redis4支持內存碎片清理功能使用

最近看到redis4支持內存碎片清理了, 之前一直期待有這麼一個功能, 因爲之前遇到內存碎片的解決辦法就是重啓, 現在終於有了優雅的解決方案.\^o^/, 這個功能其實oranagra 在2017年1月1日已經提交pr了, 相關地址: https://github.com/antirez/redis/pull/3720

版本說明:

•  Redis 4.0-RC3 以上版本才支持的
• 需要使用jemalloc作爲內存分配器(默認的)

功能介紹:

• 支持在運行期進行自動內存碎片清理 (config set activedefrag yes)
• 支持通過命令 memory purge 進行清理(與自動清理區域不同)

功能驗證流程:

(1) 首先需要拉取4.0-RC3之後的版本代碼, 編譯

(2) 啓動時限定內存大小爲1g並啓動lru, 命令如下:

./src/redis-server --maxmemory 1gb --maxmemory-policy allkeys-lru --activedefrag no --port 6383

(3) 構造大量數據並導致lru, 這樣可以觸發內存碎片, 命令如下:

redis-cli -p 6383 debug populate 7000000 asdf 150

(4) 查看當前的內存使用情況, 會發現有200多萬的數據被清理掉了

1. $ redis-cli -p 6383 info keyspace

2. # Keyspace

3. db0:keys=4649543,expires=0,avg_ttl=0

(5) 查看當前的內存碎片率, 這時碎片率(mem_fragmentation_ratio)很高 : 1.54, 意味着54%的內存浪費

1. $ redis-cli -p 6383 info memory

2. # Memory

3. used_memory:1073741736

4. used_memory_human:1024.00M

5. used_memory_rss:1650737152

6. used_memory_rss_human:1.54G

7. used_memory_peak:1608721680

8. used_memory_peak_human:1.50G

9. used_memory_peak_perc:66.75%

10. used_memory_overhead:253906398

11. used_memory_startup:766152

12. used_memory_dataset:819835338

13. used_memory_dataset_perc:76.41%

14. total_system_memory:67535904768

15. total_system_memory_human:62.90G

16. used_memory_lua:37888

17. used_memory_lua_human:37.00K

18. maxmemory:1073741824

19. maxmemory_human:1.00G

20. maxmemory_policy:allkeys-lru

21. mem_fragmentation_ratio:1.54

22. mem_allocator:jemalloc-4.0.3

23. active_defrag_running:0

24. lazyfree_pending_objects:0

(6) 看看內存分配的詳細情況, 這個地方看不懂可以看看: 科普文, 關鍵的是util指標, 指的是內存利用率, 最大的bins內存util是0.661, 說明內存利用率不高

1. $ echo "`redis-cli -p 6383 memory malloc-stats`"

2. ___ Begin jemalloc statistics ___

3. Version: 4.0.3-0-ge9192eacf8935e29fc62fddc2701f7942b1cc02c

4. Assertions disabled

5. Run-time option settings:

6. opt.abort: false

7. opt.lg_chunk: 21

8. opt.dss: "secondary"

9. opt.narenas: 48

10. opt.lg_dirty_mult: 3 (arenas.lg_dirty_mult: 3)

11. opt.stats_print: false

12. opt.junk: "false"

13. opt.quarantine: 0

14. opt.redzone: false

15. opt.zero: false

16. opt.tcache: true

17. opt.lg_tcache_max: 15

18. CPUs: 12

19. Arenas: 48

20. Pointer size: 8

21. Quantum size: 8

22. Page size: 4096

23. Min active:dirty page ratio per arena: 8:1

24. Maximum thread-cached size class: 32768

25. Chunk size: 2097152 (2^21)

26. Allocated: 1074509704, active: 1609732096, metadata: 41779072, resident: 1651118080, mapped: 1652555776

27. Current active ceiling: 1610612736

28.  

