Les processus de vidage consomment trop de CPU

Le serveur est une instance EC2, il signifie enregistrer des fichiers sur NAS (NFS) à partir de HTTPD.

Les processus comme flush-0: 32 consomment plus de 90% du processeur et la charge moyenne: 65,50, 64,02, 66,59.

Selon le graphique, il augmente chaque jour alors que la moyenne de charge initiale était de ~ 1,01, 2,02, 1,80 sur 4 cœurs. J'ai ajouté une autre instance similaire sous Load Balancer et son utilisation du processeur n'est que d'environ 6% ATM.

Que font exactement ces processus de vidage?

Peut-être devrions-nous désactiver le cache des attributs NFS si les clients ont seulement besoin d'écrire des données?

Cela pourrait-il être dû à la fragmentation des paquets?

Voici quelques statistiques de nfsstat -s -4:

=================================================================
Server 0:

Server rpc stats:
calls      badcalls   badauth    badclnt    xdrcall
715054137   0          0          0          0       

Server nfs v4:
null         compound     
993       0% 715053143 99% 

Server nfs v4 operations:
op0-unused   op1-unused   op2-future   access       close        commit       
0         0% 0         0% 0         0% 143229323  6% 78092765  3% 36693816  1% 
create       delegpurge   delegreturn  getattr      getfh        link         
3486926   0% 0         0% 0         0% 679872421 28% 158406682  6% 0         0% 
lock         lockt        locku        lookup       lookup_root  nverify      
0         0% 0         0% 0         0% 95872524  4% 0         0% 0         0% 
open         openattr     open_conf    open_dgrd    putfh        putpubfh     
78173920  3% 0         0% 46442107  1% 1668      0% 715044032 29% 0         0% 
putrootfh    read         readdir      readlink     remove       rename       
7110      0% 42081145  1% 9116904   0% 0         0% 7026      0% 1257      0% 
renew        restorefh    savefh       secinfo      setattr      setcltid     
14        0% 81591622  3% 81659244  3% 0         0% 21028018  0% 3244      0% 
setcltidconf verify       write        rellockowner bc_ctl       bind_conn    
3244      0% 0         0% 114086560  4% 0         0% 0         0% 0         0% 
exchange_id  create_ses   destroy_ses  free_stateid getdirdeleg  getdevinfo   
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
getdevlist   layoutcommit layoutget    layoutreturn secinfononam sequence     
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
set_ssv      test_stateid want_deleg   destroy_clid reclaim_comp 
0         0% 0         0% 0         0% 0         0% 0         0% 

=================================================================
Server 1:

Server rpc stats:
calls      badcalls   badauth    badclnt    xdrcall
307172153   0          0          0          0       

Server nfs v4:
null         compound     
427       0% 307171725 99% 

Server nfs v4 operations:
op0-unused   op1-unused   op2-future   access       close        commit       
0         0% 0         0% 0         0% 58451998  5% 32717934  3% 15557564  1% 
create       delegpurge   delegreturn  getattr      getfh        link         
1424670   0% 0         0% 0         0% 291829363 28% 67959378  6% 0         0% 
lock         lockt        locku        lookup       lookup_root  nverify      
0         0% 0         0% 0         0% 41790934  4% 0         0% 0         0% 
open         openattr     open_conf    open_dgrd    putfh        putpubfh     
32741459  3% 0         0% 18993781  1% 75        0% 307167167 30% 0         0% 
putrootfh    read         readdir      readlink     remove       rename       
3108      0% 18598329  1% 3892199   0% 0         0% 742       0% 2         0% 
renew        restorefh    savefh       secinfo      setattr      setcltid     
3         0% 34142743  3% 34166131  3% 0         0% 8963430   0% 1449      0% 
setcltidconf verify       write        rellockowner bc_ctl       bind_conn    
1449      0% 0         0% 54628017  5% 0         0% 0         0% 0         0% 
exchange_id  create_ses   destroy_ses  free_stateid getdirdeleg  getdevinfo   
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
getdevlist   layoutcommit layoutget    layoutreturn secinfononam sequence     
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
set_ssv      test_stateid want_deleg   destroy_clid reclaim_comp 
0         0% 0         0% 0         0% 0         0% 0         0% 

