Question about resource's start dependency in Clusterware

From: Jure Bratina <jure.bratina_at_gmail.com>
Date: Wed, 29 Jan 2014 15:45:45 +0100
Message-ID: <CAC08BHJb4xn_vqWkJU16ex1BnZ7FFD6n1bnYBb9fN+mE19t7ew_at_mail.gmail.com>



Hi,

I have Oracle Clusterware version 11.2.0.4.0 and have a question about hard start dependencies. Suppose resource A has a hard start dependency on resource B:
START_DEPENDENCIES=hard(B)

My question is: if we try to start resource A, and resource B is not running, will the resource B be automatically started and after that resource A will start? I think that is how it should work and a simple test shows that it indeed works that way. The reasons I'm asking this questions are two:

  1. The documentation at http://docs.oracle.com/cd/E11882_01/rac.112/e16794/crschp.htm#CWADD91290states about the hard dependency:
    "Define a hard start dependency for a resource if another resource must be
    running before the dependent resource can start. For example, if resource A has a hard start dependency on resource B, then resource B must be running before resource A can start."

Nowhere is stated that upon starting resource A, resource B will be automatically started first if not already running (and the "must be running" also seems to suggest that the hard dependency is only a check if the referenced resource is already running and not that if it's not, it will be implicitly started). On the other hand, for the weak dependency that's explicitly stated: "If resource A has a weak start dependency on resource B, then an attempt to start resource A attempts to start resource B, if resource B is not running."

2) I asked a similar question on dependencies in the MOS Community forum (in the section: "Database - RAC/Scalability", question title: "Question about the pullup dependency") and got that response: "The test you have done is what is happening but its not what is defined in the docs. we have filled a bug BUG 17877379 : HARD DEPENDENCY NOT WORKING AS EXPECTED , which is currently with our development." Unfortunately the bug 17877379 is not public.

I'm very grateful to the person that answered my question on the MOS forum, but that answer made me a little bit confused. If such a fundamental functionality of Clusterware as a resource's hard start dependency doesn't work correctly, I guess it's very unlikely it wouldn't be spotted until 11.2.0.4 (unless of course the bug was introduced in that version). Furthermore, many blogs and books about RAC explain that hard start dependencies do start the referenced resource, e.g. Martin Bach & Steve Shaw's "Pro Oracle Database 11g RAC on Linux" has this example:

[oracle_at_london1 ~]$ crsctl status resource ora.prod.db -p NAME=ora.prod.db
TYPE=ora.database.type
ACL=owner:oracle:rwx,pgrp:oinstall:rwx,other::r-- [...]
SPFILE=+DATA/prod/spfileprod.ora
START_DEPENDENCIES=hard(ora.DATA.dg) [...] pullup(ora.DATA.dg) STOP_DEPENDENCIES=hard(intermediate:ora.asm,shutdown:ora.DATA.dg)

"You can see that a start dependency has been created for the disk group
DATA, which will also be started if needed as part of the database start."

Thank you in advance for any answers

Regards
Jure Bratina

--
http://www.freelists.org/webpage/oracle-l
Received on Wed Jan 29 2014 - 15:45:45 CET

Original text of this message