Home » db-derby-10.5.3.0 » org.apache.derby.iapi.store.access » [javadoc | source]
org.apache.derby.iapi.store.access
public interface: TransactionController [javadoc | source]

All Implemented Interfaces:
    PersistentSet

All Known Implementing Classes:
    XATransactionController, TransactionManager, NoOpTransaction, RAMTransaction

The TransactionController interface provides methods that an access client can use to control a transaction, which include the methods for gaining access to resources (conglomerates, scans, etc.) in the transaction controller's storage manager. TransactionControllers are obtained from an AccessFactory via the getTransaction method.

Each transaction controller is associated with a transaction context which provides error cleanup when standard exceptions are thrown anywhere in the system. The transaction context performs the following actions in response to cleanupOnError:

Transactions are obtained from an AccessFactory.
Field Summary
static final  int MODE_RECORD    Constant used for the lock_level argument to openConglomerate() and openScan() calls. Pass in MODE_RECORD if you want the conglomerate to be opened with record level locking (but the system may override this choice and provide table level locking instead). 
static final  int MODE_TABLE    Constant used for the lock_level argument to openConglomerate() and openScan() calls. Pass in MODE_TABLE if you want the conglomerate to be opened with table level locking - if this mode is passed in the system will never use record level locking for the open scan or controller. 
static final  int ISOLATION_NOLOCK    No locks are requested for data that is read only. Uncommitted data may be returned. Writes only visible previous to commit. Exclusive transaction length locks are set on data that is written, no lock is set on data that is read. No table level intent lock is held so it is up to caller to insure that table is not dropped while being accessed (RESOLVE - this issue may need to be resolved differently if we can't figure out a non-locked based way to prevent ddl during read uncommitted access). ONLY USED INTERNALLY BY ACCESS, NOT VALID FOR EXTERNAL USERS. 
static final  int ISOLATION_READ_UNCOMMITTED    No locks are requested for data that is read only. Uncommitted data may be returned. Writes only visible previous to commit. Exclusive transaction length locks are set on data that is written, no lock is set on data that is read. No table level intent lock is held so it is up to caller to insure that table is not dropped while being accessed (RESOLVE - this issue may need to be resolved differently if we can't figure out a non-locked based way to prevent ddl during read uncommitted access). Note that this is currently only supported in heap scans. TODO - work in progress to support this locking mode in the 5.1 storage system. 
static final  int ISOLATION_READ_COMMITTED    No lost updates, no dirty reads, only committed data is returned. Writes only visible when committed. Exclusive transaction length locks are set on data that is written, short term locks ( possibly instantaneous duration locks) are set on data that is read. 
static final  int ISOLATION_READ_COMMITTED_NOHOLDLOCK    No lost updates, no dirty reads, only committed data is returned. Writes only visible when committed. Exclusive transaction length locks are set on data that is written, short term locks ( possibly instantaneous duration locks) are set on data that is read. Read locks are requested for "zero" duration, thus upon return from access no read row lock is held. 
static final  int ISOLATION_REPEATABLE_READ    Read and write locks are held until end of transaction, but no phantom protection is performed (ie no previous key locking). Writes only visible when committed. Note this constant is currently mapped to ISOLATION_SERIALIZABLE. The constant is provided so that code which only requires repeatable read can be coded with the right isolation level, and will just work when store provided real repeatable read isolation. 
static final  int ISOLATION_SERIALIZABLE    Gray's isolation degree 3, "Serializable, Repeatable Read". Note that some conglomerate implementations may only be able to provide phantom protection under MODE_TABLE, while others can support this under MODE_RECORD. * 
static final  int OPENMODE_USE_UPDATE_LOCKS    Use this mode to the openScan() call to indicate the scan should get update locks during scan, and either promote the update locks to exclusive locks if the row is changed or demote the lock if the row is not updated. The lock demotion depends on the isolation level of the scan. If isolation level is ISOLATION_SERIALIZABLE or ISOLATION_REPEATABLE_READ then the lock will be converted to a read lock. If the isolation level ISOLATION_READ_COMMITTED then the lock is released when the scan moves off the row.

