toasters

toasters@lists.teaparty.net

13514 discussions

RE: Locking of Files and server clusters
by Basiji, Maryam 25 May '01

25 May '01

Just saw Lance's response. Lance elaborates on how to implement the solution that is discussed in the paper I just sent over. Thanks, Maryam -----Original Message----- From: White, Lance [mailto:Lance.White@netapp.com] Sent: Friday, May 25, 2001 8:16 AM To: 'Morgan, Joshua'; 'toasters(a)mathworks.com' Subject: RE: Locking of Files and server clusters One of the steps required to ensure proper failover of SQL Server via Legato Cluster is the creation of a Resource Group (RG)connecting the application to the two nodes. The RG must include a script that gets executed as one of the first steps in bringing up the RG. The script was designed to be a workaround for the filer's normally necessary CIFS file-locking behavior. When an Legato Cluster node first fails, as far as the filer is concerned, the session between the filer and that node is still alive (the disconnect could be temporary). Because of this, it initially maintains the file locks originally established for that session. Without our script, these locks would prevent the other "takeover" node from accessing the required files during takeover. However, when our script is implemented, the file locks are terminated and the "takeover" node can access the files as necessary. In our case, the files the "takeover" node needs access to are the SQL Server database data and! log files. If the "takeover" node does not have access to these files, once the takeover is completed and the SQL Server is done recovering, the databases on the filer are declared "suspect". Following is the generic syntax for the script that must reside on each node: rsh filer_name -l cluster_admin_name cifs terminate -t 0 failed_node_name You can pass the name of the failed node via a &ft' environment variable (refer to the LCE docs for details), or you can create a nested if...then...else loop to determine which node you are starting up on, and how to terminate the CIFS session on the Filer. An example would be: if &ft'currentnode == 'failednode' then rsh Filer -l ClusterAdmin cifs terminate -t 0 failednode else rsh Filer -l ClusterAdmin cifs terminate -t 0 othernode end; Likely will be a little more coding than this, but you get the drift. Lance White Systems Engineer Network Appliance -----Original Message----- From: Morgan, Joshua [mailto:JMorgan@Trivergent.com] Sent: Friday, May 25, 2001 7:29 AM To: 'toasters(a)mathworks.com' Subject: Locking of Files and server clusters Here is the Scenario: We have SQL7 Server running with its Databases on an F760 We are using Legato Clustering Software If the main SQL Server dies and the cluster software tries to bring up the backup server, we get a message saying that the Database files are still locked by the downed server. Does anyone know of a way to script unlocking the files? We can manually do it through server manager but that is no good for Disater Recovery. Joshua Morgan PH: (864) 331-7939 jmorgan(a)trivergent.com

1 0

RE: Locking of Files and server clusters
by Basiji, Maryam 25 May '01

25 May '01

Josh, Take a look at: http://www.netapp.com/tech_library/3084.html <http://www.netapp.com/tech_library/3084.html> Similar issue and I believe Legato Clustering can use this method as well. Thanks, Maryam -----Original Message----- From: Morgan, Joshua [mailto:JMorgan@Trivergent.com] Sent: Friday, May 25, 2001 7:29 AM To: 'toasters(a)mathworks.com' Subject: Locking of Files and server clusters Here is the Scenario: We have SQL7 Server running with its Databases on an F760 We are using Legato Clustering Software If the main SQL Server dies and the cluster software tries to bring up the backup server, we get a message saying that the Database files are still locked by the downed server. Does anyone know of a way to script unlocking the files? We can manually do it through server manager but that is no good for Disater Recovery. Joshua Morgan PH: (864) 331-7939 jmorgan(a)trivergent.com

1 0

RE: Why is root not root?
by Yoder, Alan 25 May '01

25 May '01

> How about letting root delete files that are currently > locked open on the NT-side like they can with > NFS-locked files? This is burt 8500. It is fixed in 6.1; in a unix qtree you now get this behavior which is characteristic of Unix semantics. In mixed and ntfs qtrees, NTFS locking semantics still apply. Alan =============================================================== Alan G. Yoder agy(a)netapp.com ===============================================================

