaseMigrator:
             method='multi',
         )
 
+    def read_full_chunks_without_sqlalchemy(
+        self,
+        table_name: str,
+        read_limit: Optional[int] = None,
+    ):
+        """Чтение полной таблицы чанками через DBAPI-курсор (без pandas.read_sql_*)."""
+        src_connection = self.src_engine.raw_connection()
+        cursor = src_connection.cursor()
+
+        top_clause = f"TOP {int(read_limit)} " if read_limit else ""
+        sql = f"SELECT {top_clause}* FROM {self.quote_mssql_identifier(table_name)}"
+
+        try:
+            cursor.execute(sql)
+            columns = [column[0] for column in cursor.description]
+            while True:
+                rows = cursor.fetchmany(self.config.FULL_LOAD_CHUNK_SIZE)
+                if not rows:
+                    break
+                yield pd.DataFrame.from_records(rows, columns=columns)
+        finally:
+            try:
+                cursor.close()
+            finally:
+                src_connection.close()
+
+    def write_dataframe_batch_without_sqlalchemy(
+        self,
+        chunk: pd.DataFrame,
+        table_name: str,
+    ):
+        """Batch-запись DataFrame в PostgreSQL через COPY (без pandas.to_sql)."""
+        if chunk.empty:
+            return
+        if self.config.DRY_RUN:
+            self.logger.log_info(f"DRY RUN: пропущена запись {len(chunk)} строк в {table_name}")
+            return
+
+        buffer = io.StringIO()
+        chunk.to_csv(
+            buffer,
+            index=False,
+            header=False,
+            sep='\t',
+            na_rep='\\N',
+            date_format='%Y-%m-%d %H:%M:%S.%f',
+            quoting=csv.QUOTE_MINIMAL,
+            escapechar='\\',
+        )
+        buffer.seek(0)
+
+        quoted_columns = ', '.join(
+            self.quote_identifier(column_name)
+            for column_name in chunk.columns
+        )
+        copy_sql = (
+            f"COPY {self.qualify_table_name(table_name)} ({quoted_columns}) "
+            "FROM STDIN WITH (FORMAT csv, DELIMITER E'\\t', NULL '\\N', ESCAPE '\\', QUOTE '\"')"
+        )
+
+        dst_connection = self.dst_engine.raw_connection()
+        cursor = dst_connection.cursor()
+        try:
+            cursor.copy_expert(copy_sql, buffer)
+            dst_connection.commit()
+        except Exception:
+            dst_connection.rollback()
+            raise
+        finally:
+            try:
+                cursor.close()
+            finally:
+                dst_connection.close()
+
     def prepare_incremental_chunk(
         self,
         chunk: pd.DataFrame,
@@ -571,6 +662,7 @@ class DatabaseMigrator:
         chunk: pd.DataFrame,
         table_name: str,
         primary_key: List[str],
+        staging_table: Optional[str] = None,
     ):
         """Batch delete в PostgreSQL через staging-таблицу с ключами."""
         if chunk.empty:
@@ -588,7 +680,8 @@ class DatabaseMigrator:
         if missing_columns:
             raise ValueError(f"Для удаления из {table_name} не найдены ключевые поля: {missing_columns}")
 
-        staging_table = f"_stg_delete_{table_name}_{int(time.time() * 1000)}"
+        own_staging = staging_table is None
+        staging = staging_table or f"_stg_delete_{table_name}_{int(time.time() * 1000)}"
         key_chunk = chunk[primary_key].drop_duplicates()
         join_condition = ' AND '.join([
             f"target.{self.quote_identifier(column)} = source.{self.quote_identifier(column)}"
@@ -596,19 +689,30 @@ class DatabaseMigrator:
         ])
 
