P
Piotr Ryszkiewicz
Guest
Hello everyone, Recently we had strange issue at customer system, and I'd like to consult it with you. Here is the situation: - we have Java clients connecting to OE AppServer through Open Client - all communication between Java clients and AppServer goes through datasets - java clients are calling non-persistent procedures only - the first step at the beginning of each such procedure is to write complete input dataset content through dataset:write-xml method to longchar variable, which is immediately after that written into CLOB field in the database - at the end of processing output dataset is also written in the same way - this setup works without issues for few years already - minSrvInstance for AppServer process is set to 18, maxServerInstance to 30 (I will explain later why I think it's important) - AppServer is running in Stateless mode - OE 11.3.3 64-bit on AIX 7.1 Now, the problem: - some of blobs written into the database during specific 6 hours during one day in August do not contain XML as expected, but some weird binary data (looks like memory dump ?) - around 3.3% of records created during that period are corrupted this way - it never happened before nor later - in all these cases both clobs (input and output) are corrupted - datasets itself were transferred correctly - when we tried to call these procedures again with the same input data everything worked OK - there is nothing special in the appserver nor database logs, except one thing: shortly after the problem disappeared, two of appserver processes stopped and disconnected from the database (without any error message) My hypothesis: I found this KB entry: http://knowledgebase.progress.com/articles/Article/000031266 They mention memory leak in WRITE-XML method of dataset with NESTED data-relation which is our case. Maybe something bad happened to one of appserver processes because of this memory leak and since then it started to corrupt all write-xml outputs ? 3,3% of corrupted data tells that it can be 1 of of 30 processes. And when it stopped everything went back to normal ? Unfortunately at this point I cannot say if all corrupted records were written by the same appserver process, or not. I opened this case with technical support, but they didn't find the solution yet, and they don't think this problem is connected with the bug mentioned in KB. The bug itself is not solved yet. Of course I cannot reproduce the problem 
Any thoughts ? Regards, Piotr
Continue reading...
Any thoughts ? Regards, PiotrContinue reading...