2 1

RE: incremental ndmpcopy doesn't work
by Knight, Graham 25 May '01

25 May '01

I think the key to the problem might be the line that says: >DstFiler: mei 25 09:51:02 [ndmpd:0]: Cannot run an incremental >backup because the previous restore did not complete. It looks like the level 0 might not have completed correctly. Do you have the logs from the level 0 dump? Graham -----Original Message----- From: Reinoud Reynders [mailto:Reinoud.Reynders@uz.kuleuven.ac.be] Sent: Friday, May 25, 2001 6:02 AM To: tmac(a)netapp.com Cc: toasters(a)mathworks.com Subject: Re: incremental ndmpcopy doesn't work Unfortunately, all the three checks are ok. (6.1R1; no other dumps and jnmpcopy) Reinoud ----- Original Message ----- From: "Timothy A. McCarthy" <tmac(a)netapp.com> To: "Reinoud Reynders" <Reinoud.Reynders(a)uz.kuleuven.ac.be> Cc: <toasters(a)mathworks.com> Sent: Friday, May 25, 2001 12:09 PM Subject: Re: incremental ndmpcopy doesn't work > Things to check: > > 1. Make sure you are running 5.3.7R3 or later on both system to avoid > ndmp bugs. > > 2. After you run the "-level 0" are you running any other dumps? If so > it interferes > with the information and may cause a failure. > > 3. The "-level i" only works with jnmpcopy. > > **notice the space between the word level and the number or "i" > > --tmac > > Reinoud Reynders wrote: > > > Hello, I try to copy 150 GB from F740:vol0 to F820:vol1. The level0 > > frma ndmpcopy works fine. No errors. When i start the level 1, i get > > the following error: DstFiler: mei 25 09:51:02 [ndmpd:0]: RESTORE > > IS ABORTED mei 25 09:51:02 [ndmpd:0]: DstFiler just halted. > > mei 25 09:51:02 [ndmpd:0]: Reason: ndmp_data_halted_internal_error : > > Dump aborted I have also used the level i, but the same result: 1 > > time ok, incremental no go.The filers run both 6.1R1 Have someone a > > good idea? Reinoud ReyndersUZ KuleuvenHerestraat 49B- 3000 Leuven > > Belgium The complete log-file: S:\Informatiesystemen\nas>ndmpcopy > > toaster1:/vol/vol0/groepen/WebTeam mixer1:/vol/vol1/groepen/WebTeam > > -sa root:XXX -da root:XXX -level 1setting everything to version 3 > > mei 25 09:45:38 [ndmpd:0]: Now waiting for source and destination to > > finish. > > SrcFiler: mei 25 09:45:39 [ndmpd:0]: DUMP: > > SrcFiler: mei 25 09:45:39 [ndmpd:0]: creating > > "/vol/vol0/../snapshot_for_backup. > > 19" snapshot. SrcFiler: mei 25 09:46:01 [ndmpd:0]: DUMP: > > SrcFiler: mei 25 09:46:01 [ndmpd:0]: Using subtree dump SrcFiler: > > mei 25 09:46:49 [ndmpd:0]: DUMP: > > SrcFiler: mei 25 09:46:49 [ndmpd:0]: Date of this level 1 dump: Fri > > May 25 09:44 > > :58 2001. SrcFiler: mei 25 09:46:49 [ndmpd:0]: DUMP: > > SrcFiler: mei 25 09:46:49 [ndmpd:0]: Date of last level 0 dump: Fri > > May 25 09:35 > > :38 2001. SrcFiler: mei 25 09:46:49 [ndmpd:0]: DUMP: > > SrcFiler: mei 25 09:46:49 [ndmpd:0]: Dumping > > /vol/vol0/groepen/WebTeam to NDMP c > > onnection SrcFiler: mei 25 09:46:49 [ndmpd:0]: DUMP: > > SrcFiler: mei 25 09:46:49 [ndmpd:0]: mapping (Pass I)[regular > > files] SrcFiler: mei 25 09:46:49 [ndmpd:0]: DUMP: > > SrcFiler: mei 25 09:46:49 [ndmpd:0]: mapping (Pass > > II)[directories] SrcFiler: mei 25 09:50:23 [ndmpd:0]: DUMP: > > SrcFiler: mei 25 09:50:23 [ndmpd:0]: estimated 12 KB. SrcFiler: mei > > 25 09:50:23 [ndmpd:0]: DUMP: > > SrcFiler: mei 25 09:50:23 [ndmpd:0]: dumping (Pass III) > > [directories] DstFiler: mei 25 09:50:45 [ndmpd:0]: RESTORE: > > DstFiler: mei 25 09:50:45 [ndmpd:0]: Fri May 25 09:50:05 2001: Begin > > incremental > > restore SrcFiler: mei 25 09:50:59 [ndmpd:0]: DUMP: > > SrcFiler: mei 25 09:50:59 [ndmpd:0]: dumping (Pass IV) [regular > > files] DstFiler: mei 25 09:51:02 [ndmpd:0]: RESTORE: > > DstFiler: mei 25 09:51:02 [ndmpd:0]: Cannot run an incremental > > backup because th > > e previous restore did not complete. DstFiler: mei 25 09:51:02 > > [ndmpd:0]: RESTORE: > > DstFiler: mei 25 09:51:02 [ndmpd:0]: RESTORE IS ABORTED mei 25 > > 09:51:02 [ndmpd:0]: DstFiler just halted. > > mei 25 09:51:02 [ndmpd:0]: Reason: ndmp_data_halted_internal_error : > > Dump aborte > > d > > mei 25 09:51:02 [ndmpd:0]: jndmpcopy finished > > -- > ******All New Numbers!!!****** > ************* ************* > > Timothy A. McCarthy --> System Engineer, Eastern Region > Network Appliance http://www.netapp.com > 240-268-2034 Office \ / Page Me at: > 240-268-2001 Fax \/ 888-971-4468 > > >

