logs archiveIRC Archive / Freenode / #oracle / 2010 / February / 6 / 1
zasz
fine use AWR reports
you can find this information
RJarett
AWR doesnt tell you anything of what you need for this sort of issue
since the report would say the same before and after the fix
Oracle is the cause, but oracle is not the method to track it down or solve it
zasz
the EM shows how much IO oracle is using
are you telling me you can't correlate that to a sql id in oracle
RJarett
raw io isnt always the problem
zasz
em just uses the perofrmance views
RJarett
what if you have 70,000 transactions a second running
zasz
you heard me say i haven't even touched performance yet :)
RJarett
each can be tuned perfectly but still causing a lockup issue
zasz
I'm running on a big fat 3 months or so as an oracle guy?
lol
deadlocks
         

RJarett
its been longer than that
zasz
check the alert log
check listener log
RJarett
stop rambling off sh*t that you dont even know what the term means
zasz
i know what a dead lock is wise ass
RJarett
Thats the battlecry of the stupid, or the mcse
zasz
it's when one pid is waiting on the resources of another, while the other pid is doing the same
so it's a win/lose situation and the rdms has to pick a victim
RJarett
thats a very broad explaination thats not really right
no
zasz
i know exactly what it is
RJarett
the db doesnt just pick a victim in a deadlock
zasz
they are both requesting resources that are held by the opposing session
well it has a resolution system
RJarett
but the issue i speak of is nothing of the sort
zasz
so yes i know what a deadlock is
ok so you start the batch process and all users complain
do new connections into the db work fine
RJarett
yes
zasz
do they freeze up shortly after connecting?
RJarett
no
zasz
are you using the resource manager
RJarett
doesnt matter
load avg can be below 1 and have this happen
zasz
well if new connections work fine then i'm not sure
sorry to let you down :P
RJarett
like i said, its about more than knowing oracle
zasz
I would call RJarett
         

RJarett
its about knowing things like io schedulers on the OS
zasz
and he would tell me the answer while simultaneously insulting me
RJarett
I dont insult people who pay me unless ive known them a while and they really deserve it
zasz
lol
RJarett
do you run oracle on linux?
zasz
windows and linux
RJarett
do you know which io scheduler your system is using right now?
or what other ones are available?
zasz
nope
RJarett
or even further down, how to tune it or your mounts to reduce the IO load?
zasz
nah my background is MS, my linux exp from the past was simple stuff like setting up firewalls, vpn servers, etc
but I definitely need to get better with Linux
I have my plate full at the moment learning Oracle
RJarett
windows can face similiar issues, theres just not a damn thing you can do to fix it
well you said you are studying backups too
this is vital
zasz
at the moment yea, once i'm done reading i need to practice the hell out of it
i didn't realize how goddamn much you could do with RMAN
RJarett
people done have infinite amount of time tobackup
plus how can you guarentee your datas on disk from you backup, it may not be
zasz
guarantee that my data is on disk from backup ...
that makes no sense
RJarett
really?
so you know that when you insert data, its in your datafile, and then in your backup?
How?
zasz
is this a trick question... like if you insert stuff in no logged mode... like a direct insert... then it's not there in the archive log backup
so you need to do a level 1 to caputre it
RJarett
even in a logged mode
zasz
how can you guarantee that data is in the backup piece?
i would imagine the scn
RJarett
with my main DB it can be up to 15 minutes between the time i commit to the time it hits my datafile
though most wont have that extreme
but up to a minute
zasz
oh if you want to guarantee that all committed data is in the data file
do a consistent backup :p
if not you'll just need to recover with the archive logs and redo
how long have you been working with oracle for
RJarett
eh 15 years maybe
zasz
i'm smart but i can't compete with 15 years just yet :P
RJarett
but its not always been my main focus
zasz
so what's the answer?
RJarett
When you need to know, youll know the answer
zasz
you can force a checkpoint right
RJarett
yeah
zasz
i'm sure there's a way to trigger dbw
to write all the dirty buffers
RJarett
but sometimes you dont want to
zasz
is it just by issuing CHECKPOINT;
RJarett
hell with oracle adding async commits and batch nowait commits, you never know where data is
zasz
you can
shuttdown immediate;
RJarett
yeah goodluck with that
zasz
shutdown immediate that is
well hey,
how else can you do it?
RJarett
10 hours later it may not be done
zasz
the requirement that all comitted transactions need to be in the data files at the time of backup requires a consistent backup
i suppose you can quiesce the database
wait till the users go away
checkpoint
backup
RJarett
they never go away
db users are like dumb people on IRC
zasz
what version of oracle did you start with
6 or what
RJarett
7.3
zasz
so what's the answer
or is there one
there isn't
you trickster
RJarett
to what question?
zasz
your last retarded one
RJarett
You cant
zasz
lol
RJarett
You can just force certain conditions based on legal requirements
zasz
you can though
shutdown the db
take a consistent backup
bring it up
that meets the requirements of your question
hence you are incorrect
RJarett
Yeah you do that at fidelity investment and you just got a 1.5M$ SEC fine on your back, and lost customers
zasz
you didn't mention the business requirements of the company
you merely asked how
i gave you the answer
and it is correct
you saying that you cannot is incorrect
RJarett
Nope
Youre not thinking like an enterprise DBA
Where shutdowns are not an option
zasz
22:08 < RJarett> so you know that when you insert data, its in your datafile, and then in your backup?
shutdown the db
RJarett
Youre thinking very microsoftish
zasz
lol
well it's the only real answer
you can force a checkpoint but you can still have dirty buffers between the checkpoint and backup
you'll still need to perform recovery because it's an inconsistent backup
RJarett
plus if your box lost power, you could have a ton of data that was not written yet anywhere
the beauty of the new async commit system
zasz
you must protect against that
but provided that does not happen, a non abort shutdown will guarantee that all dirty buffers are written
it is stupid
it is also a stupid question
:P
you know
RJarett
Yeah when you face a crash in that situation and its your job to prevent it or address why something was lost, you think its stupid too
zasz
that happened to you?
power / component failure?
RJarett
Its just how things are
zasz
were you on irc in 2002
RJarett
You need to be the one making the educated choices on where to risk data, and how to work around it
well before 2002
since like 94
zasz
i remember i was doing a 9i course, and i was having trouble with the installation
someone in an oracle channel, i think it was efnet, essentially told me to give up lol
RJarett
me or tenchi i would think
zasz
i got it working eventually but i think it may have been you
you are very mean to new people!
that was the first time I ever touched oracle... i didn't touch it again until 2009 lol
RJarett
then i wasnt tough enough
zasz
wrong
it's much simpler now
with experience
RJarett
any idiot can do something simple
zasz
lordy i don't think oracle is simple
RJarett
But it quickly scales outside of that
zasz
but learning it is much less of a task than 2002 when i wasn't even in IT yet
give me a good year
lol
EFFINBUCKY
hello
« prev next »