Note that one must still set OPENMODE_FORUPDATE to be able to change rows in the scan. So to enable update locks for an updating scan one provides (OPENMODE_FORUPDATE | OPENMODE_USE_UPDATE_LOCKS) 

static final  int OPENMODE_SECONDARY_LOCKED    Use this mode to the openConglomerate() call which opens the base table to be used in a index to base row probe. This will cause the openConglomerate() call to not get any row locks as part of it's fetches. It is important when using this mode that the secondary index table be successfully opened before opening the base table so that proper locking protocol is followed. 
static final  int OPENMODE_BASEROW_INSERT_LOCKED    Use this mode to the openConglomerate() call used to open the secondary indices of a table for inserting new rows in the table. This will let the secondaryindex know that the base row being inserted has already been locked and only previous key locks need be obtained. It is important when using this mode that the base table be successfully opened before opening the secondaryindex so that proper locking protocol is followed. 
static final  int OPENMODE_FORUPDATE    open table for update, if not specified table will be opened for read. 
static final  int OPENMODE_FOR_LOCK_ONLY    Use this mode to the openConglomerate() call used to just get the table lock on the conglomerate without actually doing anything else. Any operations other than close() performed on the "opened" container will fail. * 
static final  int OPENMODE_LOCK_NOWAIT    The table lock request will not wait.

The request to get the table lock (any table lock including intent or "real" table level lock), will not wait if it can't be granted. A lock timeout will be returned. Note that subsequent row locks will wait if the application has not set a 0 timeout and if the call does not have a wait parameter (like OpenConglomerate.fetch(). 

public static final  int OPEN_CONGLOMERATE    Constants used for the countOpen() call. * 
public static final  int OPEN_SCAN     
public static final  int OPEN_CREATED_SORTS     
public static final  int OPEN_SORT     
public static final  int OPEN_TOTAL     
static final  byte IS_DEFAULT     
static final  byte IS_TEMPORARY     
static final  byte IS_KEPT     
public final  int RELEASE_LOCKS     
public final  int KEEP_LOCKS     
public final  int READONLY_TRANSACTION_INITIALIZATION     
Method from org.apache.derby.iapi.store.access.TransactionController Summary:
abort,   addColumnToConglomerate,   anyoneBlocked,   commit,   commitNoSync,   compressConglomerate,   conglomerateExists,   countOpens,   createAndLoadConglomerate,   createBackingStoreHashtableFromScan,   createConglomerate,   createSort,   createXATransactionFromLocalTransaction,   debugOpened,   defragmentConglomerate,   destroy,   dropConglomerate,   dropSort,   fetchMaxOnBtree,   findConglomid,   findContainerid,   getAccessManager,   getActiveStateTxIdString,   getContextManager,   getDynamicCompiledConglomInfo,   getFileHandler,   getLockSpace,   getStaticCompiledConglomInfo,   getTransactionIdString,   getUserCreateConglomPropList,   isGlobal,   isIdle,   isPristine,   logAndDo,   openCompiledConglomerate,   openCompiledScan,   openConglomerate,   openGroupFetchScan,   openScan,   openSort,   openSortCostController,   openSortRowSource,   openSortScan,   openStoreCost,   purgeConglomerate,   recreateAndLoadConglomerate,   releaseSavePoint,   rollbackToSavePoint,   setNoLockWait,   setSavePoint,   startNestedUserTransaction
Method from org.apache.derby.iapi.store.access.TransactionController Detail:
 public  void abort() throws StandardException
    Abort all changes made by this transaction since the last commit, abort or the point the transaction was started, whichever is the most recent. All savepoints within this transaction are released, and all resources are released (held or non-held).
 public  void addColumnToConglomerate(long conglomId,
    int column_id,
    Storable template_column,
    int collation_id) throws StandardException
    Add a column to a conglomerate. The Storage system will block this action until it can get an exclusive container level lock on the conglomerate. The conglomerate must not be open in the current transaction, this means that within the current transaction there must be no open ConglomerateController's or ScanControllers. It may not be possible in some implementations of the system to catch this error in the store, so it is up to the caller to insure this. The column can only be added at the spot just after the current set of columns. The template_column must be nullable. After this call has been made, all fetches of this column from rows that existed in the table prior to this call will return "null".
 public boolean anyoneBlocked()
    Return true if any transaction is blocked (even if not by this one).
 public  void commit() throws StandardException
    Commit this transaction. All savepoints within this transaction are released. All non-held conglomerates and scans are closed.
 public DatabaseInstant commitNoSync(int commitflag) throws StandardException
    "Commit" this transaction without sync'ing the log. Everything else is identical to commit(), use this at your own risk.
    bits in the commitflag can turn on to fine tuned the "commit": KEEP_LOCKS - no locks will be released by the commit and no post commit processing will be initiated. If, for some reasons, the locks cannot be kept even if this flag is set, then the commit will sync the log, i.e., it will revert to the normal commit. READONLY_TRANSACTION_INITIALIZATION - Special case used for processing while creating the transaction. Should only be used by the system while creating the transaction to commit readonly work that may have been done using the transaction while getting it setup to be used by the user. In the future we should instead use a separate tranaction to do this initialization. Will fail if called on a transaction which has done any updates.
 public  void compressConglomerate(long conglomId) throws StandardException
    Return free space from the conglomerate back to the OS.

