From 46708e46366fd846a7d9f9432a7f2ca1787ae0b6 Mon Sep 17 00:00:00 2001
From: Patrick Farrell <pfarrell@whamcloud.com>
Date: Fri, 29 Dec 2023 15:09:49 -0500
Subject: [PATCH] EX-7601 tests: tests for read-modify-write

This patch adds tests for the read-modify-write case for
EX-7601.  There's still some additional tests to be added
here, but this is a good start.

Test-Parameters: trivial
Test-Parameters: testlist=sanity-compr env=ONLY="1004 1005",ONLY_REPEAT=10
Signed-off-by: Patrick Farrell <pfarrell@whamcloud.com>
Change-Id: I5dd9e566b8274ece99283c8962e0d34225089cc0
Reviewed-on: https://review.whamcloud.com/c/ex/lustre-release/+/53230
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-by: Artem Blagodarenko <ablagodarenko@ddn.com>
---
 lustre/tests/sanity-compr.sh | 248 ++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 246 insertions(+), 2 deletions(-)

diff --git a/lustre/tests/sanity-compr.sh b/lustre/tests/sanity-compr.sh
index f80359d..a0f78ee 100644
--- a/lustre/tests/sanity-compr.sh
+++ b/lustre/tests/sanity-compr.sh
@@ -267,8 +267,6 @@ test_1002() {
 		skip "Need MDS version at least 2.14.0.91"
 
 	local tf=$DIR/$tfile
-	# We read from $tfile to this
-	local tf_copy=$DIR/$tfile.copy
 	# Larger than arm page size
 	local chunksize=128
 	local read_ahead_mb
@@ -475,6 +473,252 @@ test_1003() {
 }
 run_test 1003 "mmap test for compression"
 
