PostgreSQL 10.0 preview 功能增强 - 逻辑复制支持并行COPY初始化数据

标签

PostgreSQL , 10.0 , 逻辑复制 , 初始数据COPY

背景

PostgreSQL 已支持逻辑复制，同时对逻辑复制增加了一个初始同步的增强功能，支持通过wal receiver协议跑COPY命令（已封装在逻辑复制的内核代码中），支持多表并行。

也就是说，你可以使用PostgreSQL的逻辑复制，快速的（流式、并行）将一个实例迁移到另一个实例。

Logical replication support for initial data copy  

Add functionality for a new subscription to copy the initial data in the
tables and then sync with the ongoing apply process.  

For the copying, add a new internal COPY option to have the COPY source
data provided by a callback function.  The initial data copy works on
the subscriber by receiving COPY data from the publisher and then
providing it locally into a COPY that writes to the destination table.  

A WAL receiver can now execute full SQL commands.  This is used here to
obtain information about tables and publications.  

Several new options were added to CREATE and ALTER SUBSCRIPTION to
control whether and when initial table syncing happens.  

Change pg_dump option --no-create-subscription-slots to
--no-subscription-connect and use the new CREATE SUBSCRIPTION
... NOCONNECT option for that.  

Author: Petr Jelinek <petr.jelinek@2ndquadrant.com>
Tested-by: Erik Rijkers <er@xs4all.nl>

逻辑复制包含的初始化COPY的流程如下

主库开启事务快照(快照支持在多个会话间共享, 这也是PostgreSQL的独门秘籍之一), COPY数据, COPY结束后释放快照, 从快照对应的WAL LSN开始接收增量.

/*-------------------------------------------------------------------------
   2  * tablesync.c
   3  *    PostgreSQL logical replication
   4  *
   5  * Copyright (c) 2012-2016, PostgreSQL Global Development Group
   6  *
   7  * IDENTIFICATION
   8  *    src/backend/replication/logical/tablesync.c
   9  *
  10  * NOTES
  11  *    This file contains code for initial table data synchronization for
  12  *    logical replication.
  13  *
  14  *    The initial data synchronization is done separately for each table,
  15  *    in separate apply worker that only fetches the initial snapshot data
  16  *    from the publisher and then synchronizes the position in stream with
  17  *    the main apply worker.
  18  *
  19  *    The are several reasons for doing the synchronization this way:
  20  *     - It allows us to parallelize the initial data synchronization
  21  *       which lowers the time needed for it to happen.
  22  *     - The initial synchronization does not have to hold the xid and LSN
  23  *       for the time it takes to copy data of all tables, causing less
  24  *       bloat and lower disk consumption compared to doing the
  25  *       synchronization in single process for whole database.
  26  *     - It allows us to synchronize the tables added after the initial
  27  *       synchronization has finished.
  28  *
  29  *    The stream position synchronization works in multiple steps.
  30  *     - Sync finishes copy and sets table state as SYNCWAIT and waits
  31  *       for state to change in a loop.
  32  *     - Apply periodically checks tables that are synchronizing for SYNCWAIT.
  33  *       When the desired state appears it will compare its position in the
  34  *       stream with the SYNCWAIT position and based on that changes the
  35  *       state to based on following rules:
  36  *        - if the apply is in front of the sync in the wal stream the new
  37  *          state is set to CATCHUP and apply loops until the sync process
  38  *          catches up to the same LSN as apply
  39  *        - if the sync is in front of the apply in the wal stream the new
  40  *          state is set to SYNCDONE
  41  *        - if both apply and sync are at the same position in the wal stream
  42  *          the state of the table is set to READY
  43  *     - If the state was set to CATCHUP sync will read the stream and
  44  *       apply changes until it catches up to the specified stream
  45  *       position and then sets state to READY and signals apply that it
  46  *       can stop waiting and exits, if the state was set to something
  47  *       else than CATCHUP the sync process will simply end.
  48  *     - If the state was set to SYNCDONE by apply, the apply will
  49  *       continue tracking the table until it reaches the SYNCDONE stream
  50  *       position at which point it sets state to READY and stops tracking.
  51  *
  52  *    The catalog pg_subscription_rel is used to keep information about
  53  *    subscribed tables and their state and some transient state during
  54  *    data synchronization is kept in shared memory.
  55  *
  56  *    Example flows look like this:
  57  *     - Apply is in front:
  58  *        sync:8
  59  *          -> set SYNCWAIT
  60  *        apply:10
  61  *          -> set CATCHUP
  62  *          -> enter wait-loop
  63  *        sync:10
  64  *          -> set READY
  65  *          -> exit
  66  *        apply:10
  67  *          -> exit wait-loop
  68  *          -> continue rep
  69  *     - Sync in front:
  70  *        sync:10
  71  *          -> set SYNCWAIT
  72  *        apply:8
  73  *          -> set SYNCDONE
  74  *          -> continue per-table filtering
  75  *        sync:10
  76  *          -> exit
  77  *        apply:10
  78  *          -> set READY
  79  *          -> stop per-table filtering
  80  *          -> continue rep
  81  *-------------------------------------------------------------------------
  82  */
  83

这个patch的讨论，详见邮件组，本文末尾URL。

PostgreSQL社区的作风非常严谨，一个patch可能在邮件组中讨论几个月甚至几年，根据大家的意见反复的修正，patch合并到master已经非常成熟，所以PostgreSQL的稳定性也是远近闻名的。

参考

https://git.postgresql.org/gitweb/?p=postgresql.git;a=commit;h=7c4f52409a8c7d85ed169bbbc1f6092274d03920