-
Title
Delay in capture process -
Description
When one of Oracle Rac's nodes is idle or does not have a lot of transactions comparing to the other nodes, capture process falls behind Oracle at a certain interval.
This behavior can be monitored with show capture detail, and the delay can be different from time to time which can be few seconds to few minutes. This delay will be gone when a transaction occurs on idle nodes.
-
Cause
when shareplex works with Oracle RAC, capture was designed to delay when an activity node in RAC has very few transactions or generates no data.
There are few factors related to this design:
1. Capture has to process all nodes' redo log on the fly without waiting for a transaction to commit.
2. Although Oracle coordinates between their nodes about the most recently scn, each Oracle Rac writes in its redo log independently.Therefore, when there is no data from one node, capture needs to wait, sleep for some time, and retry to make sure it is truly have no data. Thus it is trying to avoid sending down data from other nodes too early so that sequencer starts sorting data based on scn. -
Resolution
Workaround:1. using a cron job to update a row in demo_src table (test transaction).
2. Force reader to make connection to idle node. This can be accomplished by: modifing tnsnames.ora to make sure the idle node IP address or hostname is the first entry on the tns alias.
-
Additional Information
There will be no fix for this delay, for this delay is related to the design of capture and Changing capture's may not benefit the capture performance.
Keyword search: Capture falls behind Oracle, capture delay, slow capture in Oracle RAC.