29. arenas[0]:

30. assigned threads: 1

31. dss allocation precedence: secondary

32. min active:dirty page ratio: 8:1

33. dirty pages: 393001:24 active:dirty, 0 sweeps, 0 madvises, 0 purged

34. allocated nmalloc ndalloc nrequests

35. small: 1006565256 28412640 9802493 35714594

36. large: 835584 20 11 20

37. huge: 67108864 1 0 1

38. total: 1074509704 28412661 9802504 35714615

39. active: 1609732096

40. mapped: 1650458624

41. metadata: mapped: 40202240, allocated: 491904

42. bins: size ind allocated nmalloc ndalloc nrequests curregs curruns regs pgs util nfills nflushes newruns reruns

43. 8 0 1992 319 70 357 249 1 512 1 0.486 7 8 1 0

44. 16 1 148618896 14110300 4821619 14310175 9288681 55119 256 1 0.658 141103 48825 55119 5

45. 24 2 112104360 7200400 2529385 7300348 4671015 14064 512 3 0.648 72004 25881 14064 1

46. 32 3 288 112 103 7003270 9 1 128 1 0.070 3 7 1 0

47. 40 4 360 109 100 171 9 1 512 5 0.017 3 7 1 0

48. 48 5 1248 112 86 63 26 1 256 3 0.101 2 5 1 0

49. 56 6 896 106 90 16 16 1 512 7 0.031 2 6 1 0

50. 64 7 128 64 62 5 2 1 64 1 0.031 1 3 1 0

51. 80 8 880 106 95 7 11 1 256 5 0.042 2 4 1 0

52. 96 9 9120 212 117 97 95 1 128 3 0.742 4 6 2 1

53. 112 10 336 109 106 2 3 1 256 7 0.011 3 6 3 0

54. 128 11 640 40 35 4 5 1 32 1 0.156 3 4 2 0

55. 160 12 740617440 7000148 2371289 7000001 4628859 54688 128 5 0.661 70271 24334 54689 4

