3 Replies Latest reply: Nov 25, 2012 8:56 PM by Nikolay Savvinov RSS

    rollback after a massive update/delete and "row cache objects latch"

    872581
      Hi, all.

      The oracle is 11.2.0.3 on a linux machine.

      I did a massive update/delete in PARALLEL mode on a table,
      and as expected, a lot of redo was generated.

      After the dml has completed, I issued "rollback" instead of "commit".
      And then, a number of sessions were waiting "row cache objects latch" during the rollback.

      "dc_rollback_segments" of v$rowcache indicates that there were a lot of getmisses.

      Could you explan why row cache objects latch is required when doing a rollback?

      Thanks in advance.
      Best Regards.
        • 1. Re: rollback after a massive update/delete and "row cache objects latch"
          rp0428
          >
          And then, a number of sessions were waiting "row cache objects latch" during the rollback.

          "dc_rollback_segments" of v$rowcache indicates that there were a lot of getmisses.
          >
          We can't comment on things we can't see. Is there some reason you don't want to post the information?

          Where did your info come from? Do you have an AWR? Post the queries you used to gather this information and the results of those queries using \
           tags to preserve the formatting.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                        
          • 2. Re: rollback after a massive update/delete and "row cache objects latch"
            Nikolay Savvinov
            Hi,

            no offense, but about 90% of reports of seeing abnormal waits result from confusion about Oracle Wait Interface. In particular, people often fail to understand that:

            1) V$SESSION_WAIT and other similar views display information not only for ongoing waits, but also for those that are already completed
            2) Running on CPU is not a wait

            As a result, when a session spends 99% running on CPU and 1% waiting on some exotic latch event, people may think that it was waiting on the latch the entire time.

            Without any evidence (like AWR, statspack, snapper output) how can we be sure that this is not the case here?

            Best regards,
            Nikolay
            • 3. Re: rollback after a massive update/delete and "row cache objects latch"
              sb92075
              869578 wrote:
              Hi, all.

              The oracle is 11.2.0.3 on a linux machine.

              I did a massive update/delete in PARALLEL mode on a table,
              and as expected, a lot of redo was generated.

              After the dml has completed, I issued "rollback" instead of "commit".
              And then, a number of sessions were waiting "row cache objects latch" during the rollback.

              "dc_rollback_segments" of v$rowcache indicates that there were a lot of getmisses.
              Which metric at what value does it go from being OK to being "a lot"?