RHEL iSCSI part deux

So as it turns out, the earlier post was not always the best approach to using iSCSI with RHEL. An important omission from the earlier post is that we are using Dell EqualLogic storage, and thus installed the EqualLogic Host Integration Tools (HIT). These tools help integrate iSCSI and multipathing into the server, changing the approach in the process. Below are the steps that we took to get our EqualLogic iSCSI storage online and working:

1. Create a new volume using the administration console of the SAN and ensure that a CHAP user has been created that has access to the volume


2. Install equallogic-host-tools on the server.


3. As root (or using sudo), ensure that ehcmd is started service ehcmd status


4. Edit /etc/iscsi/iscsi.conf by setting node startup as manual (as otherwise, iSCSI will try to establish too many redundant connections):
   
   #***************** # Startup settings #***************** #
   To request that the iscsi initd scripts startup a session set to "automatic". # node.startup = automatic # # To manually startup the session set to "manual". The default is automatic.
   node.startup = manual
   
   

   enable CHAP and setting the CHAP user and password

   
   
   # ************* # CHAP Settings # ************* #
   To enable CHAP authentication set node.session.auth.authmethod # to CHAP. The default is None.
   node.session.auth.authmethod = CHAP
   # To set a CHAP username and password for initiator # authentication by the target(s), uncomment the following lines:
   node.session.auth.username = user
   node.session.auth.password = password
   
   

   ...

   
   

   # To set a discovery session CHAP username and password for the initiator # authentication by the target(s), uncomment the following lines: discovery.sendtargets.auth.username = user discovery.sendtargets.auth.password = password



5. Run eqltune -v and follow any recommendations made by the tool


6. run ehcmcli -d to view the ehcmcli diagnostics information


7. Check to see which subnets EqualLogic host connection manager will manage. # rswcli -L. Exclude any subnets that you do not wish to use with multipath. In our case, we have two interfaces for data on one subnet and two interfaces for iSCSI on two other interfaces: rswcli -E -network 10.52.195.128 -mask 255.255.255.128 Run rswcli -L and ehcmcli -d to confirm that the correct adapters/interfaces are being used


8. Run Discovery:
   # iscsiadm -m discovery -p 10.10.5.50 -t st 10.10.5.50
   
   
   
   
   iqn.2001-05.com.equallogic:0-8a0906-a884a4502-fdefc658d2e4d064-oracle-data 10.10.5.50:3260,1
   
   
   iqn.2001-05.com.equallogic:0-8a0906-a884a4502-fdefc658d2e4d064-oracle-data 10.10.5.50:3260,1
   
   
   iqn.2001-05.com.equallogic:0-8a0906-a884a4502-fdefc658d2e4d064-oracle-data 10.10.5.50:3260,1
   
   
   iqn.2001-05.com.equallogic:0-8a0906-a884a4502-fdefc658d2e4d064-oracle-data 10.10.5.50:3260,1
   
   Multiple identical results will be returned as a result of the intefaces that the EqualLogic daemon creates.


9. Login to the target specifying the interface:
   # iscsiadm -m node -T iqn.2001-05.com.equallogic:0-8a0906-a884a4502-fdefc658d2e4d064-oracle-data -I eql.eth1_0 -l


10. Allow automatic logins using the node.startup property. Recall that this was set to manual in the /etc/iscsi/iscsi.conf file in the steps above, since we don't want to open sessions on all interfaces because the EqualLogic daemon creates multiple interfaces using the same adapters (eql.eth1_0, eql.eth1_1, etc). Instead, we allowed automatic startup on two interfaces, one for each adapter (eql.eth1_0 and eql.eth3_0):
    # iscsiadm –m node –T -I eql.eth1_0 –o update –n node.startup –v automatic
    
    
    # iscsiadm –m node –T -I eql.eth3_0 –o update –n node.startup –v automatic


11. Edit /etc/equallogic/eql.conf [MPIO Autologins] section to specify which adapters to use:
    
    # Autologin configuration [MPIO Autologins]
    
    
    # ehcmd will update any targets configured for autologin so the boot time logins
    # are only made through two iscsiadm ifaces using independent ethernet ports.
    # The ehcmd will create additional iSCSI sessions if necessary at a later point.
    # # If you wish to control which ethernet interfaces are used for autologin,
    # list the desired interfaces below, using a separate line for each.
    # Example: # Port = eth0
    # Port = eth1
    Port = eth1
    Port = eth3




