LVM2-2.02.133

[bdubbs@…]

New point version

[Pierre Labastie]

WHATS_NEW:

Version 2.02.133 - 30th October 2015
====================================
  Support repeated -o|--options for reporting commands.
  Support -o- and -o# for reporting commands to remove and compact fields.
  Fix missing PVs from pvs output if vgremove is run concurrently.
  Remove unwanted error message when running pvs/vgs/lvs and vgremove at once.
  Check newly created VG's metadata do not overlap in metadata ring buffer.
  Check metadata area size is at least the minimum size defined for the format.
  Thin pool targets uses low_water_mark from profile.
  Dropping 'yet' from error of unsupported thick snapshot of snapshots.
  Do not support unpartitioned DASD devices with CDL formatted with pvcreate.
  For thins use flush for suspend only when volume size is reduced.
  Enable code which detects the need of flush during suspend.
  Ensure --use-policy will resize volume to fit below threshold.
  Correct percentage evaluation when checking thin-pool over threshold.
  Fix lvmcache to move PV from VG to orphans if VG is removed and lvmetad used.
  Fix lvmcache to not cache even invalid info about PV which got removed.
  Support checking of memlock daemon counter.
  Allow all log levels to be used with the lvmetad -l option.
  Add optional shutdown when idle support for lvmetad.
  Fix missing in-sync progress info while lvconvert used with lvmpolld.
  Add report/compact_output_cols to lvm.conf to define report cols to compact.
  Do not change logging in lvm2 library when it's already set.
  Check for enough space in thin-pool in command before creating new thin.
  Make libblkid detect all copies of the same signature if use_blkid_wiping=1.
  Fix vgimportclone with -n to not add number unnecessarily to base VG name.
  Cleanup vgimportclone script and remove dependency on awk, grep, cut and tr.
  Add vg_missing_pv_count report field to report number of missing PVs in a VG.
  Properly identify internal LV holding sanlock locks within lv_role field.
  Add metadata_devices and seg_metadata_le_ranges report fields for raid vols.
  Fix lvm2-{activation,clvmd,cmirrord,monitor} service to exec before mounting.

WHATS_NEW_DM:

Version 1.02.110 - 30th October 2015
====================================
  Disable thin monitoring plugin when it fails too often (>10 times).
  Fix/restore parsing of empty field '-' when processing dmeventd event.
  Enhance dm_tree_node_size_changed() to recognize size reduction.
  Support exit on idle for dmenventd (1 hour).
  Add support to allow unmonitor device from plugin itself.
  New design for thread co-operation in dmeventd.
  Dmeventd read device status with 'noflush'.
  Dmeventd closes control device when no device is monitored.
  Thin plugin for dmeventd improved percentage usage.
  Snapshot plugin for dmeventd improved percentage usage.
  Add dm_hold_control_dev to allow holding of control device open.
  Add dm_report_compact_given_fields to remove given empty fields from report.
  Use libdm status parsing and local mem raid dmeventd plugin.
  Use local mem pool and lock only lvm2 execution for mirror dmeventd plugin.
  Lock protect only lvm2 execution for snapshot and thin dmeventd plugin.
  Use local mempool for raid and mirror plugins.
  Reworked thread initialization for dmeventd plugins.
  Dmeventd handles snapshot overflow for now equally as invalid.
  Convert dmeventd to use common logging macro system from libdm.
  Return -ENOMEM when device registration fails instead of 0 (=success).
  Enforce writethrough mode for cleaner policy.
  Add support for recognition and deactivation of MD devices to blkdeactivate.
  Move target status functions out of libdm-deptree.
  Correct use of max_write_behind parameter when generating raid target line.
  Fix dm-event systemd service to make sure it is executed before mounting.

[Pierre Labastie]

Tests again! Testing on an older VM, I got again 10 or so "failed" tests, all of which had something to do with the DM target dm-cache. Reason: configure expects that modprobe can be found in $PATH. If it finds it, it sets MODPROBE_CMD to modprobe, otherwise, it leaves it empty. But since I build as much as possible as user, I have PATH=/usr/local/bin:/bin:/usr/bin:/opt/jdk/bin, and modprobe is not found. However the cache library in lvm2 uses MODPROBE_CMD to check or load the dm-cache module. Since it is empty, it finds that it is impossible to use the cache feature and marks the test as failed, although the test is run as root and now modprobe can be found in the PATH. So I think it should be wise to add path_append /sbin; path_append /usr/sbin before the configure command, maybe saving the PATH variable before.

Note that the cache feature is broken not only for tests but also for the installed package if modprobe is not found during configure.

[Pierre Labastie]

Hmmm. Typo! It is pathappend, not path_append.

[Pierre Labastie]

Test results, without installing thin-provisionning tools, without dmeventd, without lvmetad, running make check_local as root:

### 232 tests: 185 passed, 39 skipped, 0 timed out, 6 warned, 2 failed

I have used the --with-thin-check=, etc switches. /usr/sbin and /sbin in the path during configure. Almost all config switches in the Multiple devices driver support (RAID and LVM) section selected as buit-in.

[Pierre Labastie]

Fixed at r16585