Commit causes rows to disappear [solved: rogue client]

Question

I just ran into the strangest thing.
Running XE 11g on Linux and Apex. My apex app was acting strange so I went into SQLDeveloper to run the raw commands that the apex app was supposed to be running.
I ran two insert statements, inserting data into a table (insert into a table using a select).
Then I did a select from the table to verify that the rows were there.
The two rows were there. They looked exactly as I had inserted them.
Then I issued a commit.
Guess what happened? My two rows were GONE!
I did this multiple times, each time I got the same result, the rows were gone AFTER the commit! There were no errors thrown in sqlDeveloper and nothing in the alert_XE.log.
Not knowing what else to do, I restarted the database (shutdown immediate; startup;). I reconnected and ran the exact same inserts and commits an it worked! The rows where there and my Apex app was working fine.
I've been an Oracle Programmer or DBA since 1994 and have never seen anything like this. The database has been up for 120 days and my only non-system generated tablespace is 20Meg.
The table in question has some unique constraints, but my statements didn't conflict and no errors were thrown. I'd try to open an SR, but there isn't a trace or anything to submit.
Anyone else seen this behavior in XE 11g?
Edited by: mcallister on Aug 24, 2012 8:29 AM

Answer 1

OS Details:
CentOS 5 x64 kernel 2.6.18-308.4.1.el5
Pentium 4 with 1 gig of RAM
The /u01 mount point has 73 gigs of free space and the tablespace the table is in is only 20 meg in size.
Oracle Database 11g Express Edition Release 11.2.0.2.0 - 64bit Production.
The only errors in the alert_XE.log are:
Sun Aug 19 03:00:00 2012
Errors in file /u01/app/oracle/diag/rdbms/xe/XE/trace/XE_j000_23033.trc:
ORA-12012: error on auto execute of job "SYS"."BSLN_MAINTAIN_STATS_JOB"
ORA-06550: line 1, column 807:
PLS-00201: identifier 'DBSNMP.BSLN_INTERNAL' must be declared
ORA-06550: line 1, column 807:
PL/SQL: Statement ignoredThe errors occur periodically and have been there since the database was created on 26-Feb-2012.
The table is NOT a global temporary table. No one has access to the database but me, everyone else uses the Apex app that I built. On 21-Aug-2012 a user of my site reported that he was unable to reserve a tee time for his golf league via the Apex application. I tried as well and got the same result. I went into the Apex app designer, found the pl/sql package that makes the reservation call, made the exact same call in SQLDeveloper and the row was inserted into the table. I issued a commit so that I could see the tee time reservation on the web site and it was NOT there. I did this repeatedly.
Confused, I eventually just wrote a single insert statement in SQL Developer to insert the rows manually.
1) I inserted one row, then changed the statement to avoid the unique constraint on the table.
2) inserted another row
3) then did a select on the table to see that both rows were there. They were both there.
4) then issued a commit followed by the same select as step 3. Both rows were GONE.
There was no error in the alert log. After I restarted the database, I repeated the same procedure as above with the exact same data and after the commit the rows were in the table just as they should be. The application is now working correctly.
Table details:
SELECT table_name, tablespace_name, temporary, duration
FROM all_tables
WHERE table_name = 'MATCH_TEE_TIMES';
TABLE_NAME                     TABLESPACE_NAME                TEMPORARY DURATION       
MATCH_TEE_TIMES                FACULTYSTAFFGOLF               N        Frankly, I'm very concerned. This is a data loss scenario with no outward signs of error. I've been running Oracle databases since 1994 and never seen anything like this. I run four mission-critical 11g Enterprise Edition databases in my professional life and my employer has hundreds of others that keep our university working. Granted none of them are XE, but if this is a common bug we would be in really big trouble.