[linux-lvm] vgscan fails when other nodes quit cleanly.
Xinwei Hu
hxinwei at gmail.com
Tue Mar 9 07:52:55 UTC 2010
Hi all,
Here's an interesting issue. When we shutdown the cluster stack
cleanly, all lvm commands will
fail to grab the global lock. Like this:
--->8----
sys3:~ # vgscan
cluster request failed: Host is down
Unable to obtain global lock.
---8<----
I went through the code history a bit. It seems to be caused by
e65ffb8e, which is for gulm only I think.
--->8----
commit e65ffb8e687bbce4e7edff70ebff2b3f1c0b6157
Author: Christine Caulfield <ccaulfie at redhat.com>
Date: Fri Jun 20 10:58:28 2008 +0000
Make clvmd return immediately if other nodes are down in a gulm cluster.
bz#447799
diff --git a/WHATS_NEW b/WHATS_NEW
index ec7ff54..023659e 100644
--- a/WHATS_NEW
+++ b/WHATS_NEW
@@ -1,5 +1,6 @@
Version 2.02.39 -
================================
+ Make clvmd return immediately if other nodes are down in a gulm cluster.
Improve/Fix read ahead 'auto' calculation for stripe_size
Fix lvchange output for -r auto setting if auto is already set
Add testcase for read ahead
diff --git a/daemons/clvmd/clvmd-gulm.c b/daemons/clvmd/clvmd-gulm.c
index 3a230b5..a2f2148 100644
--- a/daemons/clvmd/clvmd-gulm.c
+++ b/daemons/clvmd/clvmd-gulm.c
@@ -665,6 +665,7 @@ static int _cluster_do_node_callback(struct
local_client *master_client,
{
struct dm_hash_node *hn;
struct node_info *ninfo;
+ int somedown = 0;
dm_hash_iterate(hn, node_hash)
{
@@ -686,12 +687,14 @@ static int _cluster_do_node_callback(struct
local_client *master_client,
client = dm_hash_lookup_binary(sock_hash, csid, GULM_MAX_CSID_LEN);
}
+ DEBUGLOG("down_callback2. node %s, state = %d\n", ninfo->name, ninfo->state);
if (ninfo->state != NODE_DOWN)
callback(master_client, csid, ninfo->state == NODE_CLVMD);
-
+ if (ninfo->state != NODE_CLVMD)
+ somedown = -1;
}
- return 0;
+ return somedown;
}
/* Convert gulm error codes to unix errno numbers */
---8<----
clvmd-corosync.c is copied over from clvmd-openais.c, then from clvmd-gulm.c.
I'd suggest to remove this patch for both clvmd-corosync and clvmd-gulm.
Any comments ?
Thanks.
More information about the linux-lvm
mailing list