Re: log-begin on new transactions (7.11.UC1)

In article <E5EBqs.Mpz@nonexistent.com>, Jacob Salomon <jake@apparel.net> writes >Hi all, >Warning 1: You may have to widen your windows for this. Sorry. > Nope - luucky me! >Just some notes and wheel-inventing for OnLine admin folks. > >I was testing out some scripts on sysmaster and ran into strange >situation. > >Window1: begin work; > set isolation to cursor stability;> >window2: onstat -ux # Obviously edited to isolate my session > >Userthreads >address flags sessid user tty wait tout locks nreads >nwrites >c1fabf1c Y--P--- 13391 jake ttyq0 c2f963f0 0 1 6 >4 > >Transactions >address flags userthread locks log begin isolation retrys coordinator >c1fcba4c A---- c1fabf1c 1 0 CURSOR 0 > >Notice: That's my transaction being described; I promise it's the only >one with that isolation level. So the in-memory array of transaction >info describes this transaction but the log-begin is 0. This tells me >there is nothing on disk or the log buffer describibg this transaction. SOunds reaonsable - it too late to bother wade throught the manuals, I'll believe you. >In hindsight, I suppose this may be reasonable; if the system were to >crash now, it would not have to bother rolling back this transaction >which has not yet done a blessed thing to the database. > Correct. >Now, back to the experiment: > >Window1: delete from cust_tab where customer = "00002" >Window2: onstat -x >Transactions >address flags userthread locks log begin isolation retrys coordinator >c1fcba4c A-B-- c1fabf1c 5 157 CURSOR 0 > >Suddenly, log-begin is at uniq log sequence number 157. (Yes, that is my >current log at the moment.) Now something *has* happened to the >database. > >Why is this a problem for me? > >Well, the experiment started as a result of people asking if there was a >way to determine which logs are free-able under the newer scheme in >OnLine, wherein the F flag is never again set once a log file has been >used. (There are sound reasons for that which I do not dispute; it's >the side effect that causes the gluteal pain.) The script I was trying >to piece together involved > min(syslogs.begin) where begin > 0 >However, this discovery throws a monkey wrench into the query. > >If anyone knows a way around this, or if this behavior changes in later >releases of OnLine, please post a reply. There is no monkey wrench - looks like the begin work record only gets written to the logical log once some other record needs to be written. in-memory entry is created in transaction table to note that a transaction has started i.e. we are now in a transaction. However a statement which needs to be logged is executed then nothing is logged i.e. a begin work record is not yet written to the logical log. This means that you original calculation about which logical log is free is correct! PS I'll try onlog-ing this one tomorrow. -- David Williams