:

ORA-38701: Flashback database log, ORA-27070: async read/write failed

The Problem

I had created a guaranteed flashback restore point in my 11g (11.2.0.2 base release) database and I was running in some updates which were very intensive in terms of I/O resource. I should point out that this is on a Windows Server 2003 R2 OS. I noticed that there were disk queues showing in performance monitor of around 30 where the flashback files were stored. I then saw the following error in the alert log after I noticed that the instance had terminated.

ORA-38701: Flashback database log 131 seq 131 thread 1: “M:\DB\O1_MF_77NBRB6C_.FLB”
ORA-27070: async read/write failed
OSD-04008: WriteFile() failure, unable to write to file
O/S-Error: (OS 1450) Insufficient system resources exist to complete the requested service.
Wed Sep 21 19:21:05 2011
System state dump requested by (instance=1, osid=4112 (RVWR)), summary=[abnormal instance termination].
System State dumped to trace file D:\ORADATA\diag\rdbms\DB\trace\DB_diag_7820.trc
RVWR (ospid: 4112): terminating the instance due to error 38701

The Cause

Apparently there is a bug which requires an updated driver from EMC to fix this. It’s also reported that the cause is because Oracle is using an asynchronous method to write I/O to disk so even when the disk is not responding quickly enough it continues to throw more I/O at the disk which builds up and then it becomes so overloaded that it causes an error.

The Solution

As of yet, I don’t have one. I reverted to using SAN snapshot technology instead of flashback in the DB.

Like it, share it...

Category: Flashback


Related Posts

Leave a Reply

Your email address will not be published. Required fields are marked *