         try:
-            self.write_dataframe_batch(key_chunk, staging_table, if_exists='replace')
+            if self.table_exists(staging):
+                with self.dst_engine.connect() as conn:
+                    conn.execute(text(f'TRUNCATE TABLE {self.quote_identifier(staging)}'))
+                    conn.commit()
+            else:
+                key_chunk.iloc[0:0].to_sql(staging, self.dst_engine, if_exists='replace', index=False)
+                with self.dst_engine.connect() as conn:
+                    conn.execute(text(f'ALTER TABLE {self.quote_identifier(staging)} SET UNLOGGED'))
+                    conn.commit()
+            self.write_dataframe_batch_without_sqlalchemy(key_chunk, staging)
             sql = f"""
                 DELETE FROM {self.quote_identifier(table_name)} AS target
-                USING {self.quote_identifier(staging_table)} AS source
+                USING {self.quote_identifier(staging)} AS source
                 WHERE {join_condition}
             """
             with self.dst_engine.connect() as conn:
                 conn.execute(text(sql))
-                conn.execute(text(f'DROP TABLE IF EXISTS {self.quote_identifier(staging_table)}'))
+                if own_staging:
+                    conn.execute(text(f'DROP TABLE IF EXISTS {self.quote_identifier(staging)}'))
                 conn.commit()
         except Exception:
+            if own_staging:
                 with self.dst_engine.connect() as conn:
-                conn.execute(text(f'DROP TABLE IF EXISTS {self.quote_identifier(staging_table)}'))
+                    conn.execute(text(f'DROP TABLE IF EXISTS {self.quote_identifier(staging)}'))
                     conn.commit()
             raise
 
@@ -617,6 +721,7 @@ class DatabaseMigrator:
         chunk: pd.DataFrame,
         table_name: str,
         primary_key: List[str],
+        staging_table: Optional[str] = None,
     ):
         """Batch upsert через staging-таблицу."""
         if self.config.DRY_RUN:
@@ -627,8 +732,8 @@ class DatabaseMigrator:
             self.write_dataframe_batch(chunk, table_name, if_exists='append')
             return
 
-        chunk = self.deduplicate_incremental_chunk(chunk, primary_key)
+        own_staging = staging_table is None
-        staging_table = f"_stg_{table_name}_{int(time.time() * 1000)}"
+        staging = staging_table or f"_stg_{table_name}_{int(time.time() * 1000)}"
         columns = list(chunk.columns)
         quoted_columns = ', '.join([self.quote_identifier(column) for column in columns])
         conflict_columns = ', '.join([self.quote_identifier(column) for column in primary_key])
@@ -644,20 +749,31 @@ class DatabaseMigrator:
             conflict_action = "DO NOTHING"
 
         try:
-            self.write_dataframe_batch(chunk, staging_table, if_exists='replace')
+            if self.table_exists(staging):
+                with self.dst_engine.connect() as conn:
+                    conn.execute(text(f'TRUNCATE TABLE {self.quote_identifier(staging)}'))
+                    conn.commit()
+            else:
+                chunk.iloc[0:0].to_sql(staging, self.dst_engine, if_exists='replace', index=False)
+                with self.dst_engine.connect() as conn:
+                    conn.execute(text(f'ALTER TABLE {self.quote_identifier(staging)} SET UNLOGGED'))
+                    conn.commit()
+            self.write_dataframe_batch_without_sqlalchemy(chunk, staging)
             sql = f"""
                 INSERT INTO {self.quote_identifier(table_name)} ({quoted_columns})
                 SELECT {quoted_columns}
-                FROM {self.quote_identifier(staging_table)}
+                FROM {self.quote_identifier(staging)}
                 ON CONFLICT ({conflict_columns}) {conflict_action}
             """
             with self.dst_engine.connect() as conn:
                 conn.execute(text(sql))
-                conn.execute(text(f'DROP TABLE IF EXISTS {self.quote_identifier(staging_table)}'))
+                if own_staging:
+                    conn.execute(text(f'DROP TABLE IF EXISTS {self.quote_identifier(staging)}'))
                 conn.commit()
         except Exception:
+            if own_staging:
                 with self.dst_engine.connect() as conn:
-                conn.execute(text(f'DROP TABLE IF EXISTS {self.quote_identifier(staging_table)}'))
+                    conn.execute(text(f'DROP TABLE IF EXISTS {self.quote_identifier(staging)}'))
                     conn.commit()
             raise
 
