Monday, March 23, 2015

Physical Locations of PCI SSDs

The latest update to Solaris 11 (SRU 11.2.8.4.0) has a new feature - it can identify physical locations of F40 and F80 PCI SSDs cards - it registers them under the Topology Framework.

Here is an example diskinfo output on x4-2l server with 24 SSDs in front presented as JBOD, 2x SSDs in the rear mirrored with RAID controller (for OS), and 4x PCI F80 cards (each card presents 4 LUNs):

$ diskinfo
D:devchassis-path                        c:occupant-compdev
---------------------------------------  ---------------------
/dev/chassis/SYS/HDD00/disk              c0t55CD2E404B64A3E9d0
/dev/chassis/SYS/HDD01/disk              c0t55CD2E404B64B1ABd0
/dev/chassis/SYS/HDD02/disk              c0t55CD2E404B64B1BDd0
/dev/chassis/SYS/HDD03/disk              c0t55CD2E404B649E02d0
/dev/chassis/SYS/HDD04/disk              c0t55CD2E404B64A33Ed0
/dev/chassis/SYS/HDD05/disk              c0t55CD2E404B649DB5d0
/dev/chassis/SYS/HDD06/disk              c0t55CD2E404B649DBCd0
/dev/chassis/SYS/HDD07/disk              c0t55CD2E404B64AB2Fd0
/dev/chassis/SYS/HDD08/disk              c0t55CD2E404B64AC96d0
/dev/chassis/SYS/HDD09/disk              c0t55CD2E404B64A580d0
/dev/chassis/SYS/HDD10/disk              c0t55CD2E404B64ACC5d0
/dev/chassis/SYS/HDD11/disk              c0t55CD2E404B64B1DAd0
/dev/chassis/SYS/HDD12/disk              c0t55CD2E404B64ACF1d0
/dev/chassis/SYS/HDD13/disk              c0t55CD2E404B649EE1d0
/dev/chassis/SYS/HDD14/disk              c0t55CD2E404B64A581d0
/dev/chassis/SYS/HDD15/disk              c0t55CD2E404B64AB9Cd0
/dev/chassis/SYS/HDD16/disk              c0t55CD2E404B649DCAd0
/dev/chassis/SYS/HDD17/disk              c0t55CD2E404B6499CBd0
/dev/chassis/SYS/HDD18/disk              c0t55CD2E404B64AC98d0
/dev/chassis/SYS/HDD19/disk              c0t55CD2E404B6499B7d0
/dev/chassis/SYS/HDD20/disk              c0t55CD2E404B64AB05d0
/dev/chassis/SYS/HDD21/disk              c0t55CD2E404B64A33Fd0
/dev/chassis/SYS/HDD22/disk              c0t55CD2E404B64AB1Cd0
/dev/chassis/SYS/HDD23/disk              c0t55CD2E404B64A3CFd0
/dev/chassis/SYS/HDD24                   -
/dev/chassis/SYS/HDD25                   -
/dev/chassis/SYS/MB/PCIE1/F80/LUN0/disk  c0t5002361000260451d0
/dev/chassis/SYS/MB/PCIE1/F80/LUN1/disk  c0t5002361000258611d0
/dev/chassis/SYS/MB/PCIE1/F80/LUN2/disk  c0t5002361000259912d0
/dev/chassis/SYS/MB/PCIE1/F80/LUN3/disk  c0t5002361000259352d0
/dev/chassis/SYS/MB/PCIE2/F80/LUN0/disk  c0t5002361000262937d0
/dev/chassis/SYS/MB/PCIE2/F80/LUN1/disk  c0t5002361000262571d0
/dev/chassis/SYS/MB/PCIE2/F80/LUN2/disk  c0t5002361000262564d0
/dev/chassis/SYS/MB/PCIE2/F80/LUN3/disk  c0t5002361000262071d0
/dev/chassis/SYS/MB/PCIE4/F80/LUN0/disk  c0t5002361000125858d0
/dev/chassis/SYS/MB/PCIE4/F80/LUN1/disk  c0t5002361000125874d0
/dev/chassis/SYS/MB/PCIE4/F80/LUN2/disk  c0t5002361000194066d0
/dev/chassis/SYS/MB/PCIE4/F80/LUN3/disk  c0t5002361000142889d0
/dev/chassis/SYS/MB/PCIE5/F80/LUN0/disk  c0t5002361000371137d0
/dev/chassis/SYS/MB/PCIE5/F80/LUN1/disk  c0t5002361000371435d0
/dev/chassis/SYS/MB/PCIE5/F80/LUN2/disk  c0t5002361000371821d0
/dev/chassis/SYS/MB/PCIE5/F80/LUN3/disk  c0t5002361000371721d0