+test_1004() {
+	(( MDS1_VERSION >= $(version_code 2.14.0.91) )) ||
+		skip "Need MDS version at least 2.14.0.91"
+
+	local tf=$DIR/$tfile
+	local hdf=$LUSTRE/tests/AMSR_E_L3_DailyOcean_V05_20111003.hdf
+	local tmp_hdf=$TMP/$tfile.hdf
+	local source=$tmp_hdf
+	# Larger than arm page size
+	local chunksize=128
+
+	if [[ -f $hdf.bz2 ]] && type -p bzcat >/dev/null; then
+		bzcat $hdf.bz2 > $tmp_hdf
+	elif [[ -f $hdf.bz2 ]] && type -p bunzip2 >/dev/null; then
+		cp $hdf.bz2 $tmp_hdf.bz2 || error "cp $tmp_hdf.bz2"
+		bunzip2 $tmp_hdf.bz2 || error "bunzip2 $tmp_hdf.bz2"
+	else
+		echo "bunzip2 is not installed, skip it"
+		return 0
+	fi
+
+	# Fail test if source size changes so we catch this
+	# Source should be a few MiB in size
+	$CHECKSTAT -s 14625450 $source || error "checkstat wrong size"
+
+	stack_trap "rm -f $tf; disable_compression"
+	enable_compression
+
+	# Simple compressed layout
+	$LFS setstripe -E -1 -Z lz4:0 --compress-chunk=$chunksize $tf ||
+		error "set a compress component in $tf failed"
+
+	# Create file and verify - trivial
+	dd if=$source bs=${chunksize}K of=$tf || error "(0) dd failed"
+	flush_and_compare $source $tf "(1)"
+
+	# Do a single 4K write - this will require a read-modify-write because
+	# it is less than chunk size
+	dd if=$source bs=4K of=$tf count=1 conv=notrunc || error "(2) dd failed"
+	flush_and_compare $source $tf "(3)"
+
+	# A single write of > chunk_size but < 2 chunks
+	dd if=$source bs=$((chunksize * 3/2))K of=$tf count=1 conv=notrunc ||
+		error "(4) dd failed"
+	flush_and_compare $source $tf "(5)"
+
+	# Same test but offset slightly in to the file
+	dd if=$source seek=1 skip=1 bs=$((chunksize * 3/2))K of=$tf count=1 conv=notrunc ||
+		error "(6) dd failed"
+	flush_and_compare $source $tf "(7)"
+
+	# Later tests use fsync to force read-modify-write
+	# larger dd size with fsync (writing 4K at a time with fsync is slow)
+	dd if=$source bs=$((chunksize / 2)) of=$tf conv=fsync,notrunc ||
+		error "(8) dd failed"
+	flush_and_compare $source $tf "(9)"
+
+	# Larger than chunk size
+	dd if=$source bs=$((chunksize * 3/2)) of=$tf conv=fsync,notrunc ||
+		error "(10) dd failed"
+	flush_and_compare $source $tf "(11)"
+
+	# The above test is full pages on x86, sometimes partial on ARM
+	# This will explicitly test readup of partial chunks as part of a write
+	# This writes one full page, then part of the next page
+	# This forces a partial page read before we can do the write
+	dd if=$source bs=$((PAGE_SIZE * 2 - 1024)) of=$tf conv=notrunc ||
+		error "(12) dd failed"
+	flush_and_compare $source $tf "(13)"
+
+	# Do the same test at an offset in the file
+	dd if=$source seek=4 skip=4 bs=$((PAGE_SIZE * 2 - 1024)) of=$tf conv=notrunc ||
+		error "(14) dd failed"
+	flush_and_compare $source $tf "(15)"
+}
+run_test 1004 "initial test for write updating"
+
+test_1005() {
+	(( MDS1_VERSION >= $(version_code 2.14.0.91) )) ||
+		skip "Need MDS version at least 2.14.0.91"
+
+	local tf=$DIR/$tfile
+	# We read from $tfile to this
+	# Larger than arm page size
+	local chunksize=128
+	local read_ahead_mb
+	local hdf=$LUSTRE/tests/AMSR_E_L3_DailyOcean_V05_20111003.hdf
+	local tmp_hdf=$TMP/$tfile.hdf
+	local source=$tmp_hdf
+
+	if [[ -f $hdf.bz2 ]] && type -p bzcat >/dev/null; then
+		bzcat $hdf.bz2 > $tmp_hdf
+	elif [[ -f $hdf.bz2 ]] && type -p bunzip2 >/dev/null; then
+		cp $hdf.bz2 $tmp_hdf.bz2 || error "cp $tmp_hdf.bz2"
+		bunzip2 $tmp_hdf.bz2 || error "bunzip2 $tmp_hdf.bz2"
+	else
+		echo "bunzip2 is not installed, skip it"
+		return 0
+	fi
+
+	# Fail test if source size changes so we catch this
+	# Source should be a few MiB in size
+	$CHECKSTAT -s 14625450 $source || error "checkstat wrong size"
+
+	stack_trap "rm -f $tf; disable_compression"
+	enable_compression
+
+	# Disable readahead so reads are not expanded to full chinks
+	$LCTL set_param osc.*.rpc_stats=c
+	read_ahead_mb=$($LCTL get_param -n llite.*.max_read_ahead_mb)
+	$LCTL set_param llite.*.max_read_ahead_mb=0
+	stack_trap "$LCTL set_param llite.*.max_read_ahead_mb=$read_ahead_mb" EXIT
+
+	# Simple compressed layout
+	$LFS setstripe -E -1 -Z lz4:0 --compress-chunk=$chunksize $tf ||
+		error "set a compress component in $tf failed"
+
+	# These tests will deliberately create unusual files, using the sample
+	# hdf5 file as a reference, but cannot compare to it because they
+	# copy only part of it.  So all of them will also create an identical
+	# uncompressed file to compare against.
+	# Create the simplest possible example of a file with an incomplete
+	# chunk, just 4K at the beginning (this won't be compressed, but it's
+	# still a valid test)
+	dd if=$source bs=4K of=$tf count=1 || error "(0) dd failed"
