Long Checkpoints during an archive that occurs during blob ( text and byte ) processing
You see long checkpoints during an archive / backup that occurs during blobspace blob ( text and byte ) processing.
Checkpoint duration is longer during archives / backups that occurs during blobspace blob ( text and byte ) processing.
The archive process blocks blobspace alteration until a blobspace is backed up. Examples of blobspace alteration would be dropping a blob or inserting a blob. Checkpoint duration is drawn out longer because sessions trying to alter the blobspace are blocked until the blobspace is backed up.
Although blobspace blobs are not logged, the blob chunk free-list page is logged and this is why the blobspace must be locked during an archive.
Note: There could be other causes.
Diagnosing the problem
onstat -X will show a buffer with a pgflags of 80d. 80d refers to the blob chunk free-list page
Resolving the problem
Try to minimize the amount of planned blobspace blob activity occurring during an archive. To workaround this problem you can list out all the spaces that are to be backed up with the blobspaces listed immediately after the critical dbspaces ( rootdbs, logical log dbspace, physical log dbspace ) so they are backed up and the lock can be released sooner. Make sure the list includes ALL spaces so you have a complete backup.
More support for:
Software version: 7.3, 9.4, 10.0, 11.1, 11.5, 11.7, 12.1
Operating system(s): AIX, HP-UX, Linux, OS X, Solaris, Windows
Reference #: 1631859
Modified date: 26 June 2014