    Returns free space from the conglomerate back to the OS. Currently only the sequential free pages at the "end" of the conglomerate can be returned to the OS.

 public boolean conglomerateExists(long conglomId) throws StandardException
    Check whether a conglomerate exists.
 public int countOpens(int which_to_count) throws StandardException
    Report on the number of open conglomerates in the transaction.

    There are 4 types of open "conglomerates" that can be tracked, those opened by each of the following: openConglomerate(), openScan(), createSort(), and openSort(). Scans opened by openSortScan() are tracked the same as those opened by openScan(). This routine can be used to either report on the number of all opens, or may be used to track one particular type of open.

    This routine is expected to be used for debugging only. An implementation may only track this info under SanityManager.DEBUG mode. If the implementation does not track the info it will return -1 (so code using this call to verify that no congloms are open should check for return <= 0 rather than == 0).

    The return value depends on the "which_to_count" parameter as follows:

    • OPEN_CONGLOMERATE - return # of openConglomerate() calls not close()'d.
    • OPEN_SCAN - return # of openScan() + openSortScan() calls not close()'d.
    • OPEN_CREATED_SORTS - return # of sorts created (createSort()) in current xact. There is currently no way to get rid of these sorts before end of transaction.
    • OPEN_SORT - return # of openSort() calls not close()'d.
    • OPEN_TOTAL - return total # of all above calls not close()'d.
    - note an implementation may return -1 if it does not track the above information.

 public long createAndLoadConglomerate(String implementation,
    DataValueDescriptor[] template,
    ColumnOrdering[] columnOrder,
    int[] collationIds,
    Properties properties,
    int temporaryFlag,
    RowLocationRetRowSource rowSource,
    long[] rowCount) throws StandardException
    Create a conglomerate and load (filled) it with rows that comes from the row source without loggging.

    Individual rows that are loaded into the conglomerate are not logged. After this operation, the underlying database must be backed up with a database backup rather than an transaction log backup (when we have them). This warning is put here for the benefit of future generation.

    This function behaves the same as @see createConglomerate except it also populates the conglomerate with rows from the row source and the rows that are inserted are not logged.

 public BackingStoreHashtable createBackingStoreHashtableFromScan(long conglomId,
    int open_mode,
    int lock_level,
    int isolation_level,
    FormatableBitSet scanColumnList,
    DataValueDescriptor[] startKeyValue,
    int startSearchOperator,
    Qualifier[][] qualifier,
    DataValueDescriptor[] stopKeyValue,
    int stopSearchOperator,
    long max_rowcnt,
    int[] key_column_numbers,
    boolean remove_duplicates,
    long estimated_rowcnt,
    long max_inmemory_rowcnt,
    int initialCapacity,
    float loadFactor,
    boolean collect_runtimestats,
    boolean skipNullKeyColumns,
    boolean keepAfterCommit) throws StandardException
    Create a HashSet which contains all rows that qualify for the described scan.

    All parameters shared between openScan() and this routine are interpreted exactly the same. Logically this routine calls openScan() with the passed in set of parameters, and then places all returned rows into a newly created HashSet and returns, actual implementations will likely perform better than actually calling openScan() and doing this. For documentation of the openScan parameters see openScan().

 public long createConglomerate(String implementation,
    DataValueDescriptor[] template,
    ColumnOrdering[] columnOrder,
    int[] collationIds,
    Properties properties,
    int temporaryFlag) throws StandardException
    Create a conglomerate.