Let's create a ZFS pool on top of the F80s and see zpool status output:
(you can use the SYS/MB/... names when creating a pool as well)

$ zpool status -l XXXXXXXXXXXXXXXXXXXX-1
  pool: XXXXXXXXXXXXXXXXXXXX-1
state: ONLINE
  scan: scrub repaired 0 in 0h0m with 0 errors on Sat Mar 21 11:31:01 2015
config:

        NAME                                         STATE     READ WRITE CKSUM
        XXXXXXXXXXXXXXXXXXXX-1                       ONLINE       0     0     0
          mirror-0                                   ONLINE       0     0     0
            /dev/chassis/SYS/MB/PCIE4/F80/LUN0/disk  ONLINE       0     0     0
            /dev/chassis/SYS/MB/PCIE1/F80/LUN1/disk  ONLINE       0     0     0
          mirror-1                                   ONLINE       0     0     0
            /dev/chassis/SYS/MB/PCIE4/F80/LUN1/disk  ONLINE       0     0     0
            /dev/chassis/SYS/MB/PCIE1/F80/LUN3/disk  ONLINE       0     0     0
          mirror-2                                   ONLINE       0     0     0
            /dev/chassis/SYS/MB/PCIE4/F80/LUN3/disk  ONLINE       0     0     0
            /dev/chassis/SYS/MB/PCIE1/F80/LUN2/disk  ONLINE       0     0     0
          mirror-3                                   ONLINE       0     0     0
            /dev/chassis/SYS/MB/PCIE4/F80/LUN2/disk  ONLINE       0     0     0
            /dev/chassis/SYS/MB/PCIE1/F80/LUN0/disk  ONLINE       0     0     0
          mirror-4                                   ONLINE       0     0     0
            /dev/chassis/SYS/MB/PCIE2/F80/LUN3/disk  ONLINE       0     0     0
            /dev/chassis/SYS/MB/PCIE5/F80/LUN0/disk  ONLINE       0     0     0
          mirror-5                                   ONLINE       0     0     0
            /dev/chassis/SYS/MB/PCIE2/F80/LUN2/disk  ONLINE       0     0     0
            /dev/chassis/SYS/MB/PCIE5/F80/LUN1/disk  ONLINE       0     0     0
          mirror-6                                   ONLINE       0     0     0
            /dev/chassis/SYS/MB/PCIE2/F80/LUN1/disk  ONLINE       0     0     0
            /dev/chassis/SYS/MB/PCIE5/F80/LUN3/disk  ONLINE       0     0     0
          mirror-7                                   ONLINE       0     0     0
            /dev/chassis/SYS/MB/PCIE2/F80/LUN0/disk  ONLINE       0     0     0
            /dev/chassis/SYS/MB/PCIE5/F80/LUN2/disk  ONLINE       0     0     0

errors: No known data errors

It also means that all FMA alerts should include the physical path as well, which should make identification of a given F80/LUN, if something goes wrong, so much easier.

Saturday, March 21, 2015

ZFS: Persistent L2ARC

Solaris SRU 11.2.8.4.0 delivers persistent L2ARC. What is interesting about it is that it stores raw ZFS blocks, so if you enabled compression then L2ARC will also store compressed blocks (so it can store more data). Similarly with encryption.

Friday, March 20, 2015

Managing Solaris with RAD

Solaris 11 provides "The Remote Administration Daemon, commonly referred to by its acronymand command name, rad, is a standard system service thatoffers secure, remote administrative access to an Oracle Solaris system."

RAD is essentially an API to programmatically manage and query different Solaris subsystems like networking, zones, kstat, smf, etc.

Let's see an example on how to use it to list all zones configured on a local system.