=================================================================
Server 2:

Server rpc stats:
calls      badcalls   badauth    badclnt    xdrcall
53026598   0          0          0          0       

Server nfs v4:
null         compound     
89        0% 53026508 99% 

Server nfs v4 operations:
op0-unused   op1-unused   op2-future   access       close        commit       
0         0% 0         0% 0         0% 9605179   5% 5472897   3% 2633853   1% 
create       delegpurge   delegreturn  getattr      getfh        link         
231276    0% 0         0% 0         0% 50395149 28% 11903036  6% 0         0% 
lock         lockt        locku        lookup       lookup_root  nverify      
0         0% 0         0% 0         0% 7528324   4% 0         0% 0         0% 
open         openattr     open_conf    open_dgrd    putfh        putpubfh     
5477948   3% 0         0% 2760996   1% 15        0% 53025580 30% 0         0% 
putrootfh    read         readdir      readlink     remove       rename       
633       0% 3566001   2% 673466    0% 0         0% 11        0% 0         0% 
renew        restorefh    savefh       secinfo      setattr      setcltid     
2         0% 5704253   3% 5709223   3% 0         0% 1526967   0% 292       0% 
setcltidconf verify       write        rellockowner bc_ctl       bind_conn    
292       0% 0         0% 10439844  5% 0         0% 0         0% 0         0% 
exchange_id  create_ses   destroy_ses  free_stateid getdirdeleg  getdevinfo   
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
getdevlist   layoutcommit layoutget    layoutreturn secinfononam sequence     
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
set_ssv      test_stateid want_deleg   destroy_clid reclaim_comp 
0         0% 0         0% 0         0% 0         0% 0         0% 

=================================================================
Server 3:

Server rpc stats:
calls      badcalls   badauth    badclnt    xdrcall
63045403   0          0          0          0       

Server nfs v4:
null         compound     
121       0% 63045280 99% 

Server nfs v4 operations:
op0-unused   op1-unused   op2-future   access       close        commit       
0         0% 0         0% 0         0% 11504749  5% 6504139   3% 3119453   1% 
create       delegpurge   delegreturn  getattr      getfh        link         
271128    0% 0         0% 0         0% 59865633 28% 14058385  6% 0         0% 
lock         lockt        locku        lookup       lookup_root  nverify      
0         0% 0         0% 0         0% 8852565   4% 0         0% 0         0% 
open         openattr     open_conf    open_dgrd    putfh        putpubfh     
6521385   3% 0         0% 3365913   1% 15        0% 63043988 30% 0         0% 
putrootfh    read         readdir      readlink     remove       rename       
874       0% 4209822   2% 791702    0% 0         0% 6         0% 0         0% 
renew        restorefh    savefh       secinfo      setattr      setcltid     
9         0% 6775367   3% 6792509   3% 0         0% 1811226   0% 409       0% 
setcltidconf verify       write        rellockowner bc_ctl       bind_conn    
409       0% 0         0% 12368747  5% 0         0% 0         0% 0         0% 
exchange_id  create_ses   destroy_ses  free_stateid getdirdeleg  getdevinfo   
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
getdevlist   layoutcommit layoutget    layoutreturn secinfononam sequence     
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
set_ssv      test_stateid want_deleg   destroy_clid reclaim_comp 
0         0% 0         0% 0         0% 0         0% 0         0% 

=================================================================
Server 4:

Server rpc stats:
calls      badcalls   badauth    badclnt    xdrcall
817288490   0          0          0          0       

