[LU-1478] 1.8.6.81 client with 2.1.1 server MPIIO file locking issue Created: 05/Jun/12  Updated: 14/Nov/13  Resolved: 14/Nov/13

Status: Resolved
Project: Lustre
Component/s: None
Affects Version/s: Lustre 2.1.1, Lustre 1.8.6
Fix Version/s: None

Type: Bug Priority: Minor
Reporter: Mahmoud Hanafi Assignee: Zhenyu Xu
Resolution: Cannot Reproduce Votes: 0
Labels: None
Environment:

Clients: Lustre-1.8.6
Server: Lustre 2.1.1

Attachments: File client_locktrace.gz     File server_locktrace.gz    
Severity: 3
Rank (Obsolete): 7035

 Description   

Running a MPI-IO task requiring concurrent read locks fail using 1.8.6 clients on a 2.1.1 server. The issue is not present in 2.1.1 clients.

Easily duplicated by running a test mpi-io job on a 1.8.6 client. I used this benchmark.
http://public.lanl.gov/jnunez/benchmarks/mpiiotest.htm



 Comments   
Comment by Peter Jones [ 05/Jun/12 ]

Bobijam

Could you please look into this one?

Thanks

Peter

Comment by Zhenyu Xu [ 06/Jun/12 ]

from the log uploaded, the client is requesting to protected read flock (mode=4, PR SETFLK) on inode 144501415241318403, flags=0x40000(LDLM_FL_BLOCK_NOWAIT, SETFLK will return error if other process has protected the file with another flock)

00000080:00010000:1:1338846492.143361:0:76798:0:(file.c:3260:ll_file_flock()) inode=144501415241318403, pid=76798, flags=0x40000, mode=4, start=0, end=1048575

and other process already flock protected the file, so that later requesting client failed. It's a normal operation.

Oleg, does this relates to LU-104? Its comment said that the LU-104 patch has already landed on 1.8, but I didn't find that code in b1_8 branch.

Comment by Mahmoud Hanafi [ 14/Nov/13 ]

We can close this one

Comment by Peter Jones [ 14/Nov/13 ]

ok - thanks Mahmoud

Generated at Sat Feb 10 01:16:59 UTC 2024 using Jira 9.4.14#940014-sha1:734e6822bbf0d45eff9af51f82432957f73aa32c.