logs archiveIRC Archive / Freenode / #oracle / 2009 / December / 2 / 2
TomJ
ok, a normal kill did nothing
damn I can't eve kill -9 it
TheBonsai
status?
TomJ
Anyone know how to re-install cssd? I'm setting up a second DB server. all of the oracle software (ASM + normal) I have simply mounted on the 2nd DB server; i.e., my oracle software is written to SAN, and I 've unmounted the LUN frm the primary DB serer and nowhave mounted it on the secondary.server. but cssd won't start - no errors, it just doesn't run the binary
I manually copied over /etc/init.d/init.cssd, and I rsync' /var/opt/oracle from the primary DB
I thought it might just be easier to run the cssd installer on DB2 but I can't find it - it's not root.sh as I thought
platform is Solaris 10-x64
archstanton
Hi Can anyone help got TNS listener error messages
The server is a 32bit windows server with PAE enabled it uses about 3-4gb of memory
with 8 in Total available
It looks like TNS is just crashing every now and again is there anything i can do to check it is running out of memory and what might be using the most
DiscordianUK
I know nothing about Oracle on windows I'm afraid
archstanton
yeah i bet its great on another OS
TomJ
isn't everything?
archstanton
Not really the listener can crash on redhat
what would you use on another OS?
Dormir
TomJ: what error do you get?
I've done something similar for a failover node
         

archstanton
TNS-12518: TNS:listener could not hand off client connection
mind you im wondering if its more of a database crash as i think the test on the same box is accessible when this one isnt
it recovers itself after some time which conforms my beleive that its a memory issue
Was hoping an Oracle guru would go aha of course yes its because of....
Or does everyone just dump the default install on and its works like a dream from then onwards
Discordian
Don't Oracle Gurus cost?
archstanton
Only the good ones
and theres not many about
Dormir
lol
archstanton
hehehe
poke you with a stick there
TomJ
Dormir: no error at all. it just doesn't start. I haven tried debugging /etc/init.d/init.cssd yet though
Dormir: what I've done is: a) manually copied /etc/init.d/init.cssd to DB2 b) copied the whole of /var/opt/oracle/ to DB2 3) mv /var/opt/oracle/scls_scr/host-db1 /var/opt/oracle/scls_scr/host-db2
beore I did steps 2 and 3 I got errors in init.cssd about missing directories
all the oracle software (/opt/oracle/server/10.2.0 and /opt/oracle/asm/10.2.0) are identical across DB1 and DB2, because they're on a SAN LUN which I unmounted from DB1 and mounted to DB2
Dormir
ok, I wasn't able to copy ASM across, just the db
TomJ
ah - what happened when you tried to copy ASM?
Dormir
what we did was install ASM locally on each box, then failed over the db home and all of the disks
couldn't get it working
the consultant we were working with from Veritas said we shouldn't fail over ASM
when we did it without failing it over, it worked, so we kept it
don't remeber the exact errors, though
TomJ
so if I install ASM again, I would install the softtware obviously, but then how do I go about setting up DB2 so it gets the same diskgroup as DB1? is there any setup/config step required there?
normally I do "Setup ASM" via dbca and create a diskgroup etc
Dormir
all you have to set is asm_diskstring
TomJ
what would I do when the diskgrou pis already created, I just want to be able to mount it here?
Dormir
ASM stores the diskgroup info in the header of the disk
so, when another ASM instance mounts it, it's able to mount it
TomJ
ok, so on DB2 I would: a) install ASM software b) run DBCA c) choose "Setup ASM" d) select the diskgroup that it finds automatically ?
Dormir
I had the same questions :)
TheBonsai
it would be enough to start a ASM with the proper diskstring
Dormir
I would create an ASM instance then go into sqlplus and update asm_diskstring
TheBonsai
or that, yea
Dormir
or, liek Bonsai said, start it with the diskstring in the init.ora
TomJ
asm distring is e..g /dev/rdsk/ right?
