At Mon, 27 May 2013 14:23:01 +0800,
Liu Yuan wrote:
>
> +
> +void kick_node_recover(void)
> +{
> + struct vnode_info *old = main_thread_get(current_vnode_info);
> +
> + main_thread_set(current_vnode_info,
> + alloc_vnode_info(old->nodes, old->nr_nodes));
> + start_recovery(main_thread_get(current_vnode_info), old, false);
> + put_vnode_info(old);
> +}
> +
I wonder if starting a recovery without incrementing an epoch is a
good idea. The new free space information is not written to the
current epoch log, so I guess sheep cannot recover objects correctly
if node failure happens while (or after) reweighting.
Actually, this series doesn't pass the below test.
commit 2b0ec4ea35536be672627094d588358adfbfe18d
Author: MORITA Kazutaka <morita.kazutaka at lab.ntt.co.jp>
Date: Mon May 27 16:18:33 2013 +0900
tests: test node failure while reweighting
Signed-off-by: MORITA Kazutaka <morita.kazutaka at lab.ntt.co.jp>
diff --git a/tests/064 b/tests/064
new file mode 100755
index 0000000..2a5c6c0
--- /dev/null
+++ b/tests/064
@@ -0,0 +1,58 @@
+#!/bin/bash
+
+# Test node failure while reweighting
+seq=`basename $0`
+echo "QA output created by $seq"
+
+here=`pwd`
+tmp=/tmp/$$
+status=1 # failure is the default!
+
+# get standard environment, filters and checks
+. ./common.rc
+. ./common.filter
+
+_need_to_be_root
+
+_cleanup
+
+_make_device 0 $((100 * 1024 ** 2)) # 100 MB
+_make_device 1 $((100 * 1024 ** 2)) # 100 MB
+_make_device 2 $((200 * 1024 ** 2)) # 200 MB
+_make_device 3 $((100 * 1024 ** 3)) # 100 GB
+
+#start three in different size
+for i in 0 1 2; do
+ _start_sheep $i
+done
+_wait_for_sheep 3
+$COLLIE cluster format -c 2
+sleep 1
+
+$COLLIE vdi create test 160M -P
+$COLLIE node info
+for i in 0 1 2; do
+ $COLLIE node md info -p 700$i
+done
+
+$COLLIE node md plug $STORE/3
+_wait_for_sheep_recovery 0
+$COLLIE node info
+for i in 0 1 2; do
+ $COLLIE node md info -p 700$i
+done
+
+$COLLIE cluster reweight
+
+# restart sheep1 while reweighting
+_kill_sheep 2
+_wait_for_sheep_recovery 0
+_start_sheep 2
+_wait_for_sheep 2
+
+_wait_for_sheep_recovery 0
+$COLLIE node info
+for i in 0 1 2; do
+ $COLLIE node md info -p 700$i
+done
+$COLLIE cluster info | _filter_cluster_info
diff --git a/tests/064.out b/tests/064.out
new file mode 100644
index 0000000..e69de29
diff --git a/tests/group b/tests/group
index 40a24b0..d8824e3 100644
--- a/tests/group
+++ b/tests/group
@@ -77,3 +77,4 @@
061 auto quick cluster md
062 auto quick cluster md
063 auto quick cluster md
+064 auto quick cluster md
|
|