@@ -1040,10 +1156,6 @@ class DatabaseMigrator:
             pk_columns = self.get_mssql_primary_key(table_name)
             foreign_keys = self.get_mssql_foreign_keys(table_name)
 
-            # Очищаем целевую таблицу
-            self.logger.log_info(f"Очистка целевой таблицы {pg_table}")
-            self.truncate_table(pg_table)
-            
             # Читаем данные
             self.logger.log_info(f"Чтение данных из {table_name}")
             chunks = self.read_full_chunks(table_name, read_limit=read_limit)
@@ -1054,11 +1166,84 @@ class DatabaseMigrator:
 
             for chunk_num, chunk in enumerate(chunks, 1):
                 if first_chunk:
-                    self.write_dataframe_batch(chunk, pg_table, if_exists='fail')
+                    # Дропаем целевую таблицу только после успешного чтения первого чанка,
-                    first_chunk = False
+                    # чтобы не уничтожить данные при недоступном MSSQL
+                    self.logger.log_info(f"Очистка целевой таблицы {pg_table}")
+                    self.truncate_table(pg_table)
+                    chunk.iloc[0:0].to_sql(pg_table, self.dst_engine, if_exists='replace', index=False)
                     self.logger.log_info(f"Таблица {pg_table} создана")
-                else:
+                    first_chunk = False
-                    self.write_dataframe_batch(chunk, pg_table, if_exists='append')
+                self.write_dataframe_batch_without_sqlalchemy(chunk, pg_table)
+
+                total_rows += len(chunk)
+                if chunk_num % self.config.BATCH_SIZE == 0:
+                    self.logger.log_progress(table_name, chunk_num, total_rows)
+            
+            self.logger.log_info(f"Всего загружено строк: {total_rows}")
+
+            if total_rows > 0:
+                self.sync_target_schema(table_name, pg_table)
+                self.create_timescale_hypertable(table_config)
+
+                if self.can_create_primary_key(table_config, pk_columns):
+                    self.logger.log_info(f"Создание первичного ключа для {pg_table}")
+                    self.create_pg_primary_key(pg_table, pk_columns)
+
+                if indexes:
+                    self.logger.log_info(f"Создание {len(indexes)} индексов для {pg_table}")
+                    self.create_pg_indexes(pg_table, indexes)
+
+                if foreign_keys:
+                    self.logger.log_info(f"Создание {len(foreign_keys)} внешних ключей для {pg_table}")
+                    self.create_pg_foreign_keys(pg_table, foreign_keys)
+
+                self.logger.log_info(f"Обновление статистики для {pg_table}")
+                if total_rows > 1000000:
+                    self.vacuum_analyze_table(pg_table)
+                else:
+                    self.analyze_table(pg_table)
+            
+            self.logger.log_table_success(table_name, total_rows)
+            return True
+            
+        except Exception as e:
+            if self.is_retryable_mssql_error(e):
+                raise
+            self.logger.log_table_failure(table_name, str(e))
+            return False
+
+    def migrate_full_table_without_sqlalchemy(
+        self,
+        table_config: TableMigrationConfig,
+        read_limit: Optional[int] = None,
+    ) -> bool:
+        """Полная миграция одной таблицы без SQLAlchemy в этапе чтения/заливки данных."""
+        table_name = table_config.source_table
+        pg_table = table_config.pg_table
+        self.logger.log_table_start(table_name)
+
+        try:
+            indexes = self.get_mssql_indexes(table_name)
+            pk_columns = self.get_mssql_primary_key(table_name)
+            foreign_keys = self.get_mssql_foreign_keys(table_name)
+
+            self.logger.log_info(f"Чтение данных из {table_name} без SQLAlchemy")
+            chunks = self.read_full_chunks_without_sqlalchemy(table_name, read_limit=read_limit)
+
+            first_chunk = True
+            total_rows = 0
+
+            for chunk_num, chunk in enumerate(chunks, 1):
+                if first_chunk:
+                    # Дропаем целевую таблицу только после успешного чтения первого чанка,
+                    # чтобы не уничтожить данные при недоступном MSSQL
+                    self.logger.log_info(f"Очистка целевой таблицы {pg_table}")
+                    self.truncate_table(pg_table)
+                    self.write_dataframe_batch(chunk.iloc[0:0], pg_table, if_exists='fail')
+                    self.logger.log_info(f"Таблица {pg_table} создана")
+                    first_chunk = False
+
+                self.write_dataframe_batch_without_sqlalchemy(chunk, pg_table)
 