         

Dormir
but I don't like dbca too much
TheBonsai
my tenor was: dbca unneeded/dangerous
Dormir
something like /dev/rdsk/*
TomJ
right
Dormir
or /dev/rdsk/c0t1d*
delewis
or create devices via mknod that point to the appropriate rdsk.
TomJ
so you're saying: on DB2 a) install ASM b) run cssd c) just start ASM and it will mount the diskgroup as long as asm_diskgroup is correct?
TheBonsai
today i installed a testsystem with 11.2 and ASM dg on loopback'd files heh
delewis
and change your asm device string to something consistent like /dev/asm*
TheBonsai
(loop -> raw -> asm i mean)
Dormir
oh, wait, there's a file that will reset the cssd
but damn if I can remember it...
delewis
yeah, some OUI command you can use to reset it.
it's in ${ORACLE_HOME}/oui/bin, IIRC.
TomJ
so a switch to trigger cssd re-installation?
delewis
the upgrade/removal docs for Clusterware go over resetting CSSD.
Dormir
yeah, there's a script that can fix it
TomJ
ok i will look this up, thanks
Dormir
yep, but, the easiest way is
install ASM locally on each node
fail over only the db home and the disks
make sure each ASM has asm_diskstring set correctly
TomJ
yeah maybe I should do that
I put /opt/oracle/asm on my SAN LUN too but I can make it local instead, just have /opt/oracle/server on the LUN
TheBonsai
sounds more robust
delewis
http://download.oracle.com/docs/cd/B28359_01/rac.111/b28255/adddelclusterware.htm#CHDDCJCI
TomJ
thanks delewis
delewis
if you just want to reset CSSD then you can probably just do a localconfig reset
TheBonsai
aaaah that was it
Dormir
lol I just said the same thing
localconfig
TheBonsai
we're not getting younger, Dormir :D
TomJ
ahh localconfig, of course, I did run that for the first ASM home
Dormir
Bonsai: I know
TomJ
ok, I ran localconfig reset on DB2 and now cssd is running
now I will see if I can a) mount the DB on DB2 b) still mount the DB on DB1 :)
if I can't do b) I'll go with non-shared ASM homes
TheBonsai
you probably will loose the cssd config again
well
"loose"
er
invalidate
TomJ
yeah that's what I'm wondering
Dormir
I believe so
which is part of the reason seperate ASM homes is more appealing
TomJ
yeah
ok,DB running on DB2 ok
TheBonsai
maybe you can specify the config location somewhere
and leave the config local
no idea
TomJ
yeah worth a go
ok as expected, now cssd won't start on DB1
I'll check for a local way, otherwise I'll just unshare the homes
thanks guys, very helpful
TheBonsai
or make it a real cluster
(not a RAC, i mean use the clusterware)
TomJ
that has benefits even if I only ever want active/standby?
TheBonsai
well, both nodes share an OCR ;)
you don't invalidate it
i don't think it's worth a shot
TomJ
hehe, sounds like overkill just to resolve this issue :)
yeah
So I assume the issue I am facing is that cssd generates some kind of key (is that the OCR?) which is server specific?
TheBonsai
looks like
TomJ
incidentally is there a way to change where ASM stores its admin files? I set ORACLE_BASE but it's still going into ORACLE_HOME/admin/ for ASM
that said I dont htink I set ORACLE_BASE before installing ASM, so it migt have hardcoded it at that stage
can I change that after the fact?
ah of course it's just user_dump_dest etc
it's annoying - it appears the OCR key is /opt/oracle/asm/10.2.0/cdata/localhost/local.ocr - so you'd think that you could have cdata/hostname/local.ocr, but that doesnt seem to work
unless that's configurable with a parameter maybe
TheBonsai
maybe, but it's not wise to go that deep into the dark catacombes of oracle ware
you probably never come back
TomJ
hehe
well I'll only do it if there's a standard parameter to contorl it
it just seems like if theyt call the dir 'localhost' that it's meant to support multiple keys for multiple hostnames
of course a potentially very simple idea is ln -s /opt/oracle/asm/10.2.0/cdata /usr/local/oracle/cdata
then I can have the keys local on each instance while still having the ASM ORACLE_HOME shared
TheBonsai
worth a try
TomJ
yeah. having two homes isn't the end of the world, i'd just like to avoid it if possible. just saves on upgrade/patching time tec
etc
oh what a moron!