Server nfs v4:
null         compound     
1204      0% 817287285 99% 

Server nfs v4 operations:
op0-unused   op1-unused   op2-future   access       close        commit       
0         0% 0         0% 0         0% 164320609  6% 89471711  3% 42448842  1% 
create       delegpurge   delegreturn  getattr      getfh        link         
4101436   0% 0         0% 0         0% 778155935 28% 180629867  6% 0         0% 
lock         lockt        locku        lookup       lookup_root  nverify      
0         0% 0         0% 0         0% 109104313  4% 0         0% 0         0% 
open         openattr     open_conf    open_dgrd    putfh        putpubfh     
89598740  3% 0         0% 53534516  1% 9727      0% 817288175 29% 0         0% 
putrootfh    read         readdir      readlink     remove       rename       
8020      0% 46348481  1% 10773529  0% 0         0% 100880    0% 12342     0% 
renew        restorefh    savefh       secinfo      setattr      setcltid     
80        0% 93709338  3% 93712518  3% 0         0% 24303185  0% 3352      0% 
setcltidconf verify       write        rellockowner bc_ctl       bind_conn    
3352      0% 0         0% 127464001  4% 0         0% 0         0% 0         0% 
exchange_id  create_ses   destroy_ses  free_stateid getdirdeleg  getdevinfo   
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
getdevlist   layoutcommit layoutget    layoutreturn secinfononam sequence     
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
set_ssv      test_stateid want_deleg   destroy_clid reclaim_comp 
0         0% 0         0% 0         0% 0         0% 0         0% 

=================================================================
Server 5:

Server rpc stats:
calls      badcalls   badauth    badclnt    xdrcall
804660319   0          0          0          0       

Server nfs v4:
null         compound     
1204      0% 804659114 99% 

Server nfs v4 operations:
op0-unused   op1-unused   op2-future   access       close        commit       
0         0% 0         0% 0         0% 161331719  6% 88318366  3% 41571552  1% 
create       delegpurge   delegreturn  getattr      getfh        link         
4090384   0% 0         0% 0         0% 764533960 28% 177969216  6% 0         0% 
lock         lockt        locku        lookup       lookup_root  nverify      
0         0% 0         0% 0         0% 107385644  4% 0         0% 0         0% 
open         openattr     open_conf    open_dgrd    putfh        putpubfh     
88321805  3% 0         0% 54307425  2% 444       0% 804647492 29% 0         0% 
putrootfh    read         readdir      readlink     remove       rename       
8353      0% 45980723  1% 10410930  0% 0         0% 88471     0% 440       0% 
renew        restorefh    savefh       secinfo      setattr      setcltid     
21        0% 92410970  3% 92412629  3% 0         0% 23733174  0% 3688      0% 
setcltidconf verify       write        rellockowner bc_ctl       bind_conn    
3688      0% 0         0% 125268800  4% 0         0% 0         0% 0         0% 
exchange_id  create_ses   destroy_ses  free_stateid getdirdeleg  getdevinfo   
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
getdevlist   layoutcommit layoutget    layoutreturn secinfononam sequence     
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
set_ssv      test_stateid want_deleg   destroy_clid reclaim_comp 
0         0% 0         0% 0         0% 0         0% 0         0% 

=================================================================
Server 6:

Server rpc stats:
calls      badcalls   badauth    badclnt    xdrcall
795385017   0          0          0          0       

