MySQL master database operation large table DDL, slave database crash and system parameter error setting

Here's the thing , A user test UAT library , Frequent error reports from the database without any reason

Memory itself is OK Of . But why MYSQL The version of is official 8.011

First of all, I am getting this question , Want to pass PERCONA Tool set pt-pmp To analyze , But it's starting pt-pmp It was found that it could not run , Direct report virtual memory exhausted

But the problem is that the memory is not used up , We also have SWAP ah , In the end how can not be divided into pieces of memory . I suddenly think of an article I saw before , Is it because overcommit_memory There is a problem with the configuration of , This leads to the obvious memory , But the problem is that it can't be allocated .

First of all, let's get familiar with overcommit  The value of is 0 1 2 

By default, most systems choose 0 , The official documentation is described as applicable to a typical system , Allocations of memory that are significantly larger than expected are rejected , And will use as little as possible SWAP Of memory space

 1   Is a special Liezi , He doesn't care if the memory of the system has been over allocated , This configuration is only applicable to some special applications .

2   And 1 Counter examples of , The memory allocation provided by him will only affect the overall system 50% Distribute , SWAP + Overall memory 50% Is the most he can allocate , When memory cannot be allocated to an application , The system does not OOM application , But the application will receive a memory allocation error .

My first reaction is to query /proc/sys/vm/overcommit_memory , Sure enough, the state inside is 2 . 

Modify the parameters now   sudo  vi /etc/sysctl.conf

And then in the sysctl -p  

Start again PT-PMP command

pt-pmp --binary mysqld --iterations 2 --interval 1  --save-samples mysql.txt

After modification

In the view MYSQL Error log for ,, After modification , There are no errors in the system .  Later others DBA I think it was to test the impact of this parameter on the database , And adjusted the parameters .  Forgot to change it back . But it's good , Through this matter, we can also thoroughly understand overcommit If the parameter is set to... By default 2 ,MYSQL Possible problems .

Related code

1. int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
2. {
3. unsigned long free, allowed;
4. 
   
5. vm_acct_memory(pages);
6. 
   
7. /*
8. * Sometimes we want to use more memory than we have
9. */
10. if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) //overcommit_memory=1, Go straight back to success , No restrictions .
11. return 0;
12. 
    
13. if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { //overcommit_memory=0, Heuristics , Determine whether memory can be allocated according to the free memory status in the current system .
14. unsigned long n;
15. 
    
16. free= global_page_state(NR_FILE_PAGES);
17. free+= nr_swap_pages;
18. 
    
19. /*
20. * Any slabs which are created with the
21. * SLAB_RECLAIM_ACCOUNT flag claim to have contents
22. * which are reclaimable, under pressure. The dentry
23. * cache and most inode caches should fall into this
24. */
25. free+= global_page_state(NR_SLAB_RECLAIMABLE); 
26. 
    
27. /*
28. * Leave the last 3% for root
29. */
30. if (!cap_sys_admin)
31. free-= free / 32; //root Users can go to free less (3%) When , Allocate memory .
32. 
    
33. if (free > pages) // pages The size of memory to be allocated ,free It is calculated according to certain rules “ Free memory size ”, for the first time free Only for NR_FILE_PAGES+NR_SLAB_RECLAIMABLE, Directly or in the system “ Actual idle ” The memory cost is relatively high , So make a step-by-step judgment , Increase of efficiency .
34. return 0;
35. 
    
36. /*
37. * nr_free_pages() is very expensive on large systems,
38. * only call if we're about to fail.
    
39. */
40. n = nr_free_pages(); // When the memory allocation condition is not satisfied for the first time , Proceed again “ Actual idle ” Memory acquisition operation .
41. 
    
42. /*
43. * Leave reserved pages. The pages are not for anonymous pages.
44. */
45. if (n <= totalreserve_pages)
46. goto error;
47. else
48. n -= totalreserve_pages;
49. 
    
50. /*
51. * Leave the last 3% for root
52. */
53. if (!cap_sys_admin)
54. n -= n / 32;
55. free += n;
56. 
    
57. if (free > pages)
58. return 0;
59. 
    
60. goto error;
61. }
62. 
    
63. allowed = (totalram_pages - hugetlb_total_pages()) // When overcommit_memory=2 when , It is limited according to the total amount of virtual address space in the system .
64. * sysctl_overcommit_ratio / 100;
65. /*
66. * Leave the last 3% for root
67. */
68. if (!cap_sys_admin)
69. allowed -= allowed / 32;
70. allowed += total_swap_pages;
71. 
    
72. /* Don't let a single process grow too big:
73. leave 3% of the size of this process for other processes */
74. if (mm)
75. allowed-= mm->total_vm / 32;
76. 
    
77. if (percpu_counter_read_positive(&vm_committed_as) < allowed)
78. return 0;
79. error:
80. vm_unacct_memory(pages);
81. 
    
82. return-ENOMEM;
83. }