LVM2 test script failed, is it a read_ahead bug?

LVM2 test script failed, is it a read_ahead bug?

[Dongmao Zhang]

Dear mail-list,
    I met this in running the test case for lvm:
sh shell/read-ahead.sh FAILED:

#COMM "read ahead is properly inherited from underlying PV"
blockdev --setra 768 "$dev1"
#read-ahead.sh:34+ blockdev --setra 768 @TESTDIR@/dev/mapper/@PREFIX at pv1
vgscan
#read-ahead.sh:35+ vgscan
  Reading all physical volumes.  This may take a while...
  Found volume group "@PREFIX at vg" using metadata type lvm2
  WARNING: This metadata update is NOT backed up
lvcreate -n $lv -L4m $vg "$dev1"
#read-ahead.sh:36+ lvcreate -n LV -L4m @PREFIX at vg
@TESTDIR@/dev/mapper/@PREFIX at pv1
  WARNING: This metadata update is NOT backed up
  WARNING: This metadata update is NOT backed up
  Logical volume "LV" created
test $(blockdev --getra $DM_DEV_DIR/$vg/$lv) -eq 768
blockdev --getra $DM_DEV_DIR/$vg/$lv
##read-ahead.sh:37+ blockdev --getra @TESTDIR@/dev/@PREFIX at vg/LV
#read-ahead.sh:37+ test 1024 -eq 768

The read_ahead of underlying device was set to 768. After creating the
logical volume device, lvm found the auto read ahead was 1024 which is
bigger than 768. So lvm selected the bigger number(1024) to the
read_ahead of logical volume.

However I think we should use the minimal read_ahead in underlying
device and auto-detected kernel parameter. if so, this behavior could be
consistent with the test script.

I suggest use:

diff --git a/libdm/libdm-common.c b/libdm/libdm-common.c
index 9043352..b9000f9 100644
--- a/libdm/libdm-common.c
+++ b/libdm/libdm-common.c
@@ -1225,7 +1225,7 @@ static int _set_dev_node_read_ahead(const char
*dev_name,
                if (!get_dev_node_read_ahead(dev_name, major, minor,
&current_read_ahead))
                        return_0;

-               if (current_read_ahead >= read_ahead) {
+               if (current_read_ahead <= read_ahead) {
                        log_debug_activation("%s: retaining kernel read
ahead of %" PRIu32
                                  " (requested %" PRIu32 ")",
                                  dev_name, current_read_ahead, read_ahead);

If I am wrong, please correct me:)

thank you

LVM2 test script failed, is it a read_ahead bug?

[Zdenek Kabelac]

Dne 27.8.2013 12:00, Dongmao Zhang napsal(a):
> Dear mail-list,
>      I met this in running the test case for lvm:
> sh shell/read-ahead.sh FAILED:
>

>
> -               if (current_read_ahead >= read_ahead) {
> +               if (current_read_ahead <= read_ahead) {
>                          log_debug_activation("%s: retaining kernel read
> ahead of %" PRIu32
>                                    " (requested %" PRIu32 ")",
>                                    dev_name, current_read_ahead, read_ahead);
>
> If I am wrong, please correct me:)
>


Basic rules for 'auto'mated read_ahead setting is simple: never set smaller 
value then the one which has been already set by kernel when device has been 
initialized.

If you need any specific more 'clever' value - simply use direct value for
lvcreate/lvchange --readahead.

There is no thing like 'minimal' RA estimation in lvm2.

It's quite tricky to estimate the ideal RA for devices and it mostly depends
on the type of use.

Zdenek