Server nfs v4:
null         compound     
1204      0% 795383812 99% 

Server nfs v4 operations:
op0-unused   op1-unused   op2-future   access       close        commit       
0         0% 0         0% 0         0% 158633282  5% 87331357  3% 41400927  1% 
create       delegpurge   delegreturn  getattr      getfh        link         
4080179   0% 0         0% 0         0% 756063664 28% 176355823  6% 0         0% 
lock         lockt        locku        lookup       lookup_root  nverify      
0         0% 0         0% 0         0% 106692513  4% 0         0% 0         0% 
open         openattr     open_conf    open_dgrd    putfh        putpubfh     
87333398  3% 0         0% 53591273  2% 187       0% 795371861 29% 0         0% 
putrootfh    read         readdir      readlink     remove       rename       
8367      0% 45030006  1% 10352133  0% 0         0% 80473     0% 151       0% 
renew        restorefh    savefh       secinfo      setattr      setcltid     
28        0% 91411943  3% 91413728  3% 0         0% 23629833  0% 3707      0% 
setcltidconf verify       write        rellockowner bc_ctl       bind_conn    
3707      0% 0         0% 124033436  4% 0         0% 0         0% 0         0% 
exchange_id  create_ses   destroy_ses  free_stateid getdirdeleg  getdevinfo   
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
getdevlist   layoutcommit layoutget    layoutreturn secinfononam sequence     
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
set_ssv      test_stateid want_deleg   destroy_clid reclaim_comp 
0         0% 0         0% 0         0% 0         0% 0         0% 

=================================================================
Server 7:

Server rpc stats:
calls      badcalls   badauth    badclnt    xdrcall
801916264   0          0          0          0       

Server nfs v4:
null         compound     
1204      0% 801915059 99% 

Server nfs v4 operations:
op0-unused   op1-unused   op2-future   access       close        commit       
0         0% 0         0% 0         0% 161888929  6% 88285947  3% 41212864  1% 
create       delegpurge   delegreturn  getattr      getfh        link         
4069479   0% 0         0% 0         0% 762072130 28% 177131560  6% 0         0% 
lock         lockt        locku        lookup       lookup_root  nverify      
0         0% 0         0% 0         0% 106437411  3% 0         0% 0         0% 
open         openattr     open_conf    open_dgrd    putfh        putpubfh     
88288337  3% 0         0% 54779036  2% 191       0% 801903449 29% 0         0% 
putrootfh    read         readdir      readlink     remove       rename       
8225      0% 45488312  1% 10259565  0% 0         0% 76243     0% 177       0% 
renew        restorefh    savefh       secinfo      setattr      setcltid     
4         0% 92355900  3% 92357993  3% 0         0% 23515286  0% 3558      0% 
setcltidconf verify       write        rellockowner bc_ctl       bind_conn    
3558      0% 0         0% 123741908  4% 0         0% 0         0% 0         0% 
exchange_id  create_ses   destroy_ses  free_stateid getdirdeleg  getdevinfo   
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
getdevlist   layoutcommit layoutget    layoutreturn secinfononam sequence     
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
set_ssv      test_stateid want_deleg   destroy_clid reclaim_comp 
0         0% 0         0% 0         0% 0         0% 0         0% 

=================================================================
Server 8:

Server rpc stats:
calls      badcalls   badauth    badclnt    xdrcall
804732833   1          1          0          0       

Server nfs v4:
null         compound     
1204      0% 804731628 99% 

Server nfs v4 operations:
op0-unused   op1-unused   op2-future   access       close        commit       
0         0% 0         0% 0         0% 161428939  6% 88340891  3% 41568432  1% 
create       delegpurge   delegreturn  getattr      getfh        link         
4085332   0% 0         0% 0         0% 764396486 28% 177796853  6% 0         0% 
lock         lockt        locku        lookup       lookup_root  nverify      
0         0% 0         0% 0         0% 107176837  3% 0         0% 0         0% 
open         openattr     open_conf    open_dgrd    putfh        putpubfh     
88342790  3% 0         0% 54344886  2% 187       0% 804720008 29% 0         0% 
putrootfh    read         readdir      readlink     remove       rename       
8226      0% 46219141  1% 10381361  0% 0         0% 83380     0% 160       0% 
renew        restorefh    savefh       secinfo      setattr      setcltid     
0         0% 92426602  3% 92428282  3% 0         0% 23736349  0% 3554      0% 
setcltidconf verify       write        rellockowner bc_ctl       bind_conn    
3554      0% 0         0% 125088530  4% 0         0% 0         0% 0         0% 
exchange_id  create_ses   destroy_ses  free_stateid getdirdeleg  getdevinfo   
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
getdevlist   layoutcommit layoutget    layoutreturn secinfononam sequence     
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
set_ssv      test_stateid want_deleg   destroy_clid reclaim_comp 
0         0% 0         0% 0         0% 0         0% 0         0% 

