'RE: [Lustre-discuss] MDS died'

[prev in list] [next in list] [prev in thread] [next in thread] 

List:       lustre-announce
Subject:    RE: [Lustre-discuss] MDS died
From:       "Mc Carthy, Fergal" <fergal.mccarthy () hp ! com>
Date:       2005-03-30 16:18:56
Message-ID: 575D0CDD99F591478ADD639EF4E36E7E017D2ECE () iloexc01 ! emea ! cpqcorp ! net
[Download RAW message or body]

Alessandro,

The upcall script looks familiar. ;-)

The problem is that you need to pass the --select info to the upcall
script's lconf command also otherwise the client will not know about the
location change of the servers.

You can do this by:

a)  regenerating the XML file, but this time specifying the preferred
node as the first definition for the device(s) in question, with the
other node second. The config implicitly selects the first node defined
as being associated with a device as the preferred server.

b)  store the select info in some shared fashion. E.g. a common config
file which is used to generate the --select option to lconf

c)  store the config in LDAP and use the lactive command to update the
preferred server setting

Fergal.

--

Fergal.McCarthy@HP.com

(The contents of this message and any attachments to it are confidential
and may be legally privileged. If you have received this message in
error you should delete it from your system immediately and advise the
sender. To any recipient of this message within HP, unless otherwise
stated, you should consider this message and attachments as "HP
CONFIDENTIAL".)

-----Original Message-----
From: lustre-discuss-admin@lists.clusterfs.com
[mailto:lustre-discuss-admin@lists.clusterfs.com] On Behalf Of
Alessandro
Sent: 30 March 2005 15:57
To: lustre-discuss@lists.clusterfs.com
Subject: Re: [Lustre-discuss] MDS died

Thanks, sorry for my inattention... (I 've written 100 times 
lconf --help...:-))

Now I'm thinking:
if  "... If you use multiple --select statements I don't believe they
are
cumulative so possibly only the last set of info is used, meaning that
you may only be seeing the MDS would try to start up with no OST or vice
versa...",
what can I do if at the same time (on the same machine) MDS and OST are 
died?

In other words, if my Server1 is poweroff (where MDS and OST reside,
plus 
client1), how can I access to shared storage partition with my Server2 
(where is only the client2 when Server1 is on)???

The failover method doesn't work...
A new config script only for Server2(with MDS, OST and client) doesn't 
work...

Can you help me?

Reminder info (useful for other people, perhaps...):
Server1: MDS, OST, client1
Server2: client2
My config script

# Create nodes :
rm -f config.xml
lmc -o config.xml --add node --node Server1
lmc -m config.xml --add net --node Server1 --nid Server1 --nettype tcp
lmc -m config.xml --add node --node Server2
lmc -m config.xml --add net --node Server2 --nid Server2 --nettype tcp

# Configure MDS
lmc -m config.xml --add mds --node Server1 --mds 
mds-test --failover --fstype ext3 --dev /dev/sdb5
lmc -m config.xml --add mds --node Server2 --mds 
mds-test --failover --fstype ext3 --dev /dev/sdb5

# Configure OST
lmc -m config.xml --add ost --node Server1 --ost 
ost1-test --failover --fstype ext3 --dev /dev/sdc1
lmc -m config.xml --add ost --node Server2 --ost 
ost1-test --failover --fstype ext3 --dev /dev/sdc1

# Configure client
lmc -m config.xml --add mtpt --node Server1 --path /mnt/lustre --mds 
mds-test --ost ost1-test
lmc -m config.xml --add mtpt --node Server2 --path /mnt/lustre --mds 
mds-test --ost ost1-testStarting the system
Server1#lconf --node Server1 config.xml
Server2#lconf --node Server2 config.xml

Try (done but doesn't work)
Server1 poweroff
Server2# lconf --node Server2 --select 
mds-test=Server2,ost1-test=Server2
--lustre_upcall=/home/tests/lustre/upcall_script 
config.xml

My upcall_script (do you know this?)
#!/bin/bash

# Get the Upcall Type
UPCALL_TYPE="${1}"
shift

case "${UPCALL_TYPE}" in
FAILED_IMPORT)
	lconf --recover \
		--tgt_uuid ${1} \
		--client_uuid ${2} \
		--conn_uuid ${3} \
	;;
esac
Cheers, Alessandro

----- Original Message ----- 
From: "Mc Carthy, Fergal" <fergal.mccarthy@hp.com>
To: "Alessandro" <alessandro_avallone@hotmail.com>; 
<lustre-discuss@lists.clusterfs.com>
Sent: Tuesday, March 29, 2005 5:53 PM
Subject: RE: [Lustre-discuss] MDS died

Alessandro,

I think that you may need to specify the active server info using a
single select statement, i.e.

lconf ... --select ost1-test=Server2,mds-test=Server2 ...

This is what the lconf --help output shows:

% lconf --help | grep select
--select         service=nodeA,service2=nodeB  (default=[])

If you use multiple --select statements I don't believe they are
cumulative so possibly only the last set of info is used, meaning that
you may only be seeing the MDS would try to start up with no OST or vice
versa.

Fergal.

_______________________________________________
Lustre-discuss mailing list
Lustre-discuss@lists.clusterfs.com
https://lists.clusterfs.com/mailman/listinfo/lustre-discuss
_______________________________________________
Lustre-discuss mailing list
Lustre-discuss@lists.clusterfs.com
https://lists.clusterfs.com/mailman/listinfo/lustre-discuss

[prev in list] [next in list] [prev in thread] [next in thread] 