    Currently, only "heap"'s and ""btree secondary index"'s are supported, and all the features are not completely implemented. For now, create conglomerates like this:

    TransactionController tc;
    long conglomId = tc.createConglomerate(
    "heap", // we're requesting a heap conglomerate
    template, // a populated template is required for heap and btree.
    null, // no column order
    null, // default collation order for all columns
    null, // default properties
    0); // not temporary
    
Each implementation of a conglomerate takes a possibly different set of properties. The "heap" implementation currently takes no properties. The "btree secondary index" requires the following set of properties:
  • "baseConglomerateId" (integer). The conglomerate id of the base conglomerate is never actually accessed by the b-tree secondary index implementation, it only serves as a namespace for row locks. This property is required.
  • "rowLocationColumn" (integer). The zero-based index into the row which the b-tree secondary index will assume holds a @see RowLocation of the base row in the base conglomerate. This value will be used for acquiring locks. In this implementation RowLocationColumn must be the last key column. This property is required.
  • "allowDuplicates" (boolean). If set to true the table will allow rows which are duplicate in key column's 0 through (nUniqueColumns - 1). Currently only supports "false". This property is optional, defaults to false.
  • "nKeyFields" (integer) Columns 0 through (nKeyFields - 1) will be included in key of the conglomerate. This implementation requires that "nKeyFields" must be the same as the number of fields in the conglomerate, including the rowLocationColumn. Other implementations may relax this restriction to allow non-key fields in the index. This property is required.
  • "nUniqueColumns" (integer) Columns 0 through "nUniqueColumns" will be used to check for uniqueness. So for a standard SQL non-unique index implementation set "nUniqueColumns" to the same value as "nKeyFields"; and for a unique index set "nUniqueColumns" to "nKeyFields - 1 (ie. don't include the rowLocationColumn in the uniqueness check). This property is required.
  • "maintainParentLinks" (boolean) Whether the b-tree pages maintain the page number of their parent. Only used for consistency checking. It takes a certain amount more effort to maintain these links, but they're really handy for ensuring that the index is consistent. This property is optional, defaults to true.
A secondary index i (a, b) on table t (a, b, c) would have rows which looked like (a, b, row_location). baseConglomerateId is set to the conglomerate id of t. rowLocationColumns is set to 2. allowsDuplicates would be set to false. To create a unique secondary index set uniquenessColumns to 2, this means that the btree code will compare the key values but not the row id when determing uniqueness. To create a nonunique secondary index set uniquenessColumns to 3, this would mean that the uniqueness test would include the row location and since all row locations will be unique all rows inserted into the index will be differentiated (at least) by row location.
 public long createSort(Properties implParameters,
    DataValueDescriptor[] template,
    ColumnOrdering[] columnOrdering,
    SortObserver sortObserver,
    boolean alreadyInOrder,
    long estimatedRows,
    int estimatedRowSize) throws StandardException
    Create a sort. Rows are inserted into the sort with a sort controller, and subsequently retrieved with a sort scan controller. The rows come out in the order specified by the parameters.

    Sorts also do aggregation. The input (unaggregated) rows have the same format as the aggregated rows, and the aggregate results are part of the both rows. The sorter, when it notices that a row is a duplicate of another, calls a user-supplied aggregation method (see interface Aggregator), passing it both rows. One row is known as the 'addend' and the other the 'accumulator'. The aggregation method is assumed to merge the addend into the accumulator. The sort then discards the addend row.

    So, for the query:

    select a, sum(b) from t group by a
    The input row to the sorter would have one column for a and another column for sum(b). It is up to the caller to get the format of the row correct, and to initialize the aggregate values correctly (null for most aggregates, 0 for count).

    Nulls are always considered to be ordered in a sort, that is, null compares equal to null, and less than anything else.

 public Object createXATransactionFromLocalTransaction(int format_id,
    byte[] global_id,
    byte[] branch_id) throws StandardException
    Convert a local transaction to a global transaction.

    Get a transaction controller with which to manipulate data within the access manager. Tbis controller allows one to manipulate a global XA conforming transaction.

    Must only be called a previous local transaction was created and exists in the context. Can only be called if the current transaction is in the idle state. Upon return from this call the old tc will be unusable, and all references to it should be dropped (it will have been implicitly destroy()'d by this call.

    The (format_id, global_id, branch_id) triplet is meant to come exactly from a javax.transaction.xa.Xid. We don't use Xid so that the system can be delivered on a non-1.2 vm system and not require the javax classes in the path.

 public String debugOpened() throws StandardException
    Return a string with debug information about opened congloms/scans/sorts.

    Return a string with debugging information about current opened congloms/scans/sorts which have not been close()'d. Calls to this routine are only valid under code which is conditional on SanityManager.DEBUG.

 public GroupFetchScanController defragmentConglomerate(long conglomId,
    boolean online,
    boolean hold,
    int open_mode,
    int lock_level,
    int isolation_level) throws StandardException
    Compress table in place.

