You are viewing a plain text version of this content. The canonical link for it is here.
Posted to derby-commits@db.apache.org by mi...@apache.org on 2007/11/24 16:47:04 UTC
svn commit: r597865 - in
/db/derby/code/trunk/java/engine/org/apache/derby/impl/store/access/btree:
BTreePostCommit.java BTreeScan.java
Author: mikem
Date: Sat Nov 24 07:47:03 2007
New Revision: 597865
URL: http://svn.apache.org/viewvc?rev=597865&view=rev
Log:
DERBY-3216
If btree post commit thread can not get table level lock for row cleanup and
possible tree merging, then instead attempt just row level row purging. Without
this change some stress tests were seeing post commit queue continuous grow
as there was always some thread with a lock on the index, and thus the items
could never get executed.
Modified:
db/derby/code/trunk/java/engine/org/apache/derby/impl/store/access/btree/BTreePostCommit.java
db/derby/code/trunk/java/engine/org/apache/derby/impl/store/access/btree/BTreeScan.java
Modified: db/derby/code/trunk/java/engine/org/apache/derby/impl/store/access/btree/BTreePostCommit.java
URL: http://svn.apache.org/viewvc/db/derby/code/trunk/java/engine/org/apache/derby/impl/store/access/btree/BTreePostCommit.java?rev=597865&r1=597864&r2=597865&view=diff
==============================================================================
--- db/derby/code/trunk/java/engine/org/apache/derby/impl/store/access/btree/BTreePostCommit.java (original)
+++ db/derby/code/trunk/java/engine/org/apache/derby/impl/store/access/btree/BTreePostCommit.java Sat Nov 24 07:47:03 2007
@@ -30,7 +30,9 @@
import org.apache.derby.iapi.store.access.AccessFactoryGlobals;
import org.apache.derby.iapi.store.access.ConglomerateController;
import org.apache.derby.iapi.store.access.DynamicCompiledOpenConglomInfo;
+import org.apache.derby.iapi.store.access.RowUtil;
import org.apache.derby.iapi.store.access.TransactionController;
+
import org.apache.derby.iapi.store.access.conglomerate.LogicalUndo;
import org.apache.derby.iapi.store.access.conglomerate.TransactionManager;
@@ -132,6 +134,54 @@
}
/**
+ * Open index for either table level or row level update.
+ * <p>
+ * @param lock_level For table level use TransactionManager.MODE_TABLE,
+ * for row level use TransactionManager.MODE_RECORD
+ * @param lock_mode For table level use LockingPolicy.MODE_CONTAINER,
+ * for row level use LockingPolicy.MODE_RECORD
+ *
+ * @exception StandardException Standard exception policy.
+ **/
+ private final OpenBTree openIndex(
+ TransactionManager internal_xact,
+ int lock_level,
+ int lock_mode)
+ throws StandardException
+ {
+ OpenBTree open_btree = new OpenBTree();
+
+ ConglomerateController base_cc =
+ btree.lockTable(
+ internal_xact,
+ (ContainerHandle.MODE_FORUPDATE |
+ ContainerHandle.MODE_LOCK_NOWAIT),
+ lock_level,
+ TransactionController.ISOLATION_REPEATABLE_READ);
+
+ open_btree.init(
+ (TransactionManager) null,
+ internal_xact,
+ (ContainerHandle) null, // open the container
+ internal_xact.getRawStoreXact(),
+ false,
+ (ContainerHandle.MODE_FORUPDATE | ContainerHandle.MODE_LOCK_NOWAIT),
+ lock_level,
+ btree.getBtreeLockingPolicy(
+ internal_xact.getRawStoreXact(),
+ lock_level,
+ lock_mode,
+ TransactionController.ISOLATION_REPEATABLE_READ,
+ base_cc,
+ open_btree),
+ btree,
+ (LogicalUndo) null, // No logical undo necessry.
+ (DynamicCompiledOpenConglomInfo) null);
+
+ return(open_btree);
+ }
+
+ /**
* perform the work described in the postcommit work.
* <p>
* In this implementation the only work that can be executed by this
@@ -165,51 +215,31 @@
System.out.println("starting internal xact\n");
}
- OpenBTree open_btree = new OpenBTree();
+ OpenBTree open_btree = null;
try
{
// Get lock on base table.
- // The current space reclamation algorithm requires a table level
- // lock on the btree - this is mostly because the shrink algorithm
- // is not multi-user. This lock is requested NOWAIT as it does
- // not want to impedede normal operation on the table. If the lock
- // were to wait then the current lock manager livelock algorithm
- // would block all subsequent lock requests on this btree even if
- // they are compatible with the current holder of the lock.
