ad784282b6
resolves: rhbz#2059289
Suppress excess messages from nbdkit-nbd-plugin
resolves: rhbz#2083498
Suppress incorrect VDDK error when converting guests from vCenter
resolves: rhbz#2083617
Backport new LUKS filter from 1.32.
Add new Python binding for nbdkit_parse_size from 1.32
Cherry-picked from Fedora:
Add new luks filter.
(Fedora commit 9588e5cbc7)
121 lines
3.6 KiB
Diff
121 lines
3.6 KiB
Diff
From 5d679d01417a81a3a981520d2a0332e2370a2536 Mon Sep 17 00:00:00 2001
|
|
From: "Richard W.M. Jones" <rjones@redhat.com>
|
|
Date: Thu, 21 Apr 2022 16:14:46 +0100
|
|
Subject: [PATCH] readahead: Fix test
|
|
|
|
The previous test turned out to be pretty bad at testing the new
|
|
filter. A specific problem is that the filter starts a background
|
|
thread which issues .cache requests, while on the main connection
|
|
.pread requests are being passed through. The test used
|
|
--filter=readahead --filter=cache with the cache filter only caching
|
|
on .cache requests (since cache-on-read defaults to false), so only
|
|
caching requests made by the background thread.
|
|
|
|
main thread
|
|
client ---- .pread ----- delay-filter -------> plugin
|
|
\
|
|
\ background thread
|
|
.cache --- cache-filter
|
|
|
|
Under very high load, the background thread could be starved. This
|
|
means no requests were being cached at all, and all requests were
|
|
passing through the delay filter. It would appear that readahead was
|
|
failing (which it was, in a way).
|
|
|
|
It's not very easy to fix this since readahead is best-effort, but we
|
|
can go back to using a simpler plugin that logs reads and caches and
|
|
check that they look valid.
|
|
|
|
Update: commit 2ff548d66ad3eae87868402ec5b3319edd12090f
|
|
(cherry picked from commit db1e3311727c6ecab3264a1811d33db1aa45a4d0)
|
|
---
|
|
tests/test-readahead.sh | 61 +++++++++++++++++++++++------------------
|
|
1 file changed, 35 insertions(+), 26 deletions(-)
|
|
|
|
diff --git a/tests/test-readahead.sh b/tests/test-readahead.sh
|
|
index 17126e5a..37f4a06f 100755
|
|
--- a/tests/test-readahead.sh
|
|
+++ b/tests/test-readahead.sh
|
|
@@ -30,43 +30,52 @@
|
|
# OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
|
|
# SUCH DAMAGE.
|
|
|
|
-# Is the readahead filter faster? Copy a blank disk with a custom
|
|
-# plugin that sleeps on every request. Because the readahead filter
|
|
-# should result in fewer requests it should run faster.
|
|
-
|
|
source ./functions.sh
|
|
set -e
|
|
set -x
|
|
|
|
-requires_filter delay
|
|
+requires_plugin sh
|
|
requires nbdsh --version
|
|
requires dd iflag=count_bytes </dev/null
|
|
|
|
-files="readahead.img"
|
|
+files="readahead.out"
|
|
rm -f $files
|
|
cleanup_fn rm -f $files
|
|
|
|
-test ()
|
|
-{
|
|
- start_t=$SECONDS
|
|
- nbdkit -fv -U - "$@" null size=1M --filter=delay rdelay=5 \
|
|
- --run 'nbdsh --uri "$uri" -c "
|
|
+nbdkit -fv -U - "$@" sh - \
|
|
+ --filter=readahead \
|
|
+ --run 'nbdsh --uri "$uri" -c "
|
|
for i in range(0, 512*10, 512):
|
|
h.pread(512, i)
|
|
-"'
|
|
+"' <<'EOF'
|
|
+case "$1" in
|
|
+ thread_model)
|
|
+ echo parallel
|
|
+ ;;
|
|
+ can_cache)
|
|
+ echo native
|
|
+ ;;
|
|
+ get_size)
|
|
+ echo 1M
|
|
+ ;;
|
|
+ cache)
|
|
+ echo "$@" >> readahead.out
|
|
+ ;;
|
|
+ pread)
|
|
+ echo "$@" >> readahead.out
|
|
+ dd if=/dev/zero count=$3 iflag=count_bytes
|
|
+ ;;
|
|
+ *)
|
|
+ exit 2
|
|
+ ;;
|
|
+esac
|
|
+EOF
|
|
|
|
- end_t=$SECONDS
|
|
- echo $((end_t - start_t))
|
|
-}
|
|
+cat readahead.out
|
|
|
|
-t1=$(test --filter=readahead --filter=cache)
|
|
-t2=$(test)
|
|
-
|
|
-# In the t1 case we should make only 1 request into the plugin,
|
|
-# resulting in around 1 sleep period (5 seconds). In the t2 case we
|
|
-# make 10 requests so sleep for around 50 seconds. t1 should be < t2
|
|
-# is every reasonable scenario.
|
|
-if [ $t1 -ge $t2 ]; then
|
|
- echo "$0: readahead filter took longer, should be shorter"
|
|
- exit 1
|
|
-fi
|
|
+# We should see the pread requests, and additional cache requests for
|
|
+# the 32K region following each pread request.
|
|
+for i in `seq 0 512 $((512*10 - 512))` ; do
|
|
+ grep "pread 512 $i" readahead.out
|
|
+ grep "cache 32768 $((i+512))" readahead.out
|
|
+done
|
|
--
|
|
2.31.1
|
|
|