    Returns a GroupFetchScanController which can be used to move rows around in a table, creating a block of free pages at the end of the table. The process will move rows from the end of the table toward the beginning. The GroupFetchScanController will return the old row location, the new row location, and the actual data of any row moved. Note that this scan only returns moved rows, not an entire set of rows, the scan is designed specifically to be used by either explicit user call of the SYSCS_ONLINE_COMPRESS_TABLE() procedure, or internal background calls to compress the table. The old and new row locations are returned so that the caller can update any indexes necessary. This scan always returns all collumns of the row. All inputs work exactly as in openScan(). The return is a GroupFetchScanController, which only allows fetches of groups of rows from the conglomerate.

 public  void destroy()
    Abort the current transaction and pop the context. *
 public  void dropConglomerate(long conglomId) throws StandardException
    Drop a conglomerate. The conglomerate must not be open in the current transaction. This also means that there must not be any active scans on it.
 public  void dropSort(long sortid) throws StandardException
    Drop a sort.

    Drop a sort created by a call to createSort() within the current transaction (sorts are automatically "dropped" at the end of a transaction. This call should only be made after all openSortScan()'s and openSort()'s have been closed.

 public boolean fetchMaxOnBtree(long conglomId,
    int open_mode,
    int lock_level,
    int isolation_level,
    FormatableBitSet scanColumnList,
    DataValueDescriptor[] fetchRow) throws StandardException
    Retrieve the maximum value row in an ordered conglomerate.

    Returns true and fetches the rightmost non-null row of an ordered conglomerate into "fetchRow" if there is at least one non-null row in the conglomerate. If there are no non-null rows in the conglomerate it returns false. Any row with a first column with a Null is considered a "null" row.

    Non-ordered conglomerates will not implement this interface, calls will generate a StandardException.

    RESOLVE - this interface is temporary, long term equivalent (and more) functionality will be provided by the openBackwardScan() interface.

    ISOLATION_SERIALIZABLE and MODE_RECORD locking for btree max: The "BTREE" implementation will at the very least get a shared row lock on the max key row and the key previous to the max. This will be the case where the max row exists in the rightmost page of the btree. These locks won't be released. If the row does not exist in the last page of the btree then a scan of the entire btree will be performed, locks acquired in this scan will not be released.

    Note that under ISOLATION_READ_COMMITTED, all locks on the table are released before returning from this call.

 public long findConglomid(long containerid) throws StandardException
    For debugging, find the conglomid given the containerid.

 public long findContainerid(long conglomid) throws StandardException
    For debugging, find the containerid given the conglomid.

    Will have to change if we ever have more than one container in a conglomerate.

 public AccessFactory getAccessManager()
    Get reference to access factory which started this transaction.

 public String getActiveStateTxIdString()
    Get string id of the transaction that would be when the Transaction is IN active state. This method increments the Tx id of current Tx object if it is in idle state. Note: Use this method only getTransactionIdString() is not suitable.
 public ContextManager getContextManager()
    Get the context manager that the transaction was created with.

 public DynamicCompiledOpenConglomInfo getDynamicCompiledConglomInfo(long conglomId) throws StandardException
    Return dynamic information about the conglomerate to be dynamically reused in repeated execution of a statement.

    The dynamic info is a set of variables to be used in a given ScanController or ConglomerateController. It can only be used in one controller at a time. It is up to the caller to insure the correct thread access to this info. The type of info in this is a scratch template for btree traversal, other scratch variables for qualifier evaluation, ...

 public FileResource getFileHandler()
    Get an object to handle non-transactional files.
 public CompatibilitySpace getLockSpace()
    Return an object that when used as the compatibility space for a lock request, and the group object is the one returned by a call to getOwner() on that object, guarantees that the lock will be removed on a commit or an abort.
 public StaticCompiledOpenConglomInfo getStaticCompiledConglomInfo(long conglomId) throws StandardException
    Return static information about the conglomerate to be included in a a compiled plan.

    The static info would be valid until any ddl was executed on the conglomid, and would be up to the caller to throw away when that happened. This ties in with what language already does for other invalidation of static info. The type of info in this would be containerid and array of format id's from which templates can be created. The info in this object is read only and can be shared among as many threads as necessary.

 public String getTransactionIdString()
    Get string id of the transaction.

    This transaction "name" will be the same id which is returned in the TransactionInfo information, used by the lock and transaction vti's to identify transactions.

    Although implementation specific, the transaction id is usually a number which is bumped every time a commit or abort is issued.

 public Properties getUserCreateConglomPropList()
    A superset of properties that "users" can specify.

    A superset of properties that "users" (ie. from sql) can specify. Store may implement other properties which should not be specified by users. Layers above access may implement properties which are not known at all to Access.