1 0

RE: Why is root not root?
by Mike Sphar 25 May '01

25 May '01

Speaking of that, I ran into some interesting behavior the other day. I'd had a recurrent problem where our build scripts couldn't delete a directory from the previous build because some windows user somewhere had left a window open on that share. Well, I eventually discovered that by making the share read-only at the share level for everyone but our build engineers, that if I opened a window in that share, and then deleted it from UNIX, the delete would work and in fact my open window would automatically close. However, we later found that the behavior was slightly different if the windows client was viewing the share through an Explorer window (i.e. using the double-paned File Explorer interface rather than just an open window. Anyway, we found that when the share was accessed via Explorer, then when the rm command was executed from UNIX, the command would return successfully with no error, but the directory still remained. If I then closed the Explorer window, the directory would disappear. This was actually worse behavior from the build engineers perspective because at least before their scripts would get a permission denied error and exit, rather than continuing on assuming their work area was empty. I suspect this is for the most part standard CIFS behavior, but it's still a little bizarre to see the rules applied differently depending on which tool was being used on the client. Having said all that, yes I too would like an option whereby files or directories could be deleted no matter what their locking state was. This problem has been an unending source of complaints from our build engineers who ask questions like "Can we go back to the server where this wasn't a problem?" That server being an old Sparc running NFS and Samba. Obviously I don't want to go back to that old beast. -----Original Message---- From: Bruce Sterling Woodcock [mailto:sirbruce@ix.netcom.com] Sent: Friday, May 25, 2001 4:25 AM To: Muhlestein, Mark; White, Lance; 'Chuck Ross'; Net App List (E-mail) Subject: Re: Why is root not root? How about letting root delete files that are currently locked open on the NT-side like they can with NFS-locked files? Bruce

