cancel
Showing results for 
Search instead for 
Did you mean: 

SCSI Reserve and Mount Point

H_Sharma
Level 6

Hello Experts,

Pls help with the 2 points.

1)Difference between SPC-2 SCSI reserve and SCSI persistent reserver. What are these?

2) Lets say we have two mount points in backup selection with multistream enabled in backup selection and in Policy.
   \exchange, \firewall now we exclude this firewall mount point directory in exclude_list.
So what will happen. Job for firewall would start or get failed? How many jobs would run?

 

Thanks,

2 ACCEPTED SOLUTIONS

Accepted Solutions

mph999
Level 6
Employee Accredited

Number 2, you can research yourself, give it a go, and report back what you find.

Number 1 I will help with, again you can also check the guides.

SCSI SPC-2 reserve is quite basic - a reservation is just put on the drive, for example by media server1.  This prevents any other HBA from making a reservation on the drive (yes reservations 'come' from the HBA, so the drive is therefrore reserverd.  The only way for the reservation to be released is for a scsi reservation release to be sent to the drive, from the same HBA, or by powercycling the drive.

This is fine, apart from if say you had a cluster and the machine using the drive fails.  Another machine takes over but cannot release the drive, as this can only be done by the original HBA.

Persistent  is more advanced, basically a reservation key is used, which using various utilities can be read from the drive itself - it looks like this:

dr-media1:/usr/openv/netbackup/logs/bptm # sg_persist --read-reservation -d /dev/sg5
  IBM       ULT3580-TD5       0103
  Peripheral device type: tape
  PR generation=0xb, Reservation follows:
    Key=0x1d00006001e8488
    scope: LU_SCOPE,  type: Exclusive Access

The main difference with this, is that if the key is read, it can be determined 'who' set the reservation.   I'm not sure how, but I am told Engineering are able to determine if NBU 'made' the key.  WHat this means is that if the original server, say media server1 set the reservation, a different media server can release it, providing NBU made the reservation. 

View solution in original post

Marianne
Moderator
Moderator
Partner    VIP    Accredited Certified

Please read through the section in the manual that I have referred to above.

There is a section under this topic:

Recommended use for Enable SCSI reserve property

You will see that SCSI persistent reserve is especially important in your environment where your clustered master server nodes have tape drives attached.
You recently posted when there were issues with tape drives after master server failover.

To know if your tape drives support SCSI persistent reserve, check the NBU HCL:  http://www.symantec.com/docs/TECH76495

View solution in original post

18 REPLIES 18

Marianne
Moderator
Moderator
Partner    VIP    Accredited Certified
1) No disrespect, but have you tried to search Admin Guide I? The place I will look for the answer is in Host Properties section under 'Media Properties ' heading. 2) Test this and tell us. This looks like exam type questions, right?

mph999
Level 6
Employee Accredited

Number 2, you can research yourself, give it a go, and report back what you find.

Number 1 I will help with, again you can also check the guides.

SCSI SPC-2 reserve is quite basic - a reservation is just put on the drive, for example by media server1.  This prevents any other HBA from making a reservation on the drive (yes reservations 'come' from the HBA, so the drive is therefrore reserverd.  The only way for the reservation to be released is for a scsi reservation release to be sent to the drive, from the same HBA, or by powercycling the drive.

This is fine, apart from if say you had a cluster and the machine using the drive fails.  Another machine takes over but cannot release the drive, as this can only be done by the original HBA.

Persistent  is more advanced, basically a reservation key is used, which using various utilities can be read from the drive itself - it looks like this:

dr-media1:/usr/openv/netbackup/logs/bptm # sg_persist --read-reservation -d /dev/sg5
  IBM       ULT3580-TD5       0103
  Peripheral device type: tape
  PR generation=0xb, Reservation follows:
    Key=0x1d00006001e8488
    scope: LU_SCOPE,  type: Exclusive Access

The main difference with this, is that if the key is read, it can be determined 'who' set the reservation.   I'm not sure how, but I am told Engineering are able to determine if NBU 'made' the key.  WHat this means is that if the original server, say media server1 set the reservation, a different media server can release it, providing NBU made the reservation. 

Marianne
Moderator
Moderator
Partner    VIP    Accredited Certified

Please read through the section in the manual that I have referred to above.

There is a section under this topic:

Recommended use for Enable SCSI reserve property

You will see that SCSI persistent reserve is especially important in your environment where your clustered master server nodes have tape drives attached.
You recently posted when there were issues with tape drives after master server failover.

To know if your tape drives support SCSI persistent reserve, check the NBU HCL:  http://www.symantec.com/docs/TECH76495

mph999
Level 6
Employee Accredited

Please note that some VTLs in particular do NOT like Persistent reserve, and there is a very high risk of data loss.

I've also seen issues on Solaris with Persistent reserve.

H_Sharma
Level 6

We have set our media servers master servers as spc-2. Is there anything that needs to be done to change it to SCSI persistance?

I mean if i check the SCSI persistance on properties all is it done or need to make change from somewhere else as well?

Marianne
Moderator
Moderator
Partner    VIP    Accredited Certified
Scsi-2 reservation is enabled by default. Before you change it in Host Properties of each media server, check the HCL to see if your make/model tape drives support Persistent reservation.

mph999
Level 6
Employee Accredited

Yes ....

All devices that see a drive MUST use the same type of reservation (for example, different media servers with SSO, NDMP filers ).  If two different devices share a drive and use different reservation, you will for 100% certain get data loss.

Also, when changing reservation type, run backups and restores to test - as mentioned, sometimes particular devices are not compatible (even when the OS compatability suggests they are).  VTLs in particular can play up, and I had a case where Solaris media servers didn't want to play nicely.