    This list is a superset, as some properties may not be implemented by certain types of conglomerates. For instant an in-memory store may not implement a pageSize property. Or some conglomerates may not support pre-allocation.

    This interface is meant to be used by the SQL parser to do validation of properties passsed to the create table statement, and also by the various user interfaces which present table information back to the user.

    Currently this routine returns the following list: derby.storage.initialPages derby.storage.minimumRecordSize derby.storage.pageReservedSpace derby.storage.pageSize

 public boolean isGlobal()
    Reveals whether the transaction is a global or local transaction.
 public boolean isIdle()
    Reveals whether the transaction has ever read or written data.
 public boolean isPristine()
    Reveals whether the transaction is read only.
 public  void logAndDo(Loggable operation) throws StandardException
    Log an operation and then action it in the context of this transaction.

    This simply passes the operation to the RawStore which logs and does it.

 public ConglomerateController openCompiledConglomerate(boolean hold,
    int open_mode,
    int lock_level,
    int isolation_level,
    StaticCompiledOpenConglomInfo static_info,
    DynamicCompiledOpenConglomInfo dynamic_info) throws StandardException
    Open a conglomerate for use, optionally include "compiled" info.

    Same as openConglomerate(), except that one can optionally provide "compiled" static_info and/or dynamic_info. This compiled information must have be gotten from getDynamicCompiledConglomInfo() and/or getStaticCompiledConglomInfo() calls on the same conglomid being opened. It is up to caller that "compiled" information is still valid and is appropriately multi-threaded protected.

 public ScanController openCompiledScan(boolean hold,
    int open_mode,
    int lock_level,
    int isolation_level,
    FormatableBitSet scanColumnList,
    DataValueDescriptor[] startKeyValue,
    int startSearchOperator,
    Qualifier[][] qualifier,
    DataValueDescriptor[] stopKeyValue,
    int stopSearchOperator,
    StaticCompiledOpenConglomInfo static_info,
    DynamicCompiledOpenConglomInfo dynamic_info) throws StandardException
    Open a scan on a conglomerate, optionally providing compiled info.

    Same as openScan(), except that one can optionally provide "compiled" static_info and/or dynamic_info. This compiled information must have be gotten from getDynamicCompiledConglomInfo() and/or getStaticCompiledConglomInfo() calls on the same conglomid being opened. It is up to caller that "compiled" information is still valid and is appropriately multi-threaded protected.

 public ConglomerateController openConglomerate(long conglomId,
    boolean hold,
    int open_mode,
    int lock_level,
    int isolation_level) throws StandardException
    Open a conglomerate for use.

    The lock level indicates the minimum lock level to get locks at, the underlying conglomerate implementation may actually lock at a higher level (ie. caller may request MODE_RECORD, but the table may be locked at MODE_TABLE instead).

    The close method is on the ConglomerateController interface.

 public GroupFetchScanController openGroupFetchScan(long conglomId,
    boolean hold,
    int open_mode,
    int lock_level,
    int isolation_level,
    FormatableBitSet scanColumnList,
    DataValueDescriptor[] startKeyValue,
    int startSearchOperator,
    Qualifier[][] qualifier,
    DataValueDescriptor[] stopKeyValue,
    int stopSearchOperator) throws StandardException
    Open a scan which gets copies of multiple rows at a time.

    All inputs work exactly as in openScan(). The return is a GroupFetchScanController, which only allows fetches of groups of rows from the conglomerate.

 public ScanController openScan(long conglomId,
    boolean hold,
    int open_mode,
    int lock_level,
    int isolation_level,
    FormatableBitSet scanColumnList,
    DataValueDescriptor[] startKeyValue,
    int startSearchOperator,
    Qualifier[][] qualifier,
    DataValueDescriptor[] stopKeyValue,
    int stopSearchOperator) throws StandardException
    Open a scan on a conglomerate. The scan will return all rows in the conglomerate which are between the positions defined by {startKeyValue, startSearchOperator} and {stopKeyValue, stopSearchOperator}, which also match the qualifier.

    The way that starting and stopping keys and operators are used may best be described by example. Say there's an ordered conglomerate with two columns, where the 0-th column is named 'x', and the 1st column is named 'y'. The values of the columns are as follows:

    x: 1 3 4 4 4 5 5 5 6 7 9
    y: 1 1 2 4 6 2 4 6 1 1 1
    

A {start key, search op} pair of {{5.2}, GE} would position on {x=5, y=2}, whereas the pair {{5}, GT} would position on {x=6, y=1}.

