zpool provider get_pool_data fails to parse zpool status output on ZFS 2.3.x

[Nourdeo]

Describe the Bug

Hi,
After upgrading ZFS to 2.3.2-2bpo12+2, the zpool provider’s get_pool_data implementation in puppetlabs-zfs_core no longer parses the zpool status output correctly. This breaks ZFS-related resources that rely on get_pool_data ( zpool disk list), while the same code works fine with ZFS 2.2.7-1~bpo12+1

I’ve implemented a small parsing change that works with both ZFS 2.2.7-1bpo12+1 and 2.3.2-2bpo12+2 and would be happy to open a PR if this approach looks acceptable.

---

Environment

• OS: Debian 12 (bookworm)
• Puppet: 7.x
• Module: puppetlabs-zfs_core (version: 1.6.1)
• ZFS:
  • Working: 2.2.7-1~bpo12+1
  • Broken: 2.3.2-2~bpo12+2

---

Problem description

With ZFS 2.3.2-2bpo12+2, the format of zpool status output changed slightly. The current get_pool_data implementation in the zpool provider assumes tab-indented lines and uses this logic:

out = execute("zpool status #{zpool_opts} #{@resource[:pool]}", failonfail: false, combine: false)
zpool_data = out.lines.select { |line| line.index("\t") == 0 }.map { |l| l.strip.split("\s")[0] }
zpool_data.shift
zpool_data

On ZFS 2.3.2-2bpo12+2, the status table uses spaces and a NAME STATE READ WRITE CKSUM header.
On ZFS 2.2.7-1bpo12+1, the same code works correctly with the older zpool status formatting.

Analysis

The core issue is the parsing strategy inside get_pool_data:

It assumes that each row of interest starts with a tab and uses line.index("\t") == 0 to filter lines.

With ZFS 2.3.2-2bpo12+2, the output is still a table but the indentation and formatting changed enough that this condition no longer matches.

Proposed fix

I changed get_pool_data to:

Find the table header line matching NAME STATE READ WRITE CKSUM.

From there, collect only the indented lines (pool and device rows).

Take just the first column as the device/pool name.

def get_pool_data
  # https://docs.oracle.com/cd/E19082-01/817-2271/gbcve/index.html
  # we could also use zpool iostat -v mypool for a (little bit) cleaner output
  zpool_opts = case Facter.value(:kernel)
               # use full device names ("-P") on Linux/ZOL to prevent
               # mismatches between creation and display paths:
               when 'Linux'
                 '-P'
               else
                 ''
               end
  out = execute("zpool status #{zpool_opts} #{@resource[:pool]}", failonfail: false, combine: false)

  lines = out.lines
  # Skip until we reach the NAME/STATE/READ/WRITE/CKSUM header
  lines = lines.drop_while { |l| !(l =~ /^\s*NAME\s+STATE\s+READ\s+WRITE\s+CKSUM/) }

  # Now collect only the indented table rows (NAME, pool name, devices)
  zpool_data = lines
                 .select { |l| l =~ /^\s+/ }              # only indented lines
                 .map    { |l| l.strip.split(/\s+/)[0] }  # first column only

  # First entry is "NAME", remove it so first element is the pool name
  zpool_data.shift
  zpool_data
end

##Backwards compatibility

I’ve tested this against:

ZFS 2.2.7-1bpo12+1: returns the same device list as before (no change in behaviour).

ZFS 22.3.2-2bpo12+2: correctly returns the pool and device list where the previous code returned an empty/incorrect array.

The logic is based purely on the NAME STATE READ WRITE CKSUM header and indentation, which appears stable across both versions.