On Mon, Sep 17, 2018 at 3:51 PM pengyixiang <yxpengi386@163.com> wrote:
Hello, everyone!
      if we reload lvs, why we don't update lvs that seg_start_pe != "0" ? there's in [1] with red mark.
what seg_start_pe means?

It is kind of hidden in lvm manual, but if you read carefully lvs(8):

       -o|--options String
              Comma-separated, ordered list of fields to display in columns.  String arg syntax is: [+|-|#]Field1[,Field2 ...]  The prefix + will append the specified fields to the
              default fields, - will remove the specified fields from the default fields, and # will compact specified fields (removing them when empty for all rows.)  Use -o help to
              view the list of all available fields.  Use separate lists of fields to add, remove or compact by repeating the -o option: -o+field1,field2 -o-field3,field4 -o#field5.
              These lists are evaluated from left to right.  Use field name lv_all to view all LV fields, vg_all all VG fields, pv_all all PV fields, pvseg_all all PV segment fields,
              seg_all all LV segment fields, and pvseg_all all PV segment columns.  See the lvm.conf report section for more config options.  See lvmreport(7) for more information about
              reporting.

and you follow the instructions:

    "Use -o help to view the list of all available fields"

You will find:

$ lvs -o help 2>&1 | grep seg_start_pe
    seg_start_pe           - Offset within the LV to the start of the segment in physical extents.

 
If this lv is a vm template, may seg_start_pe is not zero, and we
should ignore it?

This is not related to templates. At this level vdsm lvm module does not know
anything about templates.

I think the issue is getting multiple lines for same lv in lvs output, when you
specify -o devices.

If an lv has several segments, lvs will return separate line for every segment.

You can simulate this case with these commands:

$ sudo losetup -f /tmp/backing --show
/dev/loop2

$ sudo pvcreate /dev/loop2
  Physical volume "/dev/loop2" successfully created.

$ sudo vgcreate test /dev/loop2
  Volume group "test" successfully created

$ sudo lvcreate -n lv1 -L 100m test
  Logical volume "lv1" created.

$ sudo lvcreate -n lv2 -L 100m test
  Logical volume "lv2" created.

$ sudo lvextend -L 200m test/lv1
  Size of logical volume test/lv1 changed from 100.00 MiB (25 extents) to 200.00 MiB (50 extents).
  Logical volume test/lv1 successfully resized.

$ sudo lvs -o lv_name,seg_start_pe,seg_pe_ranges,devices test
  LV   Start PE Ranges        Devices       
  lv1      0 /dev/loop2:0-24  /dev/loop2(0) 
  lv1     25 /dev/loop2:50-74 /dev/loop2(50)
  lv2      0 /dev/loop2:25-49 /dev/loop2(25)

Looks like the only reason we query seg_start_pe is to filter the lines about the first
segment of an lv.

... 
        updatedLVs = {}
        for line in out:
            fields = [field.strip() for field in line.split(SEPARATOR)]
            lv = makeLV(*fields)
            # For LV we are only interested in its first extent
        if lv.seg_start_pe == "0":
                self._lvs[(lv.vg_name, lv.name)] = lv
                updatedLVs[(lv.vg_name, lv.name)] = lv
...

This stores the info about the first segment in the cache, and drop possibly
other lines about the next segments.

We probably can eliminate multiple lines by using --select, which was not availabe
when this code was written:

$ sudo lvs -o lv_name,seg_pe_ranges,devices test --select 'seg_start_pe=0'
  LV   PE Ranges        Devices       
  lv1  /dev/loop2:0-24  /dev/loop2(0) 
  lv2  /dev/loop2:25-49 /dev/loop2(25)

It would be nice if someone would have time to do this.

Cheers,
Nir