Class BTreePostCommit
- java.lang.Object
-
- org.apache.derby.impl.store.access.btree.BTreePostCommit
-
- All Implemented Interfaces:
Serviceable
class BTreePostCommit extends java.lang.Object implements Serviceable
The BTreePostCommit class implements the Serviceable protocol. In it's role as a Serviceable object, it stores the state necessary to find a page in a btree that may have committed delete's to reclaim. In it's role as a PostCommitProcessor it looks up the page described, and reclaims space in the btree. It first trys to clean up any deleted commits on the page. It then will shrink the tree if it is going to delete all rows from the page (RESOLVE - not done yet).
-
-
Field Summary
Fields Modifier and Type Field Description private AccessFactory
access_factory
protected BTree
btree
private long
page_number
-
Fields inherited from interface org.apache.derby.iapi.services.daemon.Serviceable
DONE, REQUEUE
-
-
Constructor Summary
Constructors Constructor Description BTreePostCommit(AccessFactory access_factory, BTree btree, long input_page_number)
-
Method Summary
All Methods Instance Methods Concrete Methods Modifier and Type Method Description private void
doShrink(OpenBTree open_btree, DataValueDescriptor[] shrink_row)
private DataValueDescriptor[]
getShrinkKey(OpenBTree open_btree, ControlRow control_row, int slot_no)
private OpenBTree
openIndex(TransactionManager internal_xact, int lock_level, int lock_mode)
Open index for either table level or row level update.int
performWork(ContextManager contextMgr)
perform the work described in the postcommit work.private DataValueDescriptor[]
purgeCommittedDeletes(OpenBTree open_btree, long pageno)
Reclaim space taken up by committed deleted rows.private void
purgeRowLevelCommittedDeletes(OpenBTree open_btree)
Attempt to reclaim committed deleted rows from the page with row locking.boolean
serviceASAP()
The urgency of this post commit work.boolean
serviceImmediately()
If this work should be done immediately on the user thread then return true.
-
-
-
Field Detail
-
access_factory
private AccessFactory access_factory
-
page_number
private long page_number
-
btree
protected BTree btree
-
-
Constructor Detail
-
BTreePostCommit
BTreePostCommit(AccessFactory access_factory, BTree btree, long input_page_number)
-
-
Method Detail
-
serviceASAP
public boolean serviceASAP()
The urgency of this post commit work.This determines where this Serviceable is put in the post commit queue. Post commit work in the btree can be safely delayed until there is not user work to do.
- Specified by:
serviceASAP
in interfaceServiceable
- Returns:
- false, this work should not be serviced ASAP
-
serviceImmediately
public boolean serviceImmediately()
Description copied from interface:Serviceable
If this work should be done immediately on the user thread then return true. If it doesn't make any difference if this work is done on a the user thread immediately or if it is performed by another thread asynchronously later, then return false.- Specified by:
serviceImmediately
in interfaceServiceable
-
doShrink
private final void doShrink(OpenBTree open_btree, DataValueDescriptor[] shrink_row) throws StandardException
- Throws:
StandardException
-
openIndex
private final OpenBTree openIndex(TransactionManager internal_xact, int lock_level, int lock_mode) throws StandardException
Open index for either table level or row level update.- Parameters:
lock_level
- For table level use TransactionManager.MODE_TABLE, for row level use TransactionManager.MODE_RECORDlock_mode
- For table level use LockingPolicy.MODE_CONTAINER, for row level use LockingPolicy.MODE_RECORD- Throws:
StandardException
- Standard exception policy.
-
performWork
public int performWork(ContextManager contextMgr) throws StandardException
perform the work described in the postcommit work.In this implementation the only work that can be executed by this post commit processor is this class itself.
- Specified by:
performWork
in interfaceServiceable
- Parameters:
contextMgr
- the context manager started by the post commit daemon- Returns:
- Returns Serviceable.DONE when work has completed, or returns Serviceable.REQUEUE if work needs to be requeued.
- Throws:
StandardException
- Standard exception policy.
-
getShrinkKey
private final DataValueDescriptor[] getShrinkKey(OpenBTree open_btree, ControlRow control_row, int slot_no) throws StandardException
- Throws:
StandardException
-
purgeCommittedDeletes
private final DataValueDescriptor[] purgeCommittedDeletes(OpenBTree open_btree, long pageno) throws StandardException
Reclaim space taken up by committed deleted rows.This routine assumes it has been called by an internal transaction which has performed no work so far, and that it has an exclusive table lock. These assumptions mean that any deleted rows encountered must be from committed transactions (otherwise we could not have gotten the exclusive table lock).
This routine handles purging committed deletes while holding a table level exclusive lock. See purgeRowLevelCommittedDeletes() for row level purging.
- Parameters:
open_btree
- The btree already opened.pageno
- The page number of the page to look for committed deletes.- Throws:
StandardException
- Standard exception policy.
-
purgeRowLevelCommittedDeletes
private final void purgeRowLevelCommittedDeletes(OpenBTree open_btree) throws StandardException
Attempt to reclaim committed deleted rows from the page with row locking.Get exclusive latch on page, and then loop backward through page searching for deleted rows which are committed. This routine is called only from post commit processing so it will never see rows deleted by the current transaction. For each deleted row on the page it attempts to get an exclusive lock on the deleted row, NOWAIT. If it succeeds, and since this transaction did not delete the row then the row must have been deleted by a transaction which has committed, so it is safe to purge the row. It then purges the row from the page.
The latch on the leaf page containing the purged rows must be kept until after the transaction has been committed or aborted in order to insure proper undo of the purges can take place. Otherwise another transaction could use the space freed by the purge and then prevent the purge from being able to undo.
- Parameters:
open_btree
- The already open btree, which has been locked with IX table lock, to use to get latch on page.- Throws:
StandardException
- Standard exception policy.
-
-