[LU-5890] lustre-OST0001: too many transaction credits (1044 > 1024)

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.7.0
• Type: Bug
• Priority: Minor
• Reporter: Liang Zhen (Inactive)
• Assignee: WC Triage
• Resolution: Duplicate
• Severity: 3

Description

when I run IOR on my VM, I always see tons of information like this:

Lustre: 20808:0:(osd_handler.c:1003:osd_trans_start()) lustre-OST0001: too many transaction credits (1036 > 1024)
Lustre: 20808:0:(osd_handler.c:1008:osd_trans_start())   create: 0/0, destroy: 0/0
Lustre: 20808:0:(osd_handler.c:1013:osd_trans_start())   attr_set: 1/1, xattr_set: 1/1
Lustre: 20808:0:(osd_handler.c:1020:osd_trans_start())   write: 2/12, punch: 0/0, quota 4/4
Lustre: 20808:0:(osd_handler.c:1025:osd_trans_start())   insert: 0/0, delete: 0/0
Lustre: 20808:0:(osd_handler.c:1030:osd_trans_start())   ref_add: 0/0, ref_del: 0/0
Pid: 20808, comm: ll_ost_io00_002

Call Trace:
 [<ffffffffa1059895>] libcfs_debug_dumpstack+0x55/0x80 [libcfs]
 [<ffffffffa07acb9f>] osd_trans_start+0x63f/0x660 [osd_ldiskfs]
 [<ffffffffa0ac4f5c>] ofd_trans_start+0x7c/0x100 [ofd]
 [<ffffffffa0acc823>] ofd_commitrw_write+0x543/0x1060 [ofd]
 [<ffffffffa0acd903>] ofd_commitrw+0x5c3/0xae0 [ofd]
 [<ffffffffa115c361>] ? lprocfs_counter_add+0x151/0x1c0 [obdclass]
 [<ffffffffa13fa00d>] obd_commitrw.clone.0+0x11d/0x390 [ptlrpc]
 [<ffffffffa140131c>] tgt_brw_write+0xc7c/0x1530 [ptlrpc]
 [<ffffffffa135b800>] ? target_bulk_timeout+0x0/0xc0 [ptlrpc]
 [<ffffffffa14002ae>] tgt_request_handle+0x71e/0xb10 [ptlrpc]
 [<ffffffffa13afa94>] ptlrpc_main+0xe64/0x1990 [ptlrpc]
 [<ffffffffa13aec30>] ? ptlrpc_main+0x0/0x1990 [ptlrpc]
 [<ffffffff8109aef6>] kthread+0x96/0xa0
 [<ffffffff8100c20a>] child_rip+0xa/0x20
 [<ffffffff8109ae60>] ? kthread+0x0/0xa0
 [<ffffffff8100c200>] ? child_rip+0x0/0x20

Lustre: 21638:0:(osd_handler.c:1003:osd_trans_start()) lustre-OST0001: too many transaction credits (1044 > 1024)
Lustre: 21638:0:(osd_handler.c:1003:osd_trans_start()) Skipped 2 previous similar messages
Lustre: 21638:0:(osd_handler.c:1008:osd_trans_start())   create: 0/0, destroy: 0/0
Lustre: 21638:0:(osd_handler.c:1008:osd_trans_start()) Skipped 2 previous similar messages
Lustre: 21638:0:(osd_handler.c:1013:osd_trans_start())   attr_set: 1/1, xattr_set: 1/1
Lustre: 21638:0:(osd_handler.c:1013:osd_trans_start()) Skipped 2 previous similar messages
Lustre: 21638:0:(osd_handler.c:1020:osd_trans_start())   write: 2/12, punch: 0/0, quota 4/4
Lustre: 21638:0:(osd_handler.c:1020:osd_trans_start()) Skipped 2 previous similar messages
Lustre: 21638:0:(osd_handler.c:1025:osd_trans_start())   insert: 0/0, delete: 0/0
Lustre: 21638:0:(osd_handler.c:1025:osd_trans_start()) Skipped 2 previous similar messages
Lustre: 21638:0:(osd_handler.c:1030:osd_trans_start())   ref_add: 0/0, ref_del: 0/0
Lustre: 21638:0:(osd_handler.c:1030:osd_trans_start()) Skipped 2 previous similar messages
Lustre: 21638:0:(osd_handler.c:1003:osd_trans_start()) lustre-OST0001: too many transaction credits (1552 > 1024)

This is very easy to reproduce:

IOR -b 64m -t 1024 -z -X 3712 -o /mnt/lustre/liang/IOR.file

Issue Links

• is related to: LU-14918 too many ldiskfs transaction credits ...

Comments

[Andreas Dilger]

This is not a significant problem. This is debugging enabled while testing so that we can detect if there are errors in the transaction credit accounting but is not enabled in production.

[Niu Yawei (Inactive)]

There could be something wrong, I don't see why the credits for all operations don't add up right. (looks much less than the total credits)

[Alex Zhuravlev]

Niu, those are "direct" writes in osd_declare_write_commit() - those are added directly to osd_thandle. I think I fixed this in http://review.whamcloud.com/#/c/12169/