From: Cyrus Ramavarapu <cramavarapu@microsoft.com>
Date: Wed, 15 Nov 2023 14:30:37 +0000 (+0000)
Subject: LU-17733 tests: sanity test_45 fix dirty count
X-Git-Tag: 2.15.64~101
X-Git-Url: https://git.whamcloud.com/?a=commitdiff_plain;h=14df9b786cf0d40183dad916fe912bc7ca737624;p=fs%2Flustre-release.git

LU-17733 tests: sanity test_45 fix dirty count

Change test comparisons to ge (>=) instead of gt (>). A comment was
added identifying the race condition within the test related to the
completion of async io.  Also updated sync condition to check dirty
bytes after sync is equal to 0

Test-Parameters: testlist=sanity env=ONLY=45
Signed-off-by: Cyrus Ramavarapu <cramavarapu@microsoft.com>
Change-Id: I5826d82a16e20101a2eb8d415bdbde1b6bcc8d69
Reviewed-on: https://review.whamcloud.com/c/fs/lustre-release/+/54763
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-by: Timothy Day <timday@amazon.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>
---

diff --git a/lustre/tests/sanity.sh b/lustre/tests/sanity.sh
index f158339..db72a9e 100755
--- a/lustre/tests/sanity.sh
+++ b/lustre/tests/sanity.sh
@@ -5842,22 +5842,33 @@ test_45() {
 	echo blah > ${f}_grant
 	stop_writeback
 	sync
+
+	# cur_dirty_bytes is related and modified by the OSC cache granting
+	# code. It is possible for the grants to be freed during async IO
+	# completion resulting in 0 vs 0 comparisons for truncates and sync.
+	# Since it can't be reliably proven that the operation did lower the
+	# count, the comparisons are changed to >= (ge) rather than > (gt).
+	#
+	# N.B. This whole test seems flawed as the process doesn't hold file
+	# descriptors, syscalls are not explicit, and dirty data is subject to
+	# async changes outside of the MM subsystem. Some consideration should
+	# be given to removing/rewriting this test.
 	do_dirty_record "echo blah > $f"
-	[[ $before -eq $after ]] && error "write wasn't cached"
+	(( before == after )) && error "write wasn't cached"
 	do_dirty_record "> $f"
-	[[ $before -gt $after ]] || error "truncate didn't lower dirty count"
+	(( before >= after )) || error "truncate didn't lower dirty count"
 	do_dirty_record "echo blah > $f"
-	[[ $before -eq $after ]] && error "write wasn't cached"
+	(( before == after )) && error "write wasn't cached"
 	do_dirty_record "sync"
-	[[ $before -gt $after ]] || error "writeback didn't lower dirty count"
+	(( after == 0 )) || error "writeback didn't zero dirty count"
 	do_dirty_record "echo blah > $f"
-	[[ $before -eq $after ]] && error "write wasn't cached"
+	(( before == after )) && error "write wasn't cached"
 	do_dirty_record "cancel_lru_locks osc"
-	[[ $before -gt $after ]] ||
+	(( before >= after )) ||
 		error "lock cancellation didn't lower dirty count"
 	start_writeback
 }
-run_test 45 "osc io page accounting ============================"
+run_test 45 "osc io page accounting"
 
 # in a 2 stripe file (lov.sh), page 1023 maps to page 511 in its object.  this
 # test tickles a bug where re-dirtying a page was failing to be mapped to the