12. Create a physical volume using the /dev/mapper/ name, not the SD device name or one of the derivatives created by the equallogic daemon (eql-0-8a0906-d11b93e09-93720a256664d91d-a, eql-0-8a0906-d11b93e09-93720a256664d91d-oracle-datap1, etc) # pvcreate /dev/mapper/eql-0-8a0906-d11b93e09-93720a256664d91d-oracle-data


13. Create the volume group on the newly created physical volume. # vgcreate SANVolGroup00 /dev/mapper/eql-0-8a0906-d11b93e09-93720a256664d91d-oracle-data


14. Create the logical volume. In our case, we created the volume the entire size of the SAN volume: # lvcreate --extents 100%VG --name ora_data2 SANVolGroup00


15. Format the file system as ext3: # mkfs.ext3 /dev/SANVolGroup00/ora_data2


16. Mount the new lv to a mount point: # mount /dev/SANVolGroup00/ora_data2 /oracle/ora_data2


17. Finally, edit /etc/fstab to mount the logical volume on startup: /dev/SANVolGroup00/ora_data2 /oracle/ora_data2 ext3 _netdev,defaults 0 0 Ensure that the mount option _netdev is used, so the volume does not mount until network interfaces are up

RHEL Mapper stealing Filesystem

Today, we've setup our RHEL server to use iSCSI with CHAP authentication to connect to our iSCSI array volumes. We found this tutorial very helpful, although instead of open-iscsi, we simply used RedHat's iscsi-initiator-utils package (probably the same thing).

One thing to note is that after successfully authenticating to the array via CHAP and partitioning the block (which was /dev/sdd) with LVM2, we could not successfully run the pvcreate /dev/sdd1 command to create a physical volume on which we can build our LVMs. When running pvcreate, we'd get the following:

#pvcreate /dev/sdd1
Can't open /dev/sdd1 exclusively. Mounted filesystem?

Although running 'mount' wouldn't show the drive mapped anywhere. As it turns out, after writing LVM2 to the partition table, /dev/mapper grabs onto the device which effectively "mounts" it within the mapper. In our case, it was listed in the mapper as three devices:

eql-0-8a0906-8a0b93e09-01120a256614d8a6-ora-data01p1
eql-0-8a0906-8a0b93e09-01120a256614d8a6-ora-data01
eql-0-8a0906-8a0b93e09-01120a256614d8a6_a

Note the long name referencing our Equillogic iSCSI SAN. We found two workarounds:

1. Run pvcreate using the /dev/mapper/ name, so:
   #pvcreate /dev/mapper/eql-0-8a0906-8a0b93e09-01120a256614d8a6-ora-data01p1
   This was less desirable, however, because of the long file name
   
2. Remove the device from the /dev/mapper and then run pvcreate:
   
   #dmsetup ls
   #dmsetup remove eql-0-8a0906-8a0b93e09-01120a256614d8a6-ora-data01p1
   #dmsetup remove eql-0-8a0906-8a0b93e09-01120a256614d8a6-ora-data01
   #dmsetup remove eql-0-8a0906-8a0b93e09-01120a256614d8a6_a
   #pvcreate /dev/sdd1
   Physical volume "/dev/sdd1" successfully created
   
   

That solved our problem, allowing us to create the physical volume and then create our LVMs on top of it. Thanks to this forum article for the tip.

Active Directory Port Requirements

Here on the client site, we're in the process of getting our VMWare View 4.6 environment up and working. As one of the required steps, we need to get our vCenter and View Connection Servers on the domain. It so happens that our client runs their own domain (as much as I'd like to run our own), so we've had to coordinate with them for this effort. When trying to join our server to the domain, the servers (which are Windows 2003) are able to query DNS, resolved the FQDN, and identify the Domain Controllers, however we could never get to the point where we're prompted to join our servers. Instead, we receive the following error:

DNS was successfully queried for the service location (SRV) resource record
used to locate an Active Directory Domain Controller for domain local:

The query was for the SRV record for _ldap._tcp.dc._msdcs.local

The following AD DCs were identified by the query:

server1.fqdn
server2.fqdn

Common causes of this error include:

- Host (A) records that map the name of the AD DCs to its IP addresses are
missing or contain incorrect addresses.

- Active Directory Domain Controllers registered in DNS are not connected to
the network or are not running.

