[LU-4131] Changelog TRUNC flag is missing

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.4.0
• Fix Version/s: Lustre 2.6.0, Lustre 2.5.2
• Type: Bug
• Priority: Minor
• Reporter: Colin Faber [X] (Inactive)
• Assignee: Nathaniel Clark
• Resolution: Fixed
• Labels: patch

Environment

Single node test system running 2.4.0

• Severity: 3

Description

Relates to LU-2298

It seems that changelog is not logging file truncates. I've not tested 2.4.1 but when I do I'll update this.

[root@coolermaster lustre]# dd if=/dev/zero of=truncate bs=1M count=10
10+0 records in
10+0 records out

Results

7269327 01CREAT 00:36:19.512971520 2013.10.22 0x0 t=[0x200000bde:0x103e8:0x0] p=[0x200000007:0x1:0x0] truncate
7269328 11CLOSE 00:36:19.543968971 2013.10.22 0x242 t=[0x200000bde:0x103e8:0x0]

Truncate (Reduce)

truncate -s -1000 truncate

Results

7269329 14SATTR 00:36:39.677314618 2013.10.22 0xe t=[0x200000bde:0x103e8:0x0]
7269330 11CLOSE 00:36:39.678314537 2013.10.22 0x42 t=[0x200000bde:0x103e8:0x0]

Truncate

{extend}

truncate -s +1000 truncate

Results

7269331 14SATTR 00:36:54.378106399 2013.10.22 0xe t=[0x200000bde:0x103e8:0x0]
7269332 11CLOSE 00:36:54.379106360 2013.10.22 0x42 t=[0x200000bde:0x103e8:0x0]

Possibly the wrong flag 'SATTR' is being sent here, as when I reduce the mask down to just TRUNC flag nothing comes through at all.

-cf

Comments

[Nathan Rutman]

Some additional info:
"Use of truncate(2) does not produce any change log entries, and therefore the action is invisible. Since it appears that all versions of truncate(1) actually do an open(2)/ftruncate(2)/close(2) sequence, the CLOSE event informs of the possible size change. This is only an issue when there is a program using truncate(2) and not following that action with an open for writing and eventual close."
Truncate(2) is specifically referring to the C library function truncate.
This is an issue for us currently. My reading of LU-2298 suggests that this is a problem because size information is not being communicated to the MDS. (Because size on MDS is not implemented.)
Is this correct? As it stands, what would we need to do to get this logged on the MDS?

[Nathan Rutman]

note CL_TRUNC doesn't seem to be actually used anywhere.

[Rahul Deshmukh (Inactive)]

Patch : http://review.whamcloud.com/#/c/9639/

[Cliff White (Inactive)]

The first patch is marked Abandoned, will you be submitting a new patch, or should we close this issue?

[Colin Faber [X] (Inactive)]

Hi Cliff,

The first patch addressed another issue within CL_MTIME and was actually out of scope for this problem. I'll have to check to see where this issue stands on our side and will update this ticket appropriately.

-cf

[Rahul Deshmukh (Inactive)]

Apologies here, actually forgot to mention the new patch link.
I have abandoned first patch as Andreas suggested to created it
for master branch.

Here is the required patch.

Patch: http://review.whamcloud.com/#/c/9723/

I will also put one more cleanup (kind of optional/independent) patch in a day or two.

[Peter Jones]

Landed for 2.5.2 and 2.6