oracle@aircel-db1:~$ cat /var/opt/oracle/ocr.loc
ocrconfig_loc=/opt/oracle/asm/10.2.0/cdata/localhost/local.ocr
local_only=TRUE
I saw that file earlier but then forgot about it
i guess I still want local_only=TRUE
TheBonsai
:D
how obvious
shame :D
though...
wait a minute
is the OCR updated on the fly at startup when the crs/css stuff is updated?
TomJ
no just on reset I believe
TheBonsai
not that you patch server#1 and server#2 can't start because ocr is too old
TomJ
yeah if I patch server#1 then server2 is also patched
because they share all ORACLE_HOMEs
and they will never run concurrently, just active/standby
hmm, nothing actually seems to use ocr.loc though
neither localconfig reset nor init.cssd start
when I did localconfig reset it just overwrrote my changed ocr.loc and put it back in cdata/localhost/ . then I stopped cssd, manually edited ocr.loc to where I wanted it, moved local.ocr there, and tried to start cssd, but it failed to start again
very odd
oh no wait, might be luser error
ok here's a question. why on earth does shutdown STILL pause forever on a DB that's a) just been started b) has no listener c) and therefore cant possibly have any clients besides this sqlplus that I'm shutting it down from?
(I mean 'shutdown' as opposed to 'shutdown immediate')
avaia
it wants immediate to still have meaning!
TomJ
hehe
avaia
shutdown casual
TomJ
immediate does have meaning; it's shutdown itself that has none :)
woot, so now I have both DBs with all ORACLE_HOMEs shared
thanks guys :)
Dormir: did you see the solution, FYI?
Dormir
nah, what did you decide to do?
puuck
why are you sharing the oracle home?
TomJ
Dormir: the file /var/opt/oracle/ocr.loc (might be /etc/oracle on other platfors - this is Solaris) has the line ocrconfig_loc=/opt/oracle/asm/10.2.0/cdata/localhost/local.ocr . this is read by init.cssd start
however ,when you do localconfig reset it overwrites that file and puts it back to $ORACLE_HOME/cdata/localhost/ - so after doing localconfig reset on both systems, I manually moved cdata/localhost/local.ocr to cdata/$HOSTNAME/local.ocr and then I manually edited ocr.loc on each system to point to the right file
localconfig reset takes a -paramfile so it's probably possible to specify the location in there too, but I didnt research how to do that
puuck: I have two DB servers, DB1 and DB2, with shared SAN storage. I have ASM's ORACLE_HOME, and Oracle's ORACLE_HOME, stored on a LUN on the SAN. When I want to failover the DB, I unmount the software LUN from DB1, mount it to DB2, then start the db
the result is that all DB software is the same for both hosts
and I only need to patch/upgrade once
as the second DB will only ever be a failover for the primary DB, I saw no reason to install oracle or ASM twice and have it locally on each system
easier to have a centralised software store that either system can mount when it's going to be priary
*primary
puuck
ah...so you're ok with the storage and integrity of the home being a single point of failure?
TomJ
yes; if the storage fails, the whole DB goes down anyway
Dormir
TomJ: Ok, that makes sense
TomJ
(of course RMAN backs up to a different set of disks)
puuck
why not just RAC it if you're using CRS to get around hardware failures?
TomJ
puuck: that's double the license cost and one server is enough power for our needs
well, more than double as you need the RAC license as well as 2 x EE licenses
with active/standby you dont pay for the second DB server as long as it's not in use for more than 10 days a year
puuck
gotcha
TomJ
By the way, what do you Solaris users do for backups of UFS filesystems on systems running CSS? With ocssd.bin running, ufsdump faills because it can't lock the fileystem, because ocssd.bin is a real time proess or something similar
(or it might be fssnap that fails, I forget which)
TheBonsai or delewis, you encountered that problem? ^^
« prev 1 2 next »