- //
- // There are currently 3 outstanding enhancement requests:
- // track 4237 - retry the work intelligently
- // track 4238 - if can't get table lock, at least reclaim the rows
- // track 4239 - do row level lock shrink - very hard to do.
+ // First attempt to get a table lock on the btree. This lock is
+ // requested NOWAIT to not impede normal operation on the table.
+ // If the lock were to wait then the current lock manager livelock
+ // algorithm would block all subsequent lock requests on this
+ // btree even if they are compatible with the current holder of
+ // the lock.
+ //
+ // If this lock is granted then:
+ // 1) deleted rows on the page can automatically be purged as
+ // they must be committed, otherwise lock would not have been
+ // granted.
+ // 2) if all rows from page are reclaimed then a structure shrink
+ // which requires table level lock can be executed.
//
- ConglomerateController base_cc =
- btree.lockTable(
+ open_btree =
+ openIndex(
internal_xact,
- (ContainerHandle.MODE_FORUPDATE |
- ContainerHandle.MODE_LOCK_NOWAIT),
- TransactionController.MODE_TABLE,
- TransactionController.ISOLATION_REPEATABLE_READ);
-
- open_btree.init(
- (TransactionManager) null,
- internal_xact,
- (ContainerHandle) null, // open the container
- internal_xact.getRawStoreXact(),
- false,
- ContainerHandle.MODE_FORUPDATE,
- TransactionController.MODE_TABLE,
- btree.getBtreeLockingPolicy(
- internal_xact.getRawStoreXact(),
- TransactionController.MODE_TABLE,
- LockingPolicy.MODE_CONTAINER,
- TransactionController.ISOLATION_REPEATABLE_READ,
- base_cc,
- open_btree),
- btree,
- (LogicalUndo) null, // No logical undo necessry.
- (DynamicCompiledOpenConglomInfo) null);
+ TransactionController.MODE_TABLE,
+ LockingPolicy.MODE_CONTAINER);
DataValueDescriptor[] shrink_key =
purgeCommittedDeletes(open_btree, this.page_number);
@@ -222,29 +252,46 @@
}
catch (StandardException se)
{
-
-
- //2 kinds of errors here expected here. Either container not found or dead lock.
+ // 2 kinds of errors here expected here. Either container not
+ // found or could not obtain lock (LOCK_TIMEOUT or DEADLOCK).
+ //
// It is possible by the time this post commit work gets scheduled
// that the container has been dropped and that the open container
// call will return null - in this case just return assuming no
// work to be done.
- //If it is a locking error, work is requeued. (4237)
-
if (se.getMessageId().equals(SQLState.LOCK_TIMEOUT) ||
se.getMessageId().equals(SQLState.DEADLOCK))
{
- requeue_work = true;
- }
+ // Could not get exclusive table lock, so try row level
+ // reclaim of just the rows on this page. No merge is
+ // attempted.
+
+ try
+ {
+ open_btree =
+ openIndex(
+ internal_xact,
+ TransactionController.MODE_RECORD,
+ LockingPolicy.MODE_RECORD);
+
+ purgeRowLevelCommittedDeletes(open_btree);
- //RESSOLVE-mike (4238) If you can't get a table level lock for btree space recovery in
- //the post commit thread, maybe you should at least reclaim the
- //rows on the page while you are at it. Use the same algorithm
- //as exists in BTreeController.java. row level shrink is still a
- //big problem and a separate track exists for it.
-
+ open_btree.close();
+ }
+ catch (StandardException se2)
+ {
+ if (se2.getMessageId().equals(SQLState.LOCK_TIMEOUT) ||
+ se2.getMessageId().equals(SQLState.DEADLOCK))
+ {
+ // Could not get intended exclusive table lock, so
+ // requeue and hope other user gives up table level
+ // lock soon. This should not be normal case.
+ requeue_work = true;
+ }
+ }
+ }
}
finally
{
@@ -383,6 +430,98 @@
}
return(shrink_key);
+ }
+
+ /**
+ * Attempt to reclaim committed deleted rows from the page with row locking.
+ * <p>
+ * 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.
+ *
+ * @param open_btree The already open btree, which has been locked with IX
+ * table lock, to use to get latch on page.
+ *
+ * @exception StandardException Standard exception policy.
