Duplicate LVs returned from bd_lvm_lvs()

[wfpudak]

Hi,

We've noticed that if an LV has several segments as noted by lvdisplay

--- Logical volume ---
LV Path /dev/rootvg/test
LV Name test
VG Name rootvg
LV UUID TgjM5V-jhXX-tPTf-gPMa-c3w6-QEjo-7qkmVg
LV Write Access read/write
LV Creation host, time du1, 2021-10-25 15:59:20 +0000
LV Status available
"#" open 1
LV Size 68.00 MiB
Current LE 68
Segments 2
Allocation inherit
Read ahead sectors 1024
Block device 253:166

the API method bd_lvm_lvs() will return several entries since segtype is used with the -o option.

It can easily be reproduced with the lvs command itself as well. This happens due to fragmentation as described in : https://askubuntu.com/questions/380424/understanding-output-of-lvdisplay-in-lvm-linux

Is the expectation from bd_lvm_lvs() to return those duplicates or shouldn't the lib code filter them out?

Example lvs output with your flags if there are several segments :

LVM2_VG_NAME=rootvg LVM2_LV_NAME=test LVM2_LV_UUID=TgjM5V-jhXX-tPTf-gPMa-c3w6-QEjo-7qkmVg LVM2_LV_SIZE=71303168 LVM2_LV_ATTR=-wi-ao---- LVM2_SEGTYPE=linear LVM2_ORIGIN= LVM2_POOL_LV= LVM2_DATA_LV= LVM2_METADATA_LV= LVM2_LV_ROLE=public LVM2_MOVE_PV= LVM2_DATA_PERCENT= LVM2_METADATA_PERCENT= LVM2_COPY_PERCENT= LVM2_LV_TAGS=
LVM2_VG_NAME=rootvg LVM2_LV_NAME=test LVM2_LV_UUID=TgjM5V-jhXX-tPTf-gPMa-c3w6-QEjo-7qkmVg LVM2_LV_SIZE=71303168 LVM2_LV_ATTR=-wi-ao---- LVM2_SEGTYPE=linear LVM2_ORIGIN= LVM2_POOL_LV= LVM2_DATA_LV= LVM2_METADATA_LV= LVM2_LV_ROLE=public LVM2_MOVE_PV= LVM2_DATA_PERCENT= LVM2_METADATA_PERCENT= LVM2_COPY_PERCENT= LVM2_LV_TAGS=
LVM2_VG_NAME=rootvg LVM2_LV_NAME=new_test LVM2_LV_UUID=jMUqLN-DVYU-RUtV-0TnH-Qczq-L820-Bf2G00 LVM2_LV_SIZE=2097152 LVM2_LV_ATTR=-wi-a----- LVM2_SEGTYPE=linear LVM2_ORIGIN= LVM2_POOL_LV= LVM2_DATA_LV= LVM2_METADATA_LV= LVM2_LV_ROLE=public LVM2_MOVE_PV= LVM2_DATA_PERCENT= LVM2_METADATA_PERCENT= LVM2_COPY_PERCENT= LVM2_LV_TAGS=

[vojtechtrefny]

Is the expectation from bd_lvm_lvs() to return those duplicates or shouldn't the lib code filter them out?

We should definitely not return the LV twice. We might add a function to display the segments in the future, but bd_lvm_lvs should filter this out from the lvs output.

Note to myself: This can be reproduced simply by creating two linear LVs and resizing the first one to create the two segments.

[wfpudak]

Thanks @vojtechtrefny. Locally the workaround was to use lv_layout instead of segtype.
segtype_to_lv_lay.diff.zip

[vojtechtrefny]

Unfortunately lv_layout produces different results for non-linear logical volumes, for example thin,sparse instead of thin for thin LVs or thin,pool instead of thin-pool for thin pools so it isn't 1:1 replacement (and we have users that use the strings so we can't change them). I've created #671 as a workaround, that should be good enough for the stable libblockdev, but I think we need some API changes for BDLVMLVdata in 3.0, the segtype attribute doesn't really make sense.

[wfpudak]

Hi @vojtechtrefny ,

Thanks for the explanation we will test your proposed solution.

I noticed some failures on the CI runs for #671. I guess that you plan to merge it but a retrigger/rerun is needed?

Cheers,
/Filip

[vojtechtrefny]

The failed test_lvremove_extra_args case is addressed in #672

[vojtechtrefny]

Closed via #671