Chunk copying is a mechanism for moving data from one table to another and is done in operations such as online alters and the TPP migration in Jakarta. On an Oracle database, the speed at which chunk copying is done can degrade over time causing these operations to take a long time to complete.
Steps to Reproduce
On an Oracle instance with a large CMDB table, 30M+ rows:
- Upgrade to Jakarta from a previous family triggering the TPP migration on the CMDB table
- Observe the amount of time it takes to copy chunks of data to the new table(s) slowly increase in processing time
Example log messages:
2017-09-20 15:56:46 (179) worker.4 worker.4 [0:00:00.022] Copied chunk 26 of 268 to tmp_t1581683948y[ts_query]
2017-09-20 15:56:46 (201) worker.4 worker.4 [0:00:00.022] Copied chunk 27 of 268 to tmp_t1581683948y[ts_query]
2017-09-20 15:56:46 (226) worker.4 worker.4 [0:00:00.025] Copied chunk 28 of 268 to tmp_t1581683948y[ts_query]
IMPORTANT: highlighted time is the amount of time it took to complete each chunk copy. 200ms is expected, but if impacted by this PRB these times slowly creep up as high as 20 seconds per each chunk.
This issue is under review. If you are able to upgrade, check the Fixed In field below to determine the versions that have a permanent fix. You can also subscribe to this known error article (click Subscribe button at the top of the article) to receive notifications when more information is available about this issue.
Related Problem: PRB1182304