1 0

RE: Locking of Files and server clusters
by White, Lance 25 May '01

25 May '01

Come to think of it, so long as you exit with a value of 0, you shouldn't need the if...then statements. Just fire off an rsh for both nodes to terminate the sessions from the host. One will work, the other won't, but it shouldn't matter. i.e. rsh Filer -l ClusterAdmin cifs terminate -t 0 node1; rsh Filer -l ClusterAdmin cifs terminate -t 0 node2; exit 0; Again, you MIGHT want to wrap some error checking logic around this to feedback results, etc. Lance -----Original Message----- From: White, Lance [mailto:Lance.White@netapp.com] Sent: Friday, May 25, 2001 8:16 AM To: 'Morgan, Joshua'; 'toasters(a)mathworks.com' Subject: RE: Locking of Files and server clusters One of the steps required to ensure proper failover of SQL Server via Legato Cluster is the creation of a Resource Group (RG)connecting the application to the two nodes. The RG must include a script that gets executed as one of the first steps in bringing up the RG. The script was designed to be a workaround for the filer's normally necessary CIFS file-locking behavior. When an Legato Cluster node first fails, as far as the filer is concerned, the session between the filer and that node is still alive (the disconnect could be temporary). Because of this, it initially maintains the file locks originally established for that session. Without our script, these locks would prevent the other "takeover" node from accessing the required files during takeover. However, when our script is implemented, the file locks are terminated and the "takeover" node can access the files as necessary. In our case, the files the "takeover" node needs access to are the SQL Server database data and! log files. If the "takeover" node does not have access to these files, once the takeover is completed and the SQL Server is done recovering, the databases on the filer are declared "suspect". Following is the generic syntax for the script that must reside on each node: rsh filer_name -l cluster_admin_name cifs terminate -t 0 failed_node_name You can pass the name of the failed node via a &ft' environment variable (refer to the LCE docs for details), or you can create a nested if...then...else loop to determine which node you are starting up on, and how to terminate the CIFS session on the Filer. An example would be: if &ft'currentnode == 'failednode' then rsh Filer -l ClusterAdmin cifs terminate -t 0 failednode else rsh Filer -l ClusterAdmin cifs terminate -t 0 othernode end; Likely will be a little more coding than this, but you get the drift. Lance White Systems Engineer Network Appliance -----Original Message----- From: Morgan, Joshua [mailto:JMorgan@Trivergent.com] Sent: Friday, May 25, 2001 7:29 AM To: 'toasters(a)mathworks.com' Subject: Locking of Files and server clusters Here is the Scenario: We have SQL7 Server running with its Databases on an F760 We are using Legato Clustering Software If the main SQL Server dies and the cluster software tries to bring up the backup server, we get a message saying that the Database files are still locked by the downed server. Does anyone know of a way to script unlocking the files? We can manually do it through server manager but that is no good for Disater Recovery. Joshua Morgan PH: (864) 331-7939 jmorgan(a)trivergent.com

1 0

RE: Locking of Files and server clusters
by White, Lance 25 May '01

25 May '01

One of the steps required to ensure proper failover of SQL Server via Legato Cluster is the creation of a Resource Group (RG)connecting the application to the two nodes. The RG must include a script that gets executed as one of the first steps in bringing up the RG. The script was designed to be a workaround for the filer's normally necessary CIFS file-locking behavior. When an Legato Cluster node first fails, as far as the filer is concerned, the session between the filer and that node is still alive (the disconnect could be temporary). Because of this, it initially maintains the file locks originally established for that session. Without our script, these locks would prevent the other "takeover" node from accessing the required files during takeover. However, when our script is implemented, the file locks are terminated and the "takeover" node can access the files as necessary. In our case, the files the "takeover" node needs access to are the SQL Server database data and! log files. If the "takeover" node does not have access to these files, once the takeover is completed and the SQL Server is done recovering, the databases on the filer are declared "suspect". Following is the generic syntax for the script that must reside on each node: rsh filer_name -l cluster_admin_name cifs terminate -t 0 failed_node_name You can pass the name of the failed node via a &ft' environment variable (refer to the LCE docs for details), or you can create a nested if...then...else loop to determine which node you are starting up on, and how to terminate the CIFS session on the Filer. An example would be: if &ft'currentnode == 'failednode' then rsh Filer -l ClusterAdmin cifs terminate -t 0 failednode else rsh Filer -l ClusterAdmin cifs terminate -t 0 othernode end; Likely will be a little more coding than this, but you get the drift. Lance White Systems Engineer Network Appliance -----Original Message----- From: Morgan, Joshua [mailto:JMorgan@Trivergent.com] Sent: Friday, May 25, 2001 7:29 AM To: 'toasters(a)mathworks.com' Subject: Locking of Files and server clusters Here is the Scenario: We have SQL7 Server running with its Databases on an F760 We are using Legato Clustering Software If the main SQL Server dies and the cluster software tries to bring up the backup server, we get a message saying that the Database files are still locked by the downed server. Does anyone know of a way to script unlocking the files? We can manually do it through server manager but that is no good for Disater Recovery. Joshua Morgan PH: (864) 331-7939 jmorgan(a)trivergent.com