Marianne
Moderator
Moderator
Partner    VIP    Accredited Certified

I so wish that users will post all relevant details of their environment when they start new discussions - unless the question is asked out of curiosity or to prepare for exam and there is no intention to actually apply in production environment....

We know from previous posts that there is a Windows clustered master server in this environment with tape drives attached.
We also know from previous posts that there are 2 Solaris media servers in this environment.
We do not know anything about the actual tape drives as Inquiry string was removed when output was posted in Scan Command .
We know that there are no VTLs in this environment.

So, from what we know, we cannot tell you if you should or should not enable persistent SCSI reserve.

From the documentation, it is clear that persistent SCSI reserve is a good idea in a clustered environment.
Martin says that he has seen problems with persistent SCSI reserve on Solaris, but I have personally implemented the setting at a customer site with shared IBM tape drives, clustered Solaris master/media servers, about 6 other Solaris media servers and 2 Windows media servers.
Because we know nothing about the tape drives, only you can check the HCL for compatibility.

So, in closing - 
Only you can decide based on the information you have.

But I strongly recommend that you first address and solve the issue with 'disappearing' drives before making SCSI reservation changes.

mph999
Level 6
Employee Accredited

If it is clustered and if the drives are attached to machines on both sides of the cluster than persistent reservation should be used.  Why, because if the drives are in use when one side goes down they will be reserved, and the machine that takes over will be unable to break the reservation, meaning the drives will have to be powercycled.

If persistent reservation is used, then a reservation key comes into play, meaning quite simply that NBU knows it's a NBU reservation key and can break the reservation.

Personally, I prefer persistent, because the reservation key can be used for troubleshooting, at least on Linux as the sgutils package can be used to read the key off the drive and therefore in the event of reservation errors we can show it's not us causing the issue.

 

 

H_Sharma
Level 6

Hi Marianne,

Master server is on Window 2008 R2, Media servers are on Solaris 11 and Library is Oracle Storagetek SL150.

I think this is the problem in our environment because every time cluster fails over other node has some issue either 98 error codes. Sometime network errors 13, 41.

But not sure whether our SL 150 supports SPR.

Flipping through Library manual and HCL.

Marianne
Moderator
Moderator
Partner    VIP    Accredited Certified

You need to check tape drive compatibility, not tape library.

You have deleted the Enquiry String output in scan command in one of your other posts, so, we cannot help with checking the HCL.

Oh... I have just checked the HCL - it does not list SCSI Reserve support.
You will need to check with the hardware vendor.
STK can have HP or IBM tape drives.

See this note in the HCL:

3. Symantec recommends the use of SCSI Persistent Reserve (SPR) whenever all the hardware in your device path supports it. Refer to the NetBackup Administrator's Guide for more information on SPR and the types of configurations where SPR may be more effective than the default SPC-2 SCSI reserve and release method.

H_Sharma
Level 6

Hi Marianne,

 

Pls find.

 

Device Name : "Tape3"
Passthru Name: "Tape3"
Volume Header: ""
Port: 3; Bus: 0; Target: 0; LUN: 0
Inquiry : "HP Ultrium 6-SCSI 23DS"
Vendor ID : "HP "
Product ID : "Ultrium 6-SCSI "
Product Rev: "23DS"
Serial Number: "HUxxxxxxx"
WWN : ""
WWN Id Type : 0
Device Identifier: ""
Device Type : SDT_TAPE
NetBackup Drive Type: 16
Removable : Yes
Device Supports: SCSI-6
Flags : 0x0
Reason: 0x0

H_Sharma
Level 6

Hello Experts,

Coming to my second query 

I had taken the backup of four mount points.

/usr
/opt
/var
/home

Excluded /usr and still backups had taken 32 kilobyte data and all other rest mount points have got backed up.
/opt had some 5 gb data.

Now I exlcluded /opt and it got backed up with 32 kilobype.

Dont understand what is happening here? Why it is taking backup when I have already excluded it and there is also no skipping in the job.

I could see one more thing that /usr folder was never backed up in BAR gui. However exclude list is working but it should say something in job details.

Marianne
Moderator
Moderator
Partner    VIP    Accredited Certified
The size of a folder name is 32 Kb.

H_Sharma
Level 6

But why its not skipping anything in the logs and there is no information as well just 32 kb :(

 

 

H_Sharma
Level 6

Hi Marianne,

I found 2 drives.

 

HP - Support with Windows / Linux

1:- Ultrium 6250 LTO6 -
Fibre Channel
HP^^^^^^Ultrium^6-SCSI KMS, SResv, SPR, WORM

2:- Ultrium 6650
LTO6 - Fibre
Channel
HP^^^^^^Ultrium^6-SCSI KMS, SResv,
SPR, WORM

Page no. 188 in HCL.

Is it what we are looking for?

So what do you say is it supported?

Marianne
Moderator
Moderator
Partner    VIP    Accredited Certified

You said that /opt had some 5 gb data.

What do you mean with ...why its not skipping anything ...
It looks to me that EVERYTHING inside /opt is skipped.

Because you have /opt in Backup Selection, it still needs to backup /opt (you told it to in the policy).

So, NBU backs up the folder name (32 Kb) and nothing inside the folder.

If you do not want to backup a folder or filesystem, why put it in Backup Selection in the first place?

H_Sharma
Level 6

Ok got it... Every time it skips a directory it will create an empty directory 32kb. Because i have made the exclusion list.

Actually i could not test it earlier yesterday i got the chance to test it. Thats why i put it in backup selection.