redis version:2.0.4 

    Java client for redis: http://code.google.com/p/jredis/

    随机产生一个10个字节的字符串key,value与key相同，写入redis,当数据量达到1900w时，出现了如下的一些状况：刚开始测试程序全部写入db0,后来修改了测试程序，均匀写入其它数据库

    master上面的日志信息：

    [457] 25 Mar 11:35:58 - DB 0: 18308251 keys (0 volatile) in 50331648 slots HT.

    [457] 25 Mar 11:35:58 - DB 1: 61384 keys (0 volatile) in 65536 slots HT.

    [457] 25 Mar 11:35:58 - DB 2: 61203 keys (0 volatile) in 65536 slots HT.

    [457] 25 Mar 11:35:58 - DB 3: 61195 keys (0 volatile) in 65536 slots HT.

    [457] 25 Mar 11:35:58 - DB 4: 61227 keys (0 volatile) in 65536 slots HT.

    [457] 25 Mar 11:35:58 - DB 5: 61208 keys (0 volatile) in 65536 slots HT.

    [457] 25 Mar 11:35:58 - DB 6: 61207 keys (0 volatile) in 65536 slots HT.

    [457] 25 Mar 11:35:58 - DB 7: 61218 keys (0 volatile) in 65536 slots HT.

    [457] 25 Mar 11:35:58 - DB 8: 61189 keys (0 volatile) in 65536 slots HT.

    [457] 25 Mar 11:35:58 - DB 9: 61223 keys (0 volatile) in 65536 slots HT.

    [457] 25 Mar 11:35:58 - DB 10: 61179 keys (0 volatile) in 65536 slots HT.

    [457] 25 Mar 11:35:58 - DB 11: 61361 keys (0 volatile) in 65536 slots HT.

    [457] 25 Mar 11:35:58 - DB 12: 61227 keys (0 volatile) in 65536 slots HT.

    [457] 25 Mar 11:35:58 - DB 13: 61231 keys (0 volatile) in 65536 slots HT.

    [457] 25 Mar 11:35:58 - DB 14: 30792 keys (0 volatile) in 32768 slots HT.

    [457] 25 Mar 11:35:58 - DB 15: 30590 keys (0 volatile) in 32768 slots HT.

    slave开启来后，在slave的redis.log日志记录里看到如下的报错信息：

    [15029] 25 Mar 11:33:20 * Connecting to MASTER...

    [15029] 25 Mar 11:33:20 # Bad protocol from MASTER, the first byte is not \'$\', are you sure the host and port are right?

    [15029] 25 Mar 11:33:21 * Connecting to MASTER...

    [15029] 25 Mar 11:33:21 # Bad protocol from MASTER, the first byte is not \'$\', are you sure the host and port are right?

    [15029] 25 Mar 11:33:22 * Connecting to MASTER...

    [15029] 25 Mar 11:33:22 # Bad protocol from MASTER, the first byte is not \'$\', are you sure the host and port are right?

    [15029] 25 Mar 11:33:23 * Connecting to MASTER...

    [15029] 25 Mar 11:33:23 # Bad protocol from MASTER, the first byte is not \'$\', are you sure the host and port are right?

    这个报错信息不太友好，host与port肯定是对的

    查看slave的info信息：slave与master一直同步失败,同步到547w数据的时候就挂掉了

    redis> info

    redis_version:2.0.4

    redis_git_sha1:00000000

    redis_git_dirty:0

    arch_bits:64

    multiplexing_api:epoll

    process_id:15029

    uptime_in_seconds:933

    uptime_in_days:0

    connected_clients:1

    connected_slaves:0

    blocked_clients:0

    used_memory:1006633016

    used_memory_human:960.00M

    changes_since_last_save:0

    bgsave_in_progress:0

    last_save_time:1301023746

    bgrewriteaof_in_progress:0

    total_connections_received:1

    total_commands_processed:1

    expired_keys:0

    hash_max_zipmap_entries:64

    hash_max_zipmap_value:512

    pubsub_channels:0

    pubsub_patterns:0

    vm_enabled:0

    role:slave

    master_host:#.#.#.#

    master_port:6379

    master_link_status:down

    master_last_io_seconds_ago:-1

    db0:keys=5479113,expires=0

    db1:keys=3,expires=0

    db3:keys=2,expires=0

    db4:keys=1,expires=0

    然后查看redis master上面的日志信息：

    [457] 25 Mar 11:35:59 - Accepted #.#.#.#:38193

    [457] 25 Mar 11:35:59 * Slave ask for synchronization

    [457] 25 Mar 11:35:59 * Starting BGSAVE for SYNC

    [457] 25 Mar 11:35:59 # Can\'t save in background: fork: Cannot allocate memory

    [457] 25 Mar 11:35:59 * Replication failed, can\'t BGSAVE

    [457] 25 Mar 11:35:59 - Client closed connection

    再查看redis master使用内存的情况，使用3.3G内存

    redis> info

    redis_version:2.0.4

    redis_git_sha1:00000000

    redis_git_dirty:0

    arch_bits:64

    multiplexing_api:epoll

    process_id:457

    uptime_in_seconds:261600

    uptime_in_days:3

    connected_clients:2

    connected_slaves:0

    blocked_clients:0

    used_memory:3539375688

    used_memory_human:3.30G

    changes_since_last_save:19470395

    bgsave_in_progress:0

    last_save_time:1300764509

    bgrewriteaof_in_progress:0

    total_connections_received:7130

    total_commands_processed:22435670

    expired_keys:0

    hash_max_zipmap_entries:64

    hash_max_zipmap_value:512

    pubsub_channels:0

    pubsub_patterns:0

    vm_enabled:0

    role:master

    再看一下OS：原来是内存用光了

    $free -m

     total       used       free     shared    buffers     cached

    Mem:          3951       3900         50          0          0          8

    在有内存的情况下，每秒可以达到5w次tps get key[10bytes] value[10bytes]，但19148251万object*(10bytes key length+10bytes value length)/1024/1024=365M,而redis-master用了3.3G的内存，管理结构成本相当大。