RDF/IMP, IMPX, and ZLT System Management Manual

Introducing RDF
HP NonStop RDF/IMP, IMPX, and ZLT System Management Manual—524388-002
1-25
Purger Process
performance rate would be seriously degraded. Therefore, the receiver keeps general
information about all transactions it has seen across all trails.
Second, because considerable checking must be done across all trails to determine
what files can be purged based on what transactions might be represented in the
various files on the various image trails, the purger process performs this task.
The purger process is a restartable process pair that runs on the backup system (it is
started during START RDF and runs even when the updaters are stopped; image files
are purged, however, only when updating is enabled).
No image file in a given image trail can be purged until it is absolutely clear that all
updaters configured to the trail will no longer require that file for an UNDO pass and
the file contains no transactions that might need to be undone for takeover or stop-
update-to-time operations. RDF automatically keeps track of which range of
transactions is represented in each image trail file. The purger process can therefore
always determine with confidence when a particular image trail file can be purged.
For example, assume the following:
•
There are two image trails.
•
Five updaters are assigned to each trail.
•
A long-running transaction (T1000) involves all five updaters on one trail, but none
on the other.
•
T1000 became active when the current image file in each trail was AA000002, and
is still active.
•
The receiver is currently writing to image file AA000015 in both trails.
•
All updaters are currently reading audit records from AA000015.
Although all the updater restart locations are in AA000015, none of the image files from
AA000002 through AA000014 can be purged while T1000 is active or aborting
because they will be required if T1000 needs to be backed out during an RDF takeover
or stop-update-to-timestamp operation. Note that this is true for both trails, even
though none of the updaters on one trail have ever been involved with T1000. If an
UNDO pass becomes necessary, all updaters must perform that pass in search of any
audit records associated with T1000 (they must go back in each image trail to the point
where T1000 began: AA000002 in this example).
The purger process exists to avoid having the receiver keep track of all this
information, which could impact extractor-receiver throughput significantly. The purger
process interacts with the updaters to determine when image files can be purged.