[LU-13850] File in readwrite state and "No data available" error after file was added to cache on one node.

• Status: Closed
• Project: Lustre
• Affects Version/s: Lustre 2.14.0
• Type: Bug
• Priority: Major
• Reporter: Vikentsi Lapa
• Assignee: Qian Yingjin
• Resolution: Fixed
• Severity: 3

Description

 

file was created on one node and on that node was created rule to add file to cache.

lctl pcc add /mnt/lustre /mnt/pccro --param "fname={*.doc} roid=7 ropcc=1"

 

pdsh -w trevis-59vm1[0,1,2] lfs pcc state /mnt/lustre/myfile.doc
trevis-59vm12: file: /mnt/lustre/myfile.doc, type: readwrite, PCC file: /0007/0000/0402/0000/0002/0000/0x200000402:0x7:0x0, user number: 0, flags: 0
trevis-59vm10: file: /mnt/lustre/myfile.doc, type: none
trevis-59vm11: file: /mnt/lustre/myfile.doc, type: none

on attempt to read file from other nodes I seeing this error.  and file type is marked as readwrite. Epected output is readonly.

pdsh -w trevis-59vm1[0,1,2] cat /mnt/lustre/myfile.doc
trevis-59vm11: cat: /mnt/lustre/myfile.doc: No data available
trevis-59vm12: filedata
trevis-59vm10: cat: /mnt/lustre/myfile.doc: No data available

Last time when I was checked this it works successfully file showed  status

 

# pdsh -w trevis-59vm1[0,1,2]  /usr/sbin/lctl pcc list /mnt/lustre  | dshbak
----------------
trevis-59vm10
----------------
pcc:
  -
    pccpath: /mnt/pccro
    hsmtool: posix
    rwid: 0
    roid: 7
    flags: 2e
    autocache: fname={*.doc}
----------------
trevis-59vm11
----------------
pcc:
  -
    pccpath: /mnt/pccro
    hsmtool: posix
    rwid: 0
    roid: 7
    flags: 2e
    autocache: fname={*.doc}
----------------
trevis-59vm12
----------------
pcc:
  -
    pccpath: /mnt/pccro
    hsmtool: posix
    rwid: 0
    roid: 7
    flags: 2e
    autocache: fname={*.doc}

 

Comments

[James Nunez (Inactive)]

Yingjin,

Would you please review this issue and comment?

[Qian Yingjin]

Could you please give out the full command lines that caused this errors?

pdsh -w trevis-59vm1[0,1,2] lfs pcc state /mnt/lustre/myfile.doc
trevis-59vm12: file: /mnt/lustre/myfile.doc, type: readwrite, PCC file: /0007/0000/0402/0000/0002/0000/0x200000402:0x7:0x0, user number: 0, flags: 0

How the file become PCC-RW (readwrite) state?

 

Thanks,

Qian

 

[Qian Yingjin]

I reproduced the failure on my local machine.

Will update the patch soon.

 

Thanks,

Qian

[Qian Yingjin]

Fix this bug in https://review.whamcloud.com/#/c/38346/ 
LU-10918 pcc: auto RO-PCC caching when O_RDONLY open files

Please try the topmost patch:
https://review.whamcloud.com/#/c/38352
LU-12373 pcc: uncache the pcc copies when remove a PCC backend

[Vikentsi Lapa]

Thank you, I started to test it.

[Vikentsi Lapa]

Confirmation that bug was fixed.  Verification was done for lustre-2.13.55_3_g066eec1-1.el7.x86_64 version.

[root@trevis-59vm9 tests]# pdsh -w trevis-59vm[10-12]  bash /mnt/lustre/check_ro_cache.sh  | dshbak
----------------
trevis-59vm10
----------------
pcc:
  -
    pccpath: /mnt/pccro
    hsmtool: posix
    rwid: 0
    roid: 7
    flags: 2e
    autocache: fname={*.doc}
----------------
trevis-59vm11
----------------
pcc:
  -
    pccpath: /mnt/pccro
    hsmtool: posix
    rwid: 0
    roid: 7
    flags: 2e
    autocache: fname={*.doc}
----------------
trevis-59vm12
----------------
pcc:
  -
    pccpath: /mnt/pccro
    hsmtool: posix
    rwid: 0
    roid: 7
    flags: 2e
    autocache: fname={*.doc}
[root@trevis-59vm9 tests]# echo “YYYYY” > /mnt/lustre/fileabcd.doc
[root@trevis-59vm9 tests]#  pdsh -w trevis-59vm[10-12] lfs pcc state /mnt/lustre/fileabcd.doc
trevis-59vm12: file: /mnt/lustre/fileabcd.doc, type: readonly, PCC file: /002e/0000/0401/0000/0002/0000/0x200000401:0x2e:0x0, user number: 0, flags: 0
trevis-59vm10: file: /mnt/lustre/fileabcd.doc, type: readonly, PCC file: /002e/0000/0401/0000/0002/0000/0x200000401:0x2e:0x0, user number: 0, flags: 0
trevis-59vm11: file: /mnt/lustre/fileabcd.doc, type: readonly, PCC file: /002e/0000/0401/0000/0002/0000/0x200000401:0x2e:0x0, user number: 0, flags: 0

[Vikentsi Lapa]

Verified with lustre-client-2.13.55_3_g066eec1-1.el7.x86_64