We initially thought the error to be related to WINS as TCP port 137 was closed from our VLAN to the client's VLAN that hosts the Domain Controller (yes, even in a 2008 AD Environment, the client still uses WINS). We ensured that TCP 389 was open by using telnet and confirmed that it was. As it turns out, that wasn't the issue. We found that although TCP 389 was allowed, UDP 389 was being blocked. After enabling UDP 389, we were able to query AD fine and join the domain.

This was a bit perplexing to us as I've never known AD to require UDP 389, but perhaps that is because its always been open for me. After a bit of Googling, I learned that LDAP only uses UDP 389 when querying the domain to find the list of domain controllers. After querying DNS and getting a list of possible DCs, the client's netlogon service sends a datagram to each domain controller in the form of a UDP LDAP packet, over UDP 389. All available domain controllers respond with the information for DcGetDCName. The client's netlogon service then caches this information to prevent having to constatly query the DCs. Thanks to Marc Nivens at Experts-Exchange for the tip [link].

So long story short, open TCP AND UDP 389 for LDAP/Active Directory.

Addendum

As noted above, after we had opened UDP 389, we were able to get a credential prompt, however after entering correct credentials, we would still receive some odd errors and could not join the server to the domain. After a bit of Googling, we found this Microsoft Technet article outlines the ports required for 2003 and 2008 domains, most notably the high-level ports.

Lesson Learned

After not eating lunch today and doing some quick configurations withing vCenter, I accidentally deleted a vDisk off of the wrong VM. Not good. Sure enough, when logging into the ESX host, the vDisk was gone, unrecoverable. I recall VMWare ESX 3.5 had a utility named vmfs-undelete for the purpose of recovering deleted vmdk files. No longer:

vmfs-undelete utility is not available for ESX/ESXi 4.0
ESX/ESXi 3.5 Update 3 included a utility called vmfs-undelete, which could be used to recover deleted .vmdk files. This utility is not available with ESX/ESXi 4.0.

Workaround: None. Deleted .vmdk files cannot be recovered.

[VMware ESX 4.0 Update 1 Release Notes]

Fortunately the data on the vDisk can be recreated by the developer in about an hour, so all told we're not set back too far. An important lesson learned though:

1. Always double check before permanently deleting
2. Don't have multiple configuration windows open at the same time
3. Eat lunch

The perils of using Wifi for Mobile Geographic Positioning

It is pretty well known that the Google StreetView cars that capture the images used for Google Street also collect information about Wifi networks in the area, to improve its location based services. In the past, Google has been known to have been collection too much information on private Wifi networks. That aside, it is a pretty cool concept- if your mobile device, such as an ipod touch, lacks GPS capabilities, it can still find where you are based on the Wifi networks around you.

The ingenuity of their system is also their crux. Take for example what I learned today. I recently made the jump to a smartphone, an HTC Evo. I LOVE the phone and 4G, one of the primary reasons I opted for the phone was for its location-based capabilities. This weekend, I am down at the family beach house enjoying a week away from work and getting in some riding/hiking/sleeping time. However I've noticed recently that when in the house, the phone's weather report is mapping Athens, GA as my current location, although I am 250+ miles south. I wrote this inaccuracy off as a glitch in Sprint's towers or something along those lines, that is until I took a closer look at Google Maps on my phone today.

I found it curious that the phone was putting me in Athens, which is where I attended the University of Georgia for four years. Out of curiosity I wanted to see where in Athens it was putting me, as I'm pretty familiar with the city and its streets. After zooming in a bit, I noticed that it was mapping me near the intersection of Baxter and Milledge ave. After zooming in all the way (which is not as far as I've been able to do when GPS is turned on), I realized where my phone was pinpointing me- off of Springdale Street, which is where I lived for two years while at the University. Sure enough, my phone was using Google's massive database of Wifi data (MAC, SSID, etc) to locate me, the same place where it had captured the router's MAC 3-4 years ago. Although the SSID of the router has changed (as well as the encryption key, etc), it is obvious that Google maps is using the MAC (media access control) address of the router as its primary identifier- it makes sense, as the primary purpose of MAC addresses is to be a unique identifier. So although I am not really near Athens at all, I am near the router that was in Athens at the time Google captured its MAC.

This presents somewhat of a quandry using Google maps. I guess Google just didn't count on people moving very often, or they figured they update their database frequently enough to mitigate this problem (note that it has been 3 years since the router as resided at that location, it has since changed geographies twice). It begs the question; is there somewhere I can go to update their database to reflect the router's new location? Also, would this occur no matter where in the world the router was located, so long as there were not any indexed networks nearby? Regardless, it is an illustrative example of how Google Maps works using Wifi data and how this clever system also has its limitations.