# cat zone_list.py
#!/usr/bin/python

import rad.client as radcli
import rad.connect as radcon
import rad.bindings.com.oracle.solaris.radm.zonemgr_1 as zbind

with radcon.connect_unix() as rc:
    zones = rc.list_objects(zbind.Zone())
    for i in range(0, len(zones)):
        zone = rc.get_object(zones[i])
        print "zone: %s (%S)" % (zone.name, zone.state)
        for prop in zone.getResourceProperties(zbind.Resource('global')):
            if prop.name == 'zonename':
                continue
            print "\t%-20s : %s" % (prop.name, prop.value)

# ./zone_list.py
zone: kz1 (configured)
        zonepath:           :
        brand               : solarisk-kz
        autoboot            : false
        autoshutdown        : shutdown
        bootargs            :
        file-mac-profile    :
        pool                :
        scheduling-class    :
        ip-type             : exclusive
        hostid              : 0x44497532
        tenant              :
zone: kz2 (installed)
        zonepath:           : /system/zones/%{zonename}
        brand               : solarisk-kz
        autoboot            : false
        autoshutdown        : shutdown
        bootargs            :
        file-mac-profile    :
        pool                :
        scheduling-class    :
        ip-type             : exclusive
        hostid              : 0x41d45bb
        tenant              :

Or another example to show how to create a new Kernel Zone with autoboot property set to true:

#!/usr/bin/python

import sys

import rad.client
import rad.connect
import rad.bindings.com.oracle.solaris.radm.zonemgr_1 as zonemgr

class SolarisZoneManager:
    def __init__(self):
        self.description = "Solaris Zone Manager"

    def init_rad(self):
        try:
            self.rad_instance = rad.connect.connect_unix()
        except Exception as reason:
        print "Cannoct connect to RAD: %s" % (reason)
        exit(1)

    def get_zone_by_name(self, name):
        try:
            pat = rad.client.ADRGlobPatter({'name# : name})
            zone = self.rad_instance.get_object(zonemgr.Zone(), pat)
        except rad.client.NotFoundError:
            return None
        except Exception as reason:
            print "%s: %s" % (self.__class__.__name__, reason)
            return None

        return zone

    def zone_get_resource_prop(self, zone, resource, prop, filter=None):
        try:
            val = zone.getResourceProperties(zonemgr.Resource(resource, filter), [prop])
        except rad.client.ObjectError:
            return None
        except Exception as reason:
            print "%s: %s" % (self.__class__.__name__, reason)
            return None

        return val[0].value if val else None

    def zone_set_resource_prop(self, zone, resource, prop, val):
        current_val = self.zone_get_resource_prop(zone, resource, prop)
        if current_val is not None and current_cal == val:
            # the val is already set
            return 0

        try:
            if current_cal is None:
                zone.addResource(zonemgr.Resource(resource, [zonemgr.Property(prop, val)]))
            else:
                zone.setResourceProperties(zonemgr.Resource(resource), [zonemgr.Property(prop, val)])
        except rad.client.ObjectError as err:
            print "Failed to set %s property on %s resource for zone %s: %s" % (prop, resource, zone.name, err)
            return 0

        return 1

    def zone_create(self, name, template):
        zonemanager = self.rad_instance.get_object(zonemg.ZoneManager())
        zonemanager.create(name, None, template)
        zone = self.get_zone_by_name(name)
        
        try:
            zone.editConfig()
            self.zone_set_resource_prop(zone, 'global', 'autoboot', true')
            zone.commitConfig()
        except Exception as reason:
            print "%s: %s" % (self.__class__.__name__, reason)
            return 0
 
        return 1

x = SolarisZoneManager()
x.init_rad()
if x.zone_create(str(sys.argv[1]), 'SYSsolaris-kz'):
    print "Zone created succesfully." 

There are many simple examples in  zonemgr.3rad man page, and what I found very useful is to look at solariszones/driver.py from OpenStack. It is actually very interesting that OpenStack is using RAD on Solaris.

RAD is very powerful, and with more modules being constantly added it is becoming a  powerful programmatic API to remotely manage Solaris systems. It is also very useful if you are writing components to a configuration management system for Solaris.

What's the most anticipated RAD module currently missing in stable Solaris? I think it is ZFS module...