Forum Stats

  • 3,727,076 Users
  • 2,245,317 Discussions
  • 7,852,569 Comments

Discussions

log file sync and block recovery from logseq

 Hi all,

sometimes log file sync event spike on our 11.2.0.4 db on Solaris.

when we look at alert log at the same time we see following messages.


Block recovery from logseq 1462532, block 36595759 to scn 3522018885687

Recovery of Online Redo Log: Thread 1 Group 59 Seq 1462532 Reading mem 0

 Mem# 0: +REDO/DBNAME/onlinelog/group_59.272.1047579261

 Mem# 1: +REDO/DBNAME/onlinelog/group_59.356.1047581307

Block recovery completed at rba 1462532.36596324.16, scn 820.145702968

Block recovery from logseq 1462532, block 36594513 to scn 3522018885686

Recovery of Online Redo Log: Thread 1 Group 59 Seq 1462532 Reading mem 0

 Mem# 0: +REDO/DBNAME/onlinelog/group_59.272.1047579261

 Mem# 1: +REDO/DBNAME/onlinelog/group_59.356.1047581307

Block recovery completed at rba 1462532.36596324.16, scn 820.145702967

 

there is no message on lgwr trace and wait chains are below.


%This   SECONDS   AAS WAIT_CHAIN                        

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

 32%    2985  96.3 -> log file sync -> LGWR wait for redo copy -> ON CPU 

 26%    2421  78.1 -> ON CPU                        

 21%    1911  61.6 -> log file sync -> log file parallel write       


Do you have reason for that?

Regards.

  

Answers

  • Jonathan Lewis
    Jonathan Lewis Member Posts: 9,573 Gold Crown
    edited November 2020


    Do you see ANY trace files around the time of the event?

    It looks like it might be a case of the buffered copy of a block being corrupt and Oracle doing an in-memory block recovery by reading the block from disc into the buffer cache than finding and applying any redo for that block to bring the buffer up to date.

    If that's correct then it's possible there's an Oracle bug corrupting blocks occasionally, or you have a memory fault which is corrupting Oracle blocks.

    Regards

    Jonathan Lewis


    UPDATE: The last couple of paragraphs of this note of mine make a few comments about the mechanism.

    User_96XRQ
  • Jonathan Lewis
    Jonathan Lewis Member Posts: 9,573 Gold Crown

    Another follow-up (for the same blog note). If you read the conversation starting at comment #9 it suggests that your alert log ought to have some details about the corrupt block just before the bit you've supplied. (So it if doesn't there's presumably some other reason for what's going on.)


    Regards

    Jonathan Lewis

  • User_96XRQ
    User_96XRQ Member Posts: 3

    Hi Jonathan,

    thanks for reply. We did not see any trace files at that time.

    Regards.

Sign In or Register to comment.