personal_infra/ansible/roles/healthcheck/templates/checks/zfs-health.sh.j2

    # Five conditions, all of which have to hold. Ported from the check that
    # infra/nodito/32_zfs_pool_setup_playbook.yml deployed, which was correct -
    # only its reporting was tied to Uptime Kuma.
    local pool="{{ healthcheck_zfs_pool }}"
    local json issues=""

    json=$(zpool status -j "$pool" 2>&1) || { MESSAGE="zpool status failed: ${json}"; return 1; }

    # 1. pool state
    local state
    state=$(echo "$json" | jq -r --arg p "$pool" '.pools[$p].state')
    [ "$state" = "ONLINE" ] || issues="${issues}${issues:+; }pool ${state}"

    # 2. every vdev and device ONLINE
    local bad
    bad=$(echo "$json" | jq -r --arg p "$pool" '
        .pools[$p].vdevs[] | .. | objects
        | select(.state? and .state != "ONLINE")
        | "\(.name // "unknown"):\(.state)"' 2>/dev/null | paste -sd, -)
    [ -z "$bad" ] || issues="${issues}${issues:+; }devices ${bad}"

    # 3. resilver in progress
    local fn st
    fn=$(echo "$json" | jq -r --arg p "$pool" '.pools[$p].scan_stats.function // "NONE"')
    st=$(echo "$json" | jq -r --arg p "$pool" '.pools[$p].scan_stats.state // "NONE"')
    if [ "$fn" = "RESILVER" ] && [ "$st" = "SCANNING" ]; then
        issues="${issues}${issues:+; }resilvering"
    fi

    # 4. read/write/checksum errors. ZFS reports these as strings.
    local errs
    errs=$(echo "$json" | jq -r --arg p "$pool" '
        .pools[$p].vdevs[] | .. | objects
        | select(.name? and ((.read_errors // "0" | tonumber) > 0
              or (.write_errors // "0" | tonumber) > 0
              or (.checksum_errors // "0" | tonumber) > 0))
        | "\(.name) r=\(.read_errors) w=\(.write_errors) c=\(.checksum_errors)"' 2>/dev/null | paste -sd, -)
    [ -z "$errs" ] || issues="${issues}${issues:+; }errors ${errs}"

    # 5. errors from the last scrub
    local scan_err
    scan_err=$(echo "$json" | jq -r --arg p "$pool" '.pools[$p].scan_stats.errors // "0"')
    if [ -n "$scan_err" ] && [ "$scan_err" != "0" ] && [ "$scan_err" != "null" ]; then
        issues="${issues}${issues:+; }scan errors ${scan_err}"
    fi

    # 6. Capacity. Not an error condition in `zpool status` - a 95% full pool is
    #    reported perfectly ONLINE - so it has to be read separately, and it is
    #    the failure you get warning of rather than the one you discover.
    local capacity
    capacity=$(zpool list -H -o capacity "$pool" 2>/dev/null | tr -dc '0-9')
    if [ -z "$capacity" ]; then
        issues="${issues}${issues:+; }cannot read capacity"
    elif [ "$capacity" -ge {{ healthcheck_zfs_capacity_threshold }} ]; then
        issues="${issues}${issues:+; }pool ${capacity}% full (>={{ healthcheck_zfs_capacity_threshold }}%)"
    fi

    if [ -n "$issues" ]; then MESSAGE="$issues"; return 1; fi

    local scrub
    scrub=$(echo "$json" | jq -r --arg p "$pool" '.pools[$p].scan_stats.start_time // "never"')
    MESSAGE="${pool} ONLINE, ${capacity}% full, last scrub ${scrub}"
    return 0