Partial keys are used to implement partial key scans in SQL. For example, the SQL "select * from t where x = 5" would open a scan on the conglomerate (or a useful index) of t using a starting position partial key of {{5}, GE} and a stopping position partial key of {{5}, GT}.

Some more examples:

+-------------------+------------+-----------+--------------+--------------+
| predicate         | start key  | stop key  | rows         | rows locked  |
|                   | value | op | value |op | returned     |serialization |
+-------------------+-------+----+-------+---+--------------+--------------+
| x = 5             | {5}   | GE | {5}   |GT |{5,2} .. {5,6}|{4,6} .. {5,6}|
| x > 5             | {5}   | GT | null  |   |{6,1} .. {9,1}|{5,6} .. {9,1}|
| x >= 5            | {5}   | GE | null  |   |{5,2} .. {9,1}|{4,6} .. {9,1}|
| x <= 5            | null  |    | {5}   |GT |{1,1} .. {5,6}|first .. {5,6}|
| x < 5             | null  |    | {5}   |GE |{1,1} .. {4,6}|first .. {4,6}|
| x >= 5 and x <= 7 | {5},  | GE | {7}   |GT |{5,2} .. {7,1}|{4,6} .. {7,1}|
| x = 5  and y > 2  | {5,2} | GT | {5}   |GT |{5,4} .. {5,6}|{5,2} .. {5,6}|
| x = 5  and y >= 2 | {5,2} | GE | {5}   |GT |{5,2} .. {5,6}|{4,6} .. {5,6}|
| x = 5  and y < 5  | {5}   | GE | {5,5} |GE |{5,2} .. {5,4}|{4,6} .. {5,4}|
| x = 2             | {2}   | GE | {2}   |GT | none         |{1,1} .. {1,1}|
+-------------------+-------+----+-------+---+--------------+--------------+

As the above table implies, the underlying scan may lock more rows than it returns in order to guarantee serialization.

For each row which meets the start and stop position, as described above the row is "qualified" to see whether it should be returned. The qualification is a 2 dimensional array of @see Qualifiers, which represents the qualification in conjunctive normal form (CNF). Conjunctive normal form is an "and'd" set of "or'd" Qualifiers.

For example x = 5 would be represented is pseudo code as: qualifier_cnf[][] = new Qualifier[1]; qualifier_cnf[0] = new Qualifier[1]; qualifier_cnr[0][0] = new Qualifer(x = 5)

For example (x = 5) or (y = 6) would be represented is pseudo code as: qualifier_cnf[][] = new Qualifier[1]; qualifier_cnf[0] = new Qualifier[2]; qualifier_cnr[0][0] = new Qualifer(x = 5) qualifier_cnr[0][1] = new Qualifer(y = 6)

For example ((x = 5) or (x = 6)) and ((y = 1) or (y = 2)) would be represented is pseudo code as: qualifier_cnf[][] = new Qualifier[2]; qualifier_cnf[0] = new Qualifier[2]; qualifier_cnr[0][0] = new Qualifer(x = 5) qualifier_cnr[0][1] = new Qualifer(x = 6) qualifier_cnr[0][0] = new Qualifer(y = 5) qualifier_cnr[0][1] = new Qualifer(y = 6)

For each row the CNF qualfier is processed and it is determined whether or not the row should be returned to the caller. The following pseudo-code describes how this is done:

if (qualifier != null)
{
for (int and_clause; and_clause < qualifier.length; and_clause++)
{
boolean or_qualifies = false;

for (int or_clause; or_clause < qualifier[and_clause].length; or_clause++)
{
DataValueDescriptor key     = 
qualifier[and_clause][or_clause].getOrderable();

DataValueDescriptor row_col = 
get row column[qualifier[and_clause][or_clause].getColumnId()];

boolean or_qualifies = 
row_col.compare(qualifier[i].getOperator,
key,
qualifier[i].getOrderedNulls,
qualifier[i].getUnknownRV);
if (or_qualifies) { break; } } if (!or_qualifies) {
don't return this row to the client - proceed to next row;
}
}
}
 public SortController openSort(long id) throws StandardException
    Open a sort controller for a sort previously created in this transaction. Sort controllers are used to insert rows into the sort.

    There may (in the future) be multiple sort inserters for a given sort, the idea being that the various threads of a parallel query plan can all insert into the sort. For now, however, only a single sort controller per sort is supported.

 public SortCostController openSortCostController(Properties implParameters) throws StandardException
    Return an open SortCostController.

