Conference orarep::nomahs::rdb_60

Which screen is this?  Looks like this is history (since the Since is blank).

Ian

>>  <<< Note 5080.1 by NOVA::SMITHI "Don't understate or underestimate Rdb!" >>>

>>Which screen is this?  Looks like this is history (since the Since is blank).

The 'Since' being blank happens on the Active User Stall Messages screen.  
(The Stall Messages screen shows only active stalls).

This might happen if the disk the root file was on was extremely busy at 
the time all these users went to update TSNBLK information.  The fact that 
the since is blank is really okay since that means the stall is no longer 
stalled.  You're simply seeing the text from the last stall that was active 
at the time the stats displayed.

How often does this happen?  When everyone attaches?  Some other time?  All 
the time?  Are all the root files on the same shadow set?  Is the shadowed 
controller new (and whose is it)?

    /This might happen if the disk the root file was on was extremely busy
    /at the time all these users went to update TSNBLK information.  The fact
    /thatthe since is blank is really okay since that means the stall is no
    /longer stalled.  You're simply seeing the text from the last stall that was
    /active at the time the stats displayed.
     
    Ahh! forgot about that screen. He was having some locking problems 
    (caused his code) so I got confused. Now if he saw thoses messages
    (with timestamp) on the "STALL MESSAGES" screen, then, we would have
    a problem. ?
    
    thanks
    jay
    

>>     <<< Note 5080.3 by m5.us.oracle.com::JAKUHN "[email protected]" >>>
>>                       -< *ACTIVE* USER STALL != STALL  >-

>>    (caused his code) so I got confused. Now if he saw thoses messages
>>    (with timestamp) on the "STALL MESSAGES" screen, then, we would have
>>    a problem. ?
    
First, the Active User Stall Messages shows all users, stalled or not; the 
Stall Messages shows only active stalls.  So basically, the Stall Messages 
is a subset of the Active User Stall Messages screen.

The fact that you saw them at all does indicate a potential problem since 
the user did have to 'stall' or wait.  How long the stall is would give 
some indication of the level of problem.  Also remember, for the text to
even show up (even if not stalled) means that at one point during the 
refresh, the user was actively stalled and there was a time filled in (and 
just subsequently cleared on a later refresh).  So, some function of how 
significant the problem is will be based on the refresh rate.  The top of 
the screen should show that information.

    yes, and for some reason the messages cropped up in all his databases
    all at once, even though only one database was having problems with
    hangs. Maybe a controller problem or something.

    Their entire system is 4 months old. 
    They use HSD50s and HSD30s in a dual redundant configuration. 
    ie every HSD cabinet has 2 controllers in case one fails. 
    The databases are spread out across both types of contollers and I see 
    the "writing ROOT file (TSNBLK VBN xxx)" on all database regardless of 
    controller or shadow set.  They use HSD based shadow/mirroring. 
     
    Additionally, even databases that were not open at the time now show 
    this message.  He also checked unopened and opened databases from my 
    workstation(the only other member in the cluster with the production
    Alpha) 
    and his workstation also shows the "writing ROOT file (TSNBLK VBN xxx)" 
    message. The message is cluster wide regardless if a
    database was opened the other night when they had the problem. 
    ????
    thanks
    jay

    Gee Jay, try attaching to a db on your system.  Start/commit a txn, and
    see what the "Active User Stall Messages" screen says.
    
    In V7.0 all root "objects" are written to disk asynchronously. 
    Sometimes we don't do asynch because we want to make sure the updates
    get to disk before we proceed.  Thus we enter a stall msg before
    writing the object.  Since writing the TSNBLK turns out to be the last
    thing done in a txn then that is what you see in stats.