=================================================================
Server 9:

Server rpc stats:
calls      badcalls   badauth    badclnt    xdrcall
800961003   0          0          0          0       

Server nfs v4:
null         compound     
1204      0% 800959798 99% 

Server nfs v4 operations:
op0-unused   op1-unused   op2-future   access       close        commit       
0         0% 0         0% 0         0% 161394704  6% 88264733  3% 41642226  1% 
create       delegpurge   delegreturn  getattr      getfh        link         
4098314   0% 0         0% 0         0% 761225542 28% 172733291  6% 0         0% 
lock         lockt        locku        lookup       lookup_root  nverify      
0         0% 0         0% 0         0% 102217363  3% 0         0% 0         0% 
open         openattr     open_conf    open_dgrd    putfh        putpubfh     
88272429  3% 0         0% 53937975  2% 467       0% 800948292 30% 0         0% 
putrootfh    read         readdir      readlink     remove       rename       
8312      0% 45893437  1% 10409370  0% 0         0% 83127     0% 478       0% 
renew        restorefh    savefh       secinfo      setattr      setcltid     
35        0% 92369729  3% 92371221  3% 0         0% 23772628  0% 3637      0% 
setcltidconf verify       write        rellockowner bc_ctl       bind_conn    
3637      0% 0         0% 124997490  4% 0         0% 0         0% 0         0% 
exchange_id  create_ses   destroy_ses  free_stateid getdirdeleg  getdevinfo   
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
getdevlist   layoutcommit layoutget    layoutreturn secinfononam sequence     
0         0% 0         0% 0         0% 0         0% 0         0% 0         0% 
set_ssv      test_stateid want_deleg   destroy_clid reclaim_comp 
0         0% 0         0% 0         0% 0         0% 0         0%

linux amazon-ec2 linux-kernel

— Roman Newaza
source

Combien c'est trop?

— Fred Foo

puisque la question concerne un bug du noyau signalant la version du noyau aurait aidé.

— Dmitri Chubarov

les processus de vidage sont responsables de la gestion de l'écriture différée des pages sales dans le système de fichiers dont elles sont issues. Ils ne devraient jamais prendre beaucoup de CPU; la plupart de leur temps est consacré à attendre le disque (ou le réseau pour NFS et autres). Si vous constatez une utilisation élevée du processeur lors d'un processus de vidage, il peut s'agir d'un bogue du noyau - essayez de redémarrer, cela devrait effacer l'état.

— bdonlan
source

Salut! Que sont les pages sales? Oui, je viens de terminer l'instance défaillante et les autres fonctionnent normalement ATM.

— Roman Newaza

Les pages sales sont des parties de fichiers qui sont mises en cache en mémoire, ont été modifiées en mémoire par un certain processus, mais n'ont pas encore été réécrites sur le disque. Normalement, cette écriture se produit en arrière-plan, à moins que la quantité de pages sales soit trop élevée, la mémoire libre trop faible ou que le programme ne le demande explicitement - lorsque l'écriture se produit en arrière-plan, le travail de flush est de déclencher réellement cette écriture.

— bdonlan

que se passe-t-il si je cours echo 3 > /proc/sys/vm/drop_cachespour vider la pagecache, les dentiers et les inodes?

— Roman Newaza

@RomanNewaza, cela ne supprimera que les caches qui ne sont pas verrouillées ou sales. le vidage ne fait généralement rien avec les caches non sales, donc je doute que cela fasse une différence (et cela nuira gravement aux performances sur un serveur jusqu'à ce que les caches se réchauffent)

— bdonlan

Je vois. Les problèmes comme le problème ont disparu après la mise à niveau du noyau.

— Roman Newaza