    Return an open SortCostController which can be used to ask about the estimated costs of SortController() operations.

 public RowLocationRetRowSource openSortRowSource(long id) throws StandardException
    Open a scan for retrieving rows from a sort. Returns a RowSource for retrieving rows from the sort.
 public ScanController openSortScan(long id,
    boolean hold) throws StandardException
    Open a scan for retrieving rows from a sort. Returns a scan controller for retrieving rows from the sort (NOTE: the only legal methods to use on the returned sort controller are next() and fetch() - probably there should be scan controllers and updatable scan controllers).

    In the future, multiple sort scans on the same sort will be supported (for parallel execution across a uniqueness sort in which the order of the resulting rows is not important). Currently, only a single sort scan is allowed per sort.

    In the future, it will be possible to open a sort scan and start retrieving rows before the last row is inserted. The sort controller would block till rows were available to return. Currently, an attempt to retrieve a row before the sort controller is closed will cause an exception.

 public StoreCostController openStoreCost(long conglomId) throws StandardException
    Return an open StoreCostController for the given conglomid.

    Return an open StoreCostController which can be used to ask about the estimated row counts and costs of ScanController and ConglomerateController operations, on the given conglomerate.

 public  void purgeConglomerate(long conglomId) throws StandardException
    Purge all committed deleted rows from the conglomerate.

    This call will purge committed deleted rows from the conglomerate, that space will be available for future inserts into the conglomerate.

 public long recreateAndLoadConglomerate(String implementation,
    boolean recreate_ifempty,
    DataValueDescriptor[] template,
    ColumnOrdering[] columnOrder,
    int[] collationIds,
    Properties properties,
    int temporaryFlag,
    long orig_conglomId,
    RowLocationRetRowSource rowSource,
    long[] rowCount) throws StandardException
    Recreate a conglomerate and possibly load it with new rows that come from the new row source.

    This function behaves the same as @see createConglomerate except it also populates the conglomerate with rows from the row source and the rows that are inserted are not logged.

    Individual rows that are loaded into the conglomerate are not logged. After this operation, the underlying database must be backed up with a database backup rather than an transaction log backup (when we have them). This warning is put here for the benefit of future generation.

 public int releaseSavePoint(String name,
    Object kindOfSavepoint) throws StandardException
    Release the save point of the given name. Releasing a savepoint removes all knowledge from this transaction of the named savepoint and any savepoints set since the named savepoint was set.
 public int rollbackToSavePoint(String name,
    boolean close_controllers,
    Object kindOfSavepoint) throws StandardException
    Rollback all changes made since the named savepoint was set. The named savepoint is not released, it remains valid within this transaction, and thus can be named it future rollbackToSavePoint() calls. Any savepoints set since this named savepoint are released (and their changes rolled back).

    if "close_controllers" is true then all conglomerates and scans are closed (held or non-held).

    If "close_controllers" is false then no cleanup is done by the TransactionController. It is then the responsibility of the caller to close all resources that may have been affected by the statements backed out by the call. This option is meant to be used by the Language implementation of statement level backout, where the system "knows" what could be affected by the scope of the statements executed within the statement.

 public  void setNoLockWait(boolean noWait)
    Tell this transaction whether it should time out immediately if a lock cannot be granted without waiting. This mechanism can for instance be used if an operation is first attempted in a nested transaction to reduce the lifetime of locks held in the system tables (like when a stored prepared statement is compiled and stored). In such a case, the caller must catch timeout exceptions and retry the operation in the main transaction if a lock timeout occurs.
 public int setSavePoint(String name,
    Object kindOfSavepoint) throws StandardException
    Set a save point in the current transaction. A save point defines a point in time in the transaction that changes can be rolled back to. Savepoints can be nested and they behave like a stack. Setting save points "one" and "two" and the rolling back "one" will rollback all the changes made since "one" (including those made since "two") and release savepoint "two".
 public TransactionController startNestedUserTransaction(boolean readOnly) throws StandardException
    Get an nested user transaction.

    A nested user transaction can be used exactly as any other TransactionController, except as follows. For this discussion let the parent transaction be the transaction used to make the startNestedUserTransaction() call, and let the child transaction be the transaction returned by the startNestedUserTransaction() call.

    A parent transaction can nest a single readonly transaction and a single separate read/write transaction. If a subsequent nested transaction creation is attempted against the parent prior to destroying an existing nested user transaction of the same type, an exception will be thrown.

    The nesting is limited to one level deep. An exception will be thrown if a subsequent getNestedUserTransaction() is called on the child transaction.

    The locks in the