[LU-6910] Configurable values for OST reserved size

• Type: Improvement
• Resolution: Fixed
• Priority: Major
• Fix Version/s: Lustre 2.9.0
• Labels: patch
• Assignee: WC Triage
• Reporter: Alexander Boyko

Description

osp_pre_status=-ENOSPC is used to skip OST from object allocation. The error was set when OST available space is less than 0.1% of total OST size. This value is not configurable. For many cases 0.1% is too low, because append works for existing files. So, procfs files was added for configuration.

Issue Links

• is blocked by: LUDOC-317 document OST reserved space tunables
• is related to: LU-7552 sanity test_253: Error: 'File creation should fail'
• is related to: LU-7550 sanity test_27C: FAIL: Can not find 5 in obdidx 0 1 2 3 4 6

Comments

[Gerrit Updater]

Alexander Boyko ([e-mail]) uploaded a new patch: http://review.whamcloud.com/17656
Subject: LU-6910 osp: add procfs values for OST reserved size
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 5106f1ff901e24734970d8ce0294f5ed0c20c1f3

[Andreas Dilger]

It looks like the last few comments posted in LU-7550 after the patch was reverted should resolve the failure seen there when the patch is resubmitted:

diff --git a/lustre/tests/sanity.sh b/lustre/tests/sanity.sh
index 7a25e4b..19e8521 100755
--- a/lustre/tests/sanity.sh
+++ b/lustre/tests/sanity.sh
@@ -1885,6 +1885,7 @@ check_seq_oid()
                                $(facet_mntpt ost$ost)/$obj_file)
                        unmount_fstype ost$ost
                        start ost$ost $dev $OST_MOUNT_OPTS
+                       clients_up
                fi

As yet the LU-7552 problem has not been fixed.

[Andreas Dilger]

Alexander, could you please resubmit your patch once you have fixed the two test failures (LU-7550 and LU-7552. To run those tests multiple times to get a good confidence about the fix, you should use:

Test-Parameters: testlist=sanity,sanity,sanity,sanity,sanity,sanity
Test-Parameters: testlist=sanity,sanity,sanity,sanity,sanity,sanity

To run a bunch of tests and two sets in parallel so it doesn't take so long.

There was also a code late review comment internally that returning -ENOSPC from ->statfs() is probably not the right thing, because this may cause unexpected errors if the caller just wants to know the space usage. Instead, it should set OS_STATE_ENOSPC or OS_STATE_ENOINO in the os_state field and the caller should check that. Having two separate flags also avoids the need to guess what the -ENOSPC error is for.

[Gerrit Updater]

Andreas Dilger ([e-mail]) merged in patch http://review.whamcloud.com/17617/
Subject: Revert "LU-6910 osp: add procfs values for OST reserved size"
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: acc918d76856ff14306c543c74e6ceef3865bcbc

[Peter Jones]

Patch reverted. See LU-7550 for details

[Gerrit Updater]

Andreas Dilger ([e-mail]) uploaded a new patch: http://review.whamcloud.com/17617
Subject: Revert "LU-6910 osp: add procfs values for OST reserved size"
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 266fd420fb049d23f3c85b99ff56b9f920148c0a

[Peter Jones]

Landed for 2.8

[Andreas Dilger]

This also fixed a couple of bugs in the handling of nearly full OSTs.

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch http://review.whamcloud.com/15731/
Subject: LU-6910 osp: add procfs values for OST reserved size
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 0585b0fb5895a24f07ca32e830d1fa72b75f4f2b

[Gerrit Updater]

[ removed unrelated patch comment ]

[Alexander Boyko]

Patch http://review.whamcloud.com/#/c/15731/