Gnome Wallpaper Changer

One feature I really liked about Windows 7 that is lacking on F13 Gnome is the ability to automatically rotate desktop wallpapers. This minor aesthetic functionality is easily implemented using crontab and a simple python script. I set the script to run every 15 minutes using crontab and to read from a specific folder of wallpapers (.backgrounds/dual). I used Davyd's Madeley's python script:

#!/usr/bin/python
#
# change-background.py
#
# A script to change to a random background image
#
# (c) 2004, Davyd Madeley
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 2, or (at your option)
# any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software Foundation,
# Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
#

backgrounds = ".backgrounds/dual"

import gconf
import os
import random
import mimetypes

class GConfClient:
def __init__ (self):
self.__client__ = gconf.client_get_default ()
def get_background (self):
return self.__client__.get_string ("/desktop/gnome/background/picture_filename")
def set_background (self, background):
self.__client__.set_string ("/desktop/gnome/background/picture_filename", background)

client = GConfClient ()


dir_items = os.listdir (os.path.join (os.environ["HOME"], backgrounds))
items = []

for item in dir_items:
mimetype = mimetypes.guess_type (item)[0]
if mimetype and mimetype.split ('/')[0] == "image":
items.append (item)

item = random.randint (0, len (items) - 1)
current_bg = client.get_background ()

while (items[item] == current_bg):
item = random.randint (0, len (items) - 1)

client.set_background (os.path.join (os.environ["HOME"], backgrounds, items[item]))

The crontab file is set as:

*/15 * * * * DISPLAY=:0.0 /usr/local/bin/change-background.py

Citrix VDI

Haven't had much time to post lately as we're deep in our Citrix VDI build at work. Our lead VDI engineer left the firm about the same time our hardware arrived, which means I am not the full time engineer on the project, which I'm sincerely enjoying as it affords me the opportunity to learn and really familiarize myself with our XenDesktop 4, PVS 5.6 SP1, XenApp 6 environment. The downside is I definitely am caught spinning my wheels sometimes simply because I'm not that familiar with the technology. As a result, below is a list of a few things that I've learned thus far that is not documented on Citrix's site that may be useful for others experiencing similar issues:

1. When installing UPM on Windows XP VDIs, change the default installation location to C:\Citrix\ProfMgr. Supposedly, it can handle 8.3 filenames (filenames like C:\Prog~\Citrix\Prof), however we never got it working in our environment, even though 8.3 filenames are enabled.
2. When working on the GoldenImage or vdisk in private mode, make sure you delete all domain user profiles before XenConverting or putting the vdisk in standard mode. Otherwise UPM will complain of log file issues as it can't write to the log file because the log file is read-only, imprinted in the user profile on the vdisk, which is read-only
   
3. XenDesktop 4 does not like MS SQL 2008 SP2. While SQL 2008 is not officially supported, we've had no issues using it until SP2 was applied. Once SP2 was applied, XenDesktop continued to work, but could not reliably determine the power state of the VDIs and the XenDesktop Setup Wizard would always fail on the PVS with an error of "Invalid Host ID" when trying to create the desktop gropu
   
4. Make sure to check the "Manage Machine Account Password" on the VDisk File properties! Otherwise you'll have a nightmare of AD issues and the VDIs won't register using the Desktop Agent to the DDC
5. In a somewhat unique environment, we are required to stream to two separate VLANs. To accomplish this, we teamed two of our physical NICs on our PVS servers and trunked the teamed connection to each VLAN. This alone will not accomplish streaming to both VLANs as both the TFTP service and Provisioning Server Two Stage Boot Process only listen on one NIC by default. If you try opening the tftp.cpl control panel, it only allows you to check one adapter for the service. As for the PVS two-stage boot process control panel (.cpl), we couldn't even get it to open on our 2008 R2 Server. To allow the services to listen on all adapters, go into the registry HKLM\System\Controlset001\services\\parameters, clear the value of the "Adapter" reg key. If you run a netstat -a you'll see the service is listening on 0.0.0.0 *.*
6. Due to a client security requirement, TFTP is disallowed in the environment. As a result, we use boot ISOs to boot to our VDisks. Through trial and error, we discovered that to boot to an empy vdisk for the first time, you must use PXE boot (and subsequently, TFTP). You cannot use a boot iso when first creating a vdisk.

Off the top of my head, that's all I can think of right now. As we continue through our build phase, I will continue to supplement this list.