On 05/12/2015 02:09 PM, DRBD User wrote:
Hi
@Cesar: thx for your suggestion - but i don't want to do a manually fence.
from Digimer's replies to your posts:
1- the dlm "lock" will be released once the crashed node is set to a
*known* state in pacemaker. Without releasing, forget about using your
shared fs.
2- a *known* state requires a working stonith setup: either automatic
(IPMI, switched PDU, ...), or manual, as Cesar described.
Now, if you don't want to use stonith and you're brave enough to risk
having a split-brain (you have good backups, the data on the shared fs
is transient/not important, ...), I imagine you could have a shell
script with a loop running in the background that would automatically
ack a manual fence when needed. Or you could write a dummy stonith agent
that would always return success.
during testing i found out, that after pulling power plug the shared directory
it is not completely inaccessible : it is readable, only a write will block
until crashed node restarts - BUT what if crashed node never restarts ? (my
service saves it state into shared directory an should not block)
maybe its better to switch from active/active to active/passive - or is here
the situation (pull power plug, blocking..) the same ?
thx
Gesendet: Dienstag, 12. Mai 2015 um 12:33 Uhr
Von: "Cesar Peschiera" <[email protected]>
An: "DRBD User" <[email protected]>, [email protected]
Betreff: Re: [DRBD-user] Dual Primary Mode: Shared Directory blocked after node
crash until reboot
About of your problem of fence:
Instead of use a fence by Hardware, you can use a manual fence that come with
the cluster software.
Please read this:
1- It not require any hardware.
2- This option isn't advisable in production environments, but useful in
development environments.
3- The file used is "fence_ack_manual"
4- It is executed by CLI in a node that is alive for apply the fence to other
server.
5- For use it, It is advisable that first disconnect totally the electric power
on the server that will be fenced, the goal is to shut down brutally the server
that will be fenced before of run the fence command.
6- Finally, execute this command in a node that is alive:
Shell# /[PATH]/fence_ack_manual [IP or Name of the Node that will be fenced]
7- Follow the steps as directed by this command.
I hope this information is helpful.
Best regards
Cesar
----- Original Message -----
From: DRBD User[[email protected]]
To: [email protected][[email protected]]
Sent: Tuesday, May 12, 2015 5:39 AM
Subject: [DRBD-user] Dual Primary Mode: Shared Directory blocked after node
crash until reboot
the DRBD status is (regardless of 'nice' shutdown (eg reboot) or 'abrupt' kill
(eg pull power plug))
cs:WFConnection ro:Primary/Unknown ds:UpToDate/Outdated
but only with a 'nice' shutdown the shared directoy is still accessible...
Gesendet: Dienstag, 12. Mai 2015 um 09:44 Uhr
Von: Digimer <[email protected][[email protected]]>
An: "DRBD User" <[email protected]>, [email protected]
Betreff: Re: [DRBD-user] Dual Primary Mode: Shared Directory blocked after node
crash until reboot
On 12/05/15 03:42 AM, DRBD User wrote:
pacemakers pcs property stonith-enabled is currently set to false
Well there's your problem. :)
Since i don't have any (hardware) STONITH device, i have set stonith-enabled to
false.
DRBD's fencing rule is set to : 'fencing: resource-only'
My goal is: if one node crashes, the other node should take over the work
immediately. But actually i have to wait the reboot time of the crashed node. I
thought, that in such a situation the active node (rather the shared directory)
is immediately usable ?
May be i should use another fence script ?
I tried to create the resource with operation 'on-fail=restart' - but no
success ...
Any other suggestions ?
You *CAN NOT* safely proceed when a node stops responding _until_ you
have put the lost node into a known state. To do otherwise would be to
risk a split-brain.
A good fence device are switched PDUs, like the APC-brand AP7900 (not
all makes/models are supported, so check first before buying other
brands). The AP7900 can usually be found used for ~$200 and makes an
excellent external fence device.
Trying to use DRBD without proper fencing will result in pain and
heartache. The delay needed to fence a lost node is FAR preferable to
risking a split-brain.
--
Digimer
Papers and Projects: https://alteeve.ca/w/[https://alteeve.ca/w/]
What if the cure for cancer is trapped in the mind of a person without
access to education?
_______________________________________________
drbd-user mailing list
[email protected]
http://lists.linbit.com/mailman/listinfo/drbd-user[http://lists.linbit.com/mailman/listinfo/drbd-user]
------------------------------------------------------------
_______________________________________________
drbd-user mailing list
[email protected]
http://lists.linbit.com/mailman/listinfo/drbd-user_______________________________________________
drbd-user mailing list [email protected]
http://lists.linbit.com/mailman/listinfo/drbd-user[http://lists.linbit.com/mailman/listinfo/drbd-user]
_______________________________________________
drbd-user mailing list
[email protected]
http://lists.linbit.com/mailman/listinfo/drbd-user
_______________________________________________
drbd-user mailing list
[email protected]
http://lists.linbit.com/mailman/listinfo/drbd-user
