I'm very sorry. By some error I never got this message sent. So here it is, over a month too late. Fantastic...


Mogens

==============================================================

Ah, good to be back online with Tim Gorman on the old and wonderful 1575.

1575 was introduced in 7.1. Not as an error, because the code that
creates this error has been around for many years before that. 1575 was
introduced to signal an unpleasant wait situation for the ST
lock/enqueue - a warning to the DBA.

Used extents (in UET$) and free extents (in FET$) are managed
"together", meaning that 1) if you want to delete a record in UET$ and
insert it in FET$ (that means an extent has been dropped/freed), 2)
delete a record in FET$ and insert it in UET$ (extent has been
allocated) or 3) delete a bunch of records in FET$ and inserting only
one with the summary information in the same FET$ (coalescing extents) -
you have to make sure that nobody else is messing with UET$/FET$ at the
same time.

So Oracle takes out the massive ST enqueue on both UET$ and FET$ while
it performs 1, 2 or 3 mentioned above (and probably some other things I
don't recall). If somebody else tries to get the ST enqueue while it's
still being held by another session, you'll get the 1575 signalled in
the alert log - in order to simply notify you that there has been
queueing on the ST lock.

As long as you have DMTs you risk getting 1575. It might be possible to
get it with LMTs, too, but I haven't seen it personally (which is
information without value - there are so many things I haven't seen yet,
like lizards playing chess or Cary taking a quick shower).

Temporary tablespaces (in 7.3?) replaced the ST enqueue with a latch per
temp tablespace (this helped a lot in OPS environments).

Management manouvres of various kind, like having standard sizes of
extents, not coalescing ever (hence the 7.1 change whereby a tablespace
with pctincrease=0 didn't get coalesced), etc. also helped.

But it was LMTs that finally solved it. I thought. Until this thread.

So now I'm curious as to what is happening here.

Mogens

Tim Gorman wrote:

Tanel hit the nail on the head.  In the past, ORA-01575 was usually
associated with temporary tablespaces that were DMT and not tablespace type
"TEMPORARY" (which started in Oracle7.3).  First and foremost, please make
sure you are using a TEMPORARY tablespace which is locally-managed and uses
TEMPFILEs...

It might be interesting to monitor V$LOCK for TYPE = 'ST' to see what
sessions are holding this enqueue.  If the activity is too transient,
perhaps querying V$SESSION_EVENT where EVENT = 'enqueue' might indirectly
imply which sessions have waited on an enqueue (not necessarily "ST",
thought!) sometime in the past...



on 8/13/03 7:04 AM, Tanel Poder at [EMAIL PROTECTED] wrote:



Hi!

You can always schedule alter tablespace coalesce's during low usage time.
But you should check whether you have adjacent free extents in your
tablespaces at all? If you're not doing lot's of dropping or truncating
objects, then you shouldn't have. Thus no need for coalesce either. Just
check that all of your sort segments go to the temp tablespace (which should
be in temporary mode, preferrably LMT as well).

Tanel.




thanks for the info. We do have a number of DMTS in
the database. Three of them have pct_increase of 50%,
the rest - 0. Should I consider changing the
pct_increase to 0 in all tablespaces in order to get
rid of this ora 1575? Wouldn't I want to have an
automatic coalesce process for the DMTS though?

thank you

Gene
--- Tim Gorman <[EMAIL PROTECTED]> wrote:


Haven't seen this error since Oracle7...

If the message is hitting the "alert.log", then
chances are good it is
coming from SMON.  SMON is attempting to acquire the
"ST" (a.k.a. Space
transaction) enqueue in preparation for coalescing
free space in some
tablespaces.  However, if it is unable to acquire
"ST" after a couple
seconds, it times out and issues ORA-01575 to the
alert.log.

So, based on experiences from 6-7 years ago:

   * do you have a lot of "dictionary-managed"
tablespaces?
   * do these DMT's have default PCTINCREASE
non-zero, thus attacting
     SMON to do coalescing?

If so, I'd suggest going to "locally-managed"
tablespaces if at all
possible...



on 8/12/03 12:44 PM, Gurelei at [EMAIL PROTECTED]
wrote:



Hi all:

I'm seeing the ora-01575 error in the alert


logfile.


The article on the metalink refers to the


parameter


which I think is obsolete in the ORacle version we


are


running (8.1.7). What does this error refer to?


Any


thoughts? references?

thanks

gene

__________________________________
Do you Yahoo!?
Yahoo! SiteBuilder - Free, easy-to-use web site


design software


http://sitebuilder.yahoo.com


--
Please see the official ORACLE-L FAQ:
http://www.orafaq.net
--
Author: Tim Gorman
 INET: [EMAIL PROTECTED]

Fat City Network Services    -- 858-538-5051
http://www.fatcity.com
San Diego, California        -- Mailing list and web
hosting services



---------------------------------------------------------------------


To REMOVE yourself from this mailing list, send an
E-Mail message
to: [EMAIL PROTECTED] (note EXACT spelling of
'ListGuru') and in
the message BODY, include a line containing: UNSUB
ORACLE-L
(or the name of mailing list you want to be removed
from). You may
also send the HELP command for other information
(like subscribing).


__________________________________
Do you Yahoo!?
Yahoo! SiteBuilder - Free, easy-to-use web site design software
http://sitebuilder.yahoo.com
--
Please see the official ORACLE-L FAQ: http://www.orafaq.net
--
Author: Gurelei
 INET: [EMAIL PROTECTED]

Fat City Network Services    -- 858-538-5051 http://www.fatcity.com
San Diego, California        -- Mailing list and web hosting services
---------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: [EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
(or the name of mailing list you want to be removed from).  You may
also send the HELP command for other information (like subscribing).









Hesin e-postur er kanna�ur fyri virus av F�roya Tele.


This e-mail was virus scanned by Faroese Telecom.

--
Please see the official ORACLE-L FAQ: http://www.orafaq.net
--
Author: =?ISO-8859-1?Q?Mogens_N=F8rgaard?=
 INET: [EMAIL PROTECTED]

Fat City Network Services    -- 858-538-5051 http://www.fatcity.com
San Diego, California        -- Mailing list and web hosting services
---------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: [EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
(or the name of mailing list you want to be removed from).  You may
also send the HELP command for other information (like subscribing).

Reply via email to