1 0

How to configure?
by Jordan Share 25 May '01

25 May '01

So, we are looking at getting a 720 (possibly a used 740) in the near future. How should I configure it wrt number of volumes and so forth? Is there a FAQ somewhere where I can read about the theory behind volume configuration? Basically, I've seen posts that make me think it might be nice to have one volume just for the root filesystem, and another for the actual data we'll be using. What are the restrictions on volumes and so forth? Can I have a 2 volumes in a RAID set, or no? I'd just like a pointer towards a FAQ, when you come down to it, I realize these questions are quite vague. :) Thanks, Jordan

2 1

Locking of Files and server clusters
by Morgan, Joshua 25 May '01

25 May '01

Here is the Scenario: We have SQL7 Server running with its Databases on an F760 We are using Legato Clustering Software If the main SQL Server dies and the cluster software tries to bring up the backup server, we get a message saying that the Database files are still locked by the downed server. Does anyone know of a way to script unlocking the files? We can manually do it through server manager but that is no good for Disater Recovery. Joshua Morgan PH: (864) 331-7939 jmorgan(a)trivergent.com

1 0

RE: Why is root not root?
by Chuck Ross 25 May '01

25 May '01

Yep, set the ignore_acl option and it works great! Thanks guys. Chuck Ross > -----Original Message----- > From: White, Lance [mailto:Lance.White@netapp.com] > Sent: Thursday, May 24, 2001 5:59 PM > To: Knight, Graham; Chuck Ross; Net App List (E-mail) > Subject: RE: Why is root not root? > > > Ah yes, I remember using this one. Nice memory, Graham! > Hopefully that's the right answer for you, Chuck. Let us know! > > LW > > -----Original Message----- > From: Knight, Graham [mailto:Graham.Knight@netapp.com] > Sent: Thursday, May 24, 2001 2:49 PM > To: 'Chuck Ross'; Net App List (E-mail) > Subject: RE: Why is root not root? > > > > Probably because the file has an NT ACL that has been set so > tightly that > the root user cannot access it without taking ownership > first, which cannot > be done very easily from the command line. > > Setting this (hidden) option will make this type of problem a > thing of the past: > > options cifs.nfs_root_ignore_acl on > > I used this extensively at my last position with very good results. > > Graham > > > -----Original Message----- > From: Chuck Ross [mailto:chuck.ross@signiant.com] > Sent: Thursday, May 24, 2001 3:06 PM > To: Net App List (E-mail) > Subject: Why is root not root? > > > Hi all, > > Why is root on a Unix admin host for a filer in a mixed env not really > root? By that I mean why is it possible to create directories on the > NetApp via a Cifs share that root has no access to? > > Chuck Ross > > SIGNIANT - A Nortel Networks Venture > 1550 Carling Avenue, Suite 101 > Ottawa, ON > K1Z 8S8 > > email: Chuck.Ross(a)signiant.com > tel: (613) 761-3605 > fax: (613) 761-3629 >

1 0

Jump to page:

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

2003

2002

2001

2000

1999

1998

1997

toasters