56. 192 13 768 68 64 1 4 1 64 3 0.062 2 4 2 0

57. 224 14 4683616 100000 79091 99946 20909 781 128 7 0.209 1000 1641 782 0

58. 256 15 0 16 16 4 0 0 16 1 1 1 3 1 0

59. 320 16 5120 64 48 16 16 1 64 5 0.250 1 3 1 0

60. 384 17 768 33 31 2 2 1 32 3 0.062 1 3 1 0

61. 448 18 28672 64 0 0 64 1 64 7 1 1 0 1 0

62. 512 19 1024 10 8 4 2 1 8 1 0.250 1 2 2 0

63. 640 20 0 32 32 1 0 0 32 5 1 1 3 1 0

64. ---

65. 896 22 48384 85 31 50 54 2 32 7 0.843 2 3 2 0

66. 1024 23 3072 10 7 3 3 1 4 1 0.750 1 2 3 0

67. 1280 24 20480 16 0 0 16 1 16 5 1 1 0 1 0

68. 1536 25 15360 10 0 0 10 2 8 3 0.625 1 0 2 0

69. 1792 26 28672 16 0 0 16 1 16 7 1 1 0 1 0

70. 2048 27 4096 10 8 2 2 1 2 1 1 1 2 5 0

71. ---

72. 3584 30 35840 10 0 0 10 2 8 7 0.625 1 0 2 0

73. ---

74. 5120 32 250880 49 0 49 49 13 4 5 0.942 0 0 13 0

75. ---

76. 8192 35 81920 10 0 0 10 10 1 2 1 1 0 10 0

77. ---

78. large: size ind allocated nmalloc ndalloc nrequests curruns

79. 16384 39 16384 2 1 2 1

80. 20480 40 40960 2 0 2 2

81. ---

82. 32768 43 32768 1 0 1 1

83. 40960 44 40960 10 9 10 1

84. ---

85. 81920 48 81920 1 0 1 1

86. ---

87. 131072 51 131072 1 0 1 1

88. 163840 52 163840 1 0 1 1

89. ---

90. 327680 56 327680 1 0 1 1

91. ---

92. 1048576 63 0 1 1 1 0

93. ---

94. huge: size ind allocated nmalloc ndalloc nrequests curhchunks

95. ---

96. 67108864 87 67108864 1 0 1 1

97. ---

98. --- End jemalloc statistics ---

(7) 開啓自動內存碎片整理

1. $ redis-cli -p 6383 config set activedefrag yes

2. OK

(8) 等會兒再看看, 發現內存碎片降低了

1. $ redis-cli -p 6383 info memory

2. # Memory

3. used_memory:1073740712

4. used_memory_human:1024.00M

5. used_memory_rss:1253371904

6. used_memory_rss_human:1.17G

7. used_memory_peak:1608721680

8. used_memory_peak_human:1.50G

9. used_memory_peak_perc:66.74%

10. used_memory_overhead:253906398

11. used_memory_startup:766152

12. used_memory_dataset:819834314

13. used_memory_dataset_perc:76.41%

14. total_system_memory:67535904768

15. total_system_memory_human:62.90G

16. used_memory_lua:37888

17. used_memory_lua_human:37.00K

18. maxmemory:1073741824

19. maxmemory_human:1.00G

20. maxmemory_policy:allkeys-lru

21. mem_fragmentation_ratio:1.17

22. mem_allocator:jemalloc-4.0.3

23. active_defrag_running:0

24. lazyfree_pending_objects:0

(9) 可以再看看內存利用率, 可以看到已經上升到0.82

1. $ echo "`redis-cli -p 6383 memory malloc-stats`"

2. ___ Begin jemalloc statistics ___

3. Version: 4.0.3-0-ge9192eacf8935e29fc62fddc2701f7942b1cc02c

4. Assertions disabled

5. Run-time option settings:

6. opt.abort: false

7. opt.lg_chunk: 21

8. opt.dss: "secondary"

9. opt.narenas: 48

10. opt.lg_dirty_mult: 3 (arenas.lg_dirty_mult: 3)

11. opt.stats_print: false

12. opt.junk: "false"

13. opt.quarantine: 0

14. opt.redzone: false

15. opt.zero: false

16. opt.tcache: true

17. opt.lg_tcache_max: 15

18. CPUs: 12

19. Arenas: 48

20. Pointer size: 8

21. Quantum size: 8

22. Page size: 4096

23. Min active:dirty page ratio per arena: 8:1

24. Maximum thread-cached size class: 32768

25. Chunk size: 2097152 (2^21)

26. Allocated: 1074509800, active: 1307602944, metadata: 41779072, resident: 1512247296, mapped: 1652555776

27. Current active ceiling: 1308622848

28.  

29. arenas[0]:

30. assigned threads: 1

31. dss allocation precedence: secondary

32. min active:dirty page ratio: 8:1

33. dirty pages: 319239:39882 active:dirty, 4878 sweeps, 6343 madvises, 33915 purged

34. allocated nmalloc ndalloc nrequests

35. small: 1006565352 35456589 16846439 45126633

36. large: 835584 24 15 24

37. huge: 67108864 1 0 1

38. total: 1074509800 35456614 16846454 45126658

39. active: 1307602944

40. mapped: 1650458624

41. metadata: mapped: 40202240, allocated: 491904

42. bins: size ind allocated nmalloc ndalloc nrequests curregs curruns regs pgs util nfills nflushes newruns reruns

43. 8 0 1992 319 70 357 249 1 512 1 0.486 7 8 1 0

44. 16 1 148618896 17658482 8369801 17858357 9288681 44332 256 1 0.818 141103 48825 55119 26364

45. 24 2 112104360 8897298 4226283 8997246 4671015 11525 512 3 0.791 72004 25881 14064 6205

46. 32 3 384 115 103 9371363 12 1 128 1 0.093 4 7 1 0

47. 40 4 360 109 100 171 9 1 512 5 0.017 3 7 1 0

48. 48 5 1248 112 86 63 26 1 256 3 0.101 2 5 1 0

49. 56 6 896 106 90 16 16 1 512 7 0.031 2 6 1 0

50. 64 7 128 64 62 5 2 1 64 1 0.031 1 3 1 0

51. 80 8 880 106 95 7 11 1 256 5 0.042 2 4 1 0

52. 96 9 9120 212 117 97 95 1 128 3 0.742 4 6 2 1

53. 112 10 336 109 106 2 3 1 256 7 0.011 3 6 3 0

54. 128 11 640 40 35 4 5 1 32 1 0.156 3 4 2 0

55. 160 12 740617440 8788058 4159199 8787911 4628859 44056 128 5 0.820 70271 24334 54689 26488