+ **/
+ private final void purgeRowLevelCommittedDeletes(
+ OpenBTree open_btree)
+ throws StandardException
+ {
+ ControlRow controlRow = null;
+
+ try
+ {
+
+ if ((controlRow = ControlRow.get(open_btree, page_number)) == null)
+ return;
+
+ LeafControlRow leaf = (LeafControlRow) controlRow;
+
+ BTreeLockingPolicy btree_locking_policy =
+ open_btree.getLockingPolicy();
+
+ // The number records that can be reclaimed is:
+ // total recs - control row - recs_not_deleted
+ int num_possible_commit_delete =
+ leaf.page.recordCount() - 1 - leaf.page.nonDeletedRecordCount();
+
+ if ((num_possible_commit_delete > 0) &&
+ (btree_locking_policy.lockScanForReclaimSpace(leaf)))
+ {
+ DataValueDescriptor[] scratch_template =
+ open_btree.getRuntimeMem().get_template(
+ open_btree.getRawTran());
+
+ // Need to get an exclusive scan lock on the page before we can
+ // do any sort of purges, otherwise other concurrent scans would
+ // not work. If we can't get the lock NOWAIT, just give up on
+ // purging rows.
+ Page page = leaf.page;
+
+
+ // RowLocation column is in last column of template.
+ FetchDescriptor lock_fetch_desc =
+ RowUtil.getFetchDescriptorConstant(
+ scratch_template.length - 1);
+
+ // loop backward so that purges which affect the slot table
+ // don't affect the loop (ie. they only move records we
+ // have already looked at).
+ for (int slot_no = page.recordCount() - 1;
+ slot_no > 0;
+ slot_no--)
+ {
+ if (page.isDeletedAtSlot(slot_no))
+ {
+ // try to get an exclusive lock on the row, if we can
+ // then the row is a committed deleted row and it is
+ // safe to purge it.
+ if (btree_locking_policy.lockScanCommittedDeletedRow(
+ open_btree, leaf, scratch_template,
+ lock_fetch_desc, slot_no))
+ {
+ // the row is a committed deleted row, purge it.
+ page.purgeAtSlot(slot_no, 1, true);
+ }
+ }
+ }
+
+ }
+ }
+ finally
+ {
+ if (controlRow != null)
+ controlRow.release();
+
+ return;
+ }
}
}
Modified: db/derby/code/trunk/java/engine/org/apache/derby/impl/store/access/btree/BTreeScan.java
URL: http://svn.apache.org/viewvc/db/derby/code/trunk/java/engine/org/apache/derby/impl/store/access/btree/BTreeScan.java?rev=597865&r1=597864&r2=597865&view=diff
==============================================================================
--- db/derby/code/trunk/java/engine/org/apache/derby/impl/store/access/btree/BTreeScan.java (original)
+++ db/derby/code/trunk/java/engine/org/apache/derby/impl/store/access/btree/BTreeScan.java Sat Nov 24 07:47:03 2007
@@ -1366,7 +1366,8 @@
}
}
- if (SanityManager.DEBUG) {
+ if (SanityManager.DEBUG)
+ {
// DERBY-2197: Assume no row locking here. If locking policy
// requires row locking, we would need to obtain a row lock at
// this point.
@@ -1377,9 +1378,12 @@
}
if (scan_position.current_leaf.page.isDeletedAtSlot(
- scan_position.current_slot)) {
+ scan_position.current_slot))
+ {
ret_val = false;
- } else {
+ }
+ else
+ {
scan_position.current_leaf.page.deleteAtSlot(
scan_position.current_slot, true, this.btree_undo);
ret_val = true;
@@ -1387,16 +1391,18 @@
// See if we just deleted the last row on the page, in a btree a
// page with all rows still has 1 left - the control row.
- // Beetle 5750: we do not reclaim the root page of the btree if
- // there are no children since we were
- // doing too many post commit actions in a benchmark which does an
- // insert/commit/delete/commit operations in a single user system. now ,
- // with this change the work will move to the user
- // thread which does the insert
+ // Do not reclaim the root page of the btree if there are no
+ // children since we were doing too many post commit actions in a
+ // benchmark which does an insert/commit/delete/commit operations
+ // in a single user system. Now with this change the work will
+ // move to the user thread which does the insert and finds no space
+ // on the root page. In that case it will try a split, which
+ // automatically first checks if there is committed deleted space
+ // that can be reclaimed.
if (scan_position.current_leaf.page.nonDeletedRecordCount() == 1 &&
- !(scan_position.current_leaf.getIsRoot() &&
- scan_position.current_leaf.getLevel() == 0 ))
+ !(scan_position.current_leaf.getIsRoot() &&
+ scan_position.current_leaf.getLevel() == 0))
{
this.getXactMgr().addPostCommitWork(new BTreePostCommit(
this.getXactMgr().getAccessManager(),