+	dd if=$source bs=4K of=$tf.2 count=1 || error "(1) dd failed"
+
+	flush_and_compare $tf $tf.2 "(2)"
+
+	# Now we write adjacent in this first chunk, confirming updating it works
+	dd if=$source bs=4K of=$tf conv=notrunc count=2 seek=1 skip=1 || error "(3) dd failed"
+	dd if=$source bs=4K of=$tf.2 conv=notrunc count=2 seek=1 skip=1 || error "(4) dd failed"
+
+	flush_and_compare $tf $tf.2 "(5)"
+
+	# Now we jump a bit further on and do a write
+	dd if=$source bs=8K of=$tf conv=notrunc count=1 seek=4 skip=4 || error "(6) dd failed"
+	dd if=$source bs=8K of=$tf.2 conv=notrunc count=1 seek=4 skip=4 || error "(7) dd failed"
+	sync
+
+	flush_and_compare $tf $tf.2 "(8)"
+
+	# None of the above was compressed, because the first write was too
+	# small and the later writes weren't chunk aligned.  Confirm the server
+	# didn't do any decompression/compression
+	#FIXME: Stats check goes here once available
+
+	# OK, now we truncate the file back to zero and start with a compressed
+	# chunk at the beginning.
+	# 16K - this will be compressed
+	dd if=$source bs=16K of=$tf count=1 || error "(9) dd failed"
+	dd if=$source bs=16K of=$tf.2 count=1 || error "(10) dd failed"
+
+	flush_and_compare $tf $tf.2 "(11)"
+
+	# Now we do the same size of write slightly further down
+	# 16K, in to an existing chunk - leads to read-modify-write
+	dd if=$source bs=16K of=$tf count=1 seek=2 skip=2 conv=notrunc || error "(12) dd failed"
+	dd if=$source bs=16K of=$tf.2 count=1 seek=2 skip=2 conv=notrunc || error "(13) dd failed"
+
+	flush_and_compare $tf $tf.2 "(14)"
+
+	# OK, now we're going to create a complete compressed chunk further
+	# along in the file
+	dd if=$source bs=128K count=1 skip=1 seek=1 conv=notrunc of=$tf ||
+		error "(15) dd failed"
+	dd if=$source bs=128K count=1 skip=1 seek=1 conv=notrunc of=$tf.2 ||
+		error "(16) dd failed"
+
+	flush_and_compare $tf $tf.2 "(17)"
+
+	# Now we're going to write *into* the middle of this compressed chunk
+	dd if=$source bs=32K count=1 skip=5 seek=5 conv=notrunc of=$tf ||
+		error "(18) dd failed"
+	dd if=$source bs=32K count=1 skip=5 seek=5 conv=notrunc of=$tf.2 ||
+		error "(19) dd failed"
+
+	flush_and_compare $tf $tf.2 "(20)"
+
+	# OK, now we're going to add another incomplete chunk after those
+	# two - starting in the third chunk, ie, offset of 256K
+	# This will be compressed
+	dd if=$source bs=64K count=1 skip=4 seek=4 conv=notrunc of=$tf ||
+		error "(21) dd failed"
+	dd if=$source bs=64K count=1 skip=4 seek=4 conv=notrunc of=$tf.2 ||
+		error "(22) dd failed"
+
+	flush_and_compare $tf $tf.2 "(23)"
+
+	# Now we're going to write the second 'half' of this chunk
+	dd if=$source bs=64K count=1 skip=5 seek=5 conv=notrunc of=$tf ||
+		error "(24) dd failed"
+	dd if=$source bs=64K count=1 skip=5 seek=5 conv=notrunc of=$tf.2 ||
+		error "(25) dd failed"
+
+	flush_and_compare $tf $tf.2 "(26)"
+
+	# And now we're going to add a chunk that doesn't start at offset 0,
+	# so it won't be compressed.  This is the fourth chunk, so starts at
+	# 384K.  So we'll start at 448K
+	dd if=$source bs=64K count=1 skip=7 seek=7 conv=notrunc of=$tf ||
+		error "(27) dd failed"
+	dd if=$source bs=64K count=1 skip=7 seek=7 conv=notrunc of=$tf.2 ||
+		error "(28) dd failed"
+
+	flush_and_compare $tf $tf.2 "(29)"
+
+	# Now we'll write a few KiB in to the middle of the first part of this
+	# chunk - starting at 400K
+	dd if=$source bs=8K count=2 skip=50 seek=50 conv=notrunc of=$tf ||
+		error "(30) dd failed"
+	dd if=$source bs=8K count=2 skip=50 seek=50 conv=notrunc of=$tf.2 ||
+		error "(31) dd failed"
+
+	flush_and_compare $tf $tf.2 "(32)"
+
+	# Now let's skip an entire chunk and do a full chunk at 640K
+	dd if=$source bs=128K count=1 skip=5 seek=5 conv=notrunc of=$tf ||
+		error "(33) dd failed"
+	dd if=$source bs=128K count=1 skip=5 seek=5 conv=notrunc of=$tf.2 ||
+		error "(34) dd failed"
+
+	flush_and_compare $tf $tf.2 "(35)"
+
+	# Then one more partial, but compressible, chunk after that at
+	# 768K
+	dd if=$source bs=64K count=1 skip=12 seek=12 conv=notrunc of=$tf ||
+		error "(36) dd failed"
+	dd if=$source bs=64K count=1 skip=12 seek=12 conv=notrunc of=$tf.2 ||
+		error "(37) dd failed"
+
+	flush_and_compare $tf $tf.2 "(38)"
+
+	# OK, now we have this complex file, let's test reading it at a
+	# number of block sizes to give us unusual offsets
+	echo "copying from compressed file with $bs"
+	for bs in 3 4 7 32 97 128 130 192; do
+		dd if=$tf bs=${bs}K of=$tf.3 ||
+			error "(39) dd with block size ${bs}K failed"
+
+		flush_and_compare $tf $tf.3 "(40)"
+		rm -f $tf.3
+	done
+}
+run_test 1005 "test for write updating with partial chunks"
+
 complete_test $SECONDS
 check_and_cleanup_lustre
 declare -a logs=($ONLY)
-- 
1.8.3.1