2 Chapter 26. High Availability, Load Balancing, and Replication
6 26.1. Comparison of Different Solutions
7 26.2. Log-Shipping Standby Servers
10 26.2.2. Standby Server Operation
11 26.2.3. Preparing the Primary for Standby Servers
12 26.2.4. Setting Up a Standby Server
13 26.2.5. Streaming Replication
14 26.2.6. Replication Slots
15 26.2.7. Cascading Replication
16 26.2.8. Synchronous Replication
17 26.2.9. Continuous Archiving in Standby
22 26.4.1. User's Overview
23 26.4.2. Handling Query Conflicts
24 26.4.3. Administrator's Overview
25 26.4.4. Hot Standby Parameter Reference
28 Database servers can work together to allow a second server to take
29 over quickly if the primary server fails (high availability), or to
30 allow several computers to serve the same data (load balancing).
31 Ideally, database servers could work together seamlessly. Web servers
32 serving static web pages can be combined quite easily by merely
33 load-balancing web requests to multiple machines. In fact, read-only
34 database servers can be combined relatively easily too. Unfortunately,
35 most database servers have a read/write mix of requests, and read/write
36 servers are much harder to combine. This is because though read-only
37 data needs to be placed on each server only once, a write to any server
38 has to be propagated to all servers so that future read requests to
39 those servers return consistent results.
41 This synchronization problem is the fundamental difficulty for servers
42 working together. Because there is no single solution that eliminates
43 the impact of the sync problem for all use cases, there are multiple
44 solutions. Each solution addresses this problem in a different way, and
45 minimizes its impact for a specific workload.
47 Some solutions deal with synchronization by allowing only one server to
48 modify the data. Servers that can modify data are called read/write,
49 master or primary servers. Servers that track changes in the primary
50 are called standby or secondary servers. A standby server that cannot
51 be connected to until it is promoted to a primary server is called a
52 warm standby server, and one that can accept connections and serves
53 read-only queries is called a hot standby server.
55 Some solutions are synchronous, meaning that a data-modifying
56 transaction is not considered committed until all servers have
57 committed the transaction. This guarantees that a failover will not
58 lose any data and that all load-balanced servers will return consistent
59 results no matter which server is queried. In contrast, asynchronous
60 solutions allow some delay between the time of a commit and its
61 propagation to the other servers, opening the possibility that some
62 transactions might be lost in the switch to a backup server, and that
63 load balanced servers might return slightly stale results. Asynchronous
64 communication is used when synchronous would be too slow.
66 Solutions can also be categorized by their granularity. Some solutions
67 can deal only with an entire database server, while others allow
68 control at the per-table or per-database level.
70 Performance must be considered in any choice. There is usually a
71 trade-off between functionality and performance. For example, a fully
72 synchronous solution over a slow network might cut performance by more
73 than half, while an asynchronous one might have a minimal performance
76 The remainder of this section outlines various failover, replication,
77 and load balancing solutions.