lars.ellenberg at linbit
Feb 20, 2012, 12:40 PM
Views: 296
Permalink
|
What do you say?
commit e57f9f7a7b8f838e750864ce332436a0f36321d7
Author: Lars Ellenberg <lars.ellenberg [at] linbit>
Date: Mon Feb 20 20:38:49 2012 +0100
Medium: Filesystem: improve read/write checks for CHECK_LEVEL 10, 20
Some devices don't support 512 byte direct writes anymore,
the "physical" block size is 4kByte.
Changing "bs=512 count=16" to "bs=4k count=1" also helps to avoid
spurious timeouts. I've seen badly tuned (and/or totally overloaded) IO
stacks exposing a synchronous single-request IO latency of many seconds
(under load). If we need 16 of those, we can easily violate timeout
settings.
diff --git a/heartbeat/Filesystem b/heartbeat/Filesystem
index a31931b..5f1ecfe 100755
--- a/heartbeat/Filesystem
+++ b/heartbeat/Filesystem
@@ -837,7 +837,7 @@ Filesystem_monitor_10()
ocf_log warn "$DEVICE is not a block device, monitor 10 is noop"
return $OCF_SUCCESS
fi
- dd_opts="iflag=direct bs=512 count=16"
+ dd_opts="iflag=direct bs=4k count=1"
err_output=`dd if=$DEVICE $dd_opts 2>&1 >/dev/null`
if [ $? -ne 0 ]; then
ocf_log err "Failed to read device $DEVICE"
@@ -853,11 +853,11 @@ Filesystem_monitor_20()
{
if [ "$blockdevice" = "no" ] ; then
# O_DIRECT not supported on cifs/smbfs
- dd_opts="oflag=sync bs=512 conv=fsync,sync"
+ dd_opts="oflag=sync bs=4k conv=fsync,sync"
else
# Writing to the device in O_DIRECT mode is imperative
# to bypass caches.
- dd_opts="oflag=direct,sync bs=512 conv=fsync,sync"
+ dd_opts="oflag=direct,sync bs=4k conv=fsync,sync"
fi
status_dir=`dirname $STATUSFILE`
[ -d "$status_dir" ] ||
_______________________________________________________
Linux-HA-Dev: Linux-HA-Dev [at] lists
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
|
[RfC] [Patch] Filesystem