56. 192 13 768 68 64 1 4 1 64 3 0.062 2 4 2 0

57. 224 14 4683616 110956 90047 110902 20909 467 128 7 0.349 1000 1641 782 105

58. 256 15 0 16 16 4 0 0 16 1 1 1 3 1 0

59. 320 16 5120 64 48 16 16 1 64 5 0.250 1 3 1 0

60. 384 17 768 33 31 2 2 1 32 3 0.062 1 3 1 0

61. 448 18 28672 64 0 0 64 1 64 7 1 1 0 1 0

62. 512 19 1024 10 8 4 2 1 8 1 0.250 1 2 2 0

63. 640 20 0 32 32 1 0 0 32 5 1 1 3 1 0

64. ---

65. 896 22 48384 85 31 50 54 2 32 7 0.843 2 3 2 0

66. 1024 23 3072 10 7 3 3 1 4 1 0.750 1 2 3 0

67. 1280 24 20480 16 0 0 16 1 16 5 1 1 0 1 0

68. 1536 25 15360 10 0 0 10 2 8 3 0.625 1 0 2 0

69. 1792 26 28672 16 0 0 16 1 16 7 1 1 0 1 0

70. 2048 27 4096 10 8 2 2 1 2 1 1 1 2 5 0

71. ---

72. 3584 30 35840 10 0 0 10 2 8 7 0.625 1 0 2 0

73. ---

74. 5120 32 250880 49 0 49 49 13 4 5 0.942 0 0 13 0

75. ---

76. 8192 35 81920 10 0 0 10 10 1 2 1 1 0 10 0

77. ---

78. large: size ind allocated nmalloc ndalloc nrequests curruns

79. 16384 39 16384 2 1 2 1

80. 20480 40 40960 2 0 2 2

81. ---

82. 32768 43 32768 1 0 1 1

83. 40960 44 40960 14 13 14 1

84. ---

85. 81920 48 81920 1 0 1 1

86. ---

87. 131072 51 131072 1 0 1 1

88. 163840 52 163840 1 0 1 1

89. ---

90. 327680 56 327680 1 0 1 1

91. ---

92. 1048576 63 0 1 1 1 0

93. ---

94. huge: size ind allocated nmalloc ndalloc nrequests curhchunks

95. ---

96. 67108864 87 67108864 1 0 1 1

97. ---

98. --- End jemalloc statistics ---

(10) 別急, 還有一個大招: 手動清理

$ redis-cli -p 6383 memory purge

(11) 再次查看內存使用情況: 發現碎片率降到1.04, 內存利用率到0.998, 內存碎片基本上消滅了^_^

1. $ redis-cli -p 6383 info memory

2. # Memory

3. used_memory:1073740904

4. used_memory_human:1024.00M

5. used_memory_rss:1118720000

6. used_memory_rss_human:1.04G

7. used_memory_peak:1608721680

8. used_memory_peak_human:1.50G

9. used_memory_peak_perc:66.74%

10. used_memory_overhead:253906398

11. used_memory_startup:766152

12. used_memory_dataset:819834506

13. used_memory_dataset_perc:76.41%

14. total_system_memory:67535904768

15. total_system_memory_human:62.90G

16. used_memory_lua:37888

17. used_memory_lua_human:37.00K

18. maxmemory:1073741824

19. maxmemory_human:1.00G

20. maxmemory_policy:allkeys-lru

21. mem_fragmentation_ratio:1.04

22. mem_allocator:jemalloc-4.0.3

23. active_defrag_running:0

24. lazyfree_pending_objects:0

25. $

配置說明:

1. # Enabled active defragmentation

2. # 碎片整理總開關

3. # activedefrag yes

4.  

5. # Minimum amount of fragmentation waste to start active defrag

6. # 內存碎片達到多少的時候開啓整理

7. active-defrag-ignore-bytes 100mb

8.  

9. # Minimum percentage of fragmentation to start active defrag

10. # 碎片率達到百分之多少開啓整理

11. active-defrag-threshold-lower 10

12.  

13. # Maximum percentage of fragmentation at which we use maximum effort

14. # 碎片率小余多少百分比開啓整理

15. active-defrag-threshold-upper 100

16.  

17. # Minimal effort for defrag in CPU percentage

18. active-defrag-cycle-min 25

19.  

20. # Maximal effort for defrag in CPU percentage

21. active-defrag-cycle-max 75

總結: 

從測試的結果看, 效果還是非常不錯的, 另外在配置中我們可以看到如下一段聲明:

說明現在這個功能還是實驗性質的, 對應的命令在官方文檔中都沒有看到. 但是它也說經過了壓力測試, 而且現在也一年多了, 經受了一些考驗, 可以嘗試小流量上線觀察

TODO redis 內存碎片整理實現