                 total_rows += len(chunk)
                 if chunk_num % self.config.BATCH_SIZE == 0:
@@ -1068,26 +1253,20 @@ class DatabaseMigrator:
 
             if total_rows > 0:
                 self.sync_target_schema(table_name, pg_table)
-
-            if total_rows > 0:
                 self.create_timescale_hypertable(table_config)
 
-            # Создаем первичный ключ
                 if self.can_create_primary_key(table_config, pk_columns):
                     self.logger.log_info(f"Создание первичного ключа для {pg_table}")
                     self.create_pg_primary_key(pg_table, pk_columns)
 
-            # Создаем индексы
                 if indexes:
                     self.logger.log_info(f"Создание {len(indexes)} индексов для {pg_table}")
                     self.create_pg_indexes(pg_table, indexes)
 
-            # Создаем внешние ключи
                 if foreign_keys:
                     self.logger.log_info(f"Создание {len(foreign_keys)} внешних ключей для {pg_table}")
                     self.create_pg_foreign_keys(pg_table, foreign_keys)
 
-            # Обновляем статистику
                 self.logger.log_info(f"Обновление статистики для {pg_table}")
                 if total_rows > 1000000:
                     self.vacuum_analyze_table(pg_table)
@@ -1108,6 +1287,8 @@ class DatabaseMigrator:
         table_name = table_config.source_table
         pg_table = table_config.pg_table
         self.logger.log_table_start(f"{table_name} ({table_config.read_table})")
+        upsert_staging = f"_stg_upsert_{pg_table}"
+        delete_staging = f"_stg_delete_{pg_table}"
 
         try:
             if not table_config.life_table:
@@ -1188,18 +1369,19 @@ class DatabaseMigrator:
                     )
 
                 if not delete_chunk.empty:
-                    self.delete_dataframe_batch(delete_chunk, pg_table, table_config.primary_key)
+                    self.delete_dataframe_batch(delete_chunk, pg_table, table_config.primary_key, staging_table=delete_staging)
 
                 if write_chunk.empty:
                     pass
                 elif first_chunk:
-                    self.write_dataframe_batch(write_chunk, pg_table, if_exists='append')
+                    write_chunk.iloc[0:0].to_sql(pg_table, self.dst_engine, if_exists='replace', index=False)
+                    self.write_dataframe_batch_without_sqlalchemy(write_chunk, pg_table)
                     self.create_timescale_hypertable(table_config)
                     if self.can_create_primary_key(table_config, table_config.primary_key):
                         self.create_pg_primary_key(pg_table, table_config.primary_key)
                     first_chunk = False
                 elif table_config.primary_key:
-                    self.upsert_dataframe_batch(write_chunk, pg_table, table_config.primary_key)
+                    self.upsert_dataframe_batch(write_chunk, pg_table, table_config.primary_key, staging_table=upsert_staging)
                 else:
                     self.write_dataframe_batch(write_chunk, pg_table, if_exists='append')
 
@@ -1260,6 +1442,15 @@ class DatabaseMigrator:
             self.logger.log_table_failure(table_name, str(e))
             return False
 
+        finally:
+            try:
+                with self.dst_engine.connect() as conn:
+                    conn.execute(text(f'DROP TABLE IF EXISTS {self.quote_identifier(upsert_staging)}'))
+                    conn.execute(text(f'DROP TABLE IF EXISTS {self.quote_identifier(delete_staging)}'))
+                    conn.commit()
+            except Exception as cleanup_error:
+                self.logger.log_warning(f"Не удалось очистить staging-таблицы для {pg_table}: {cleanup_error}")
+
     def run_migration(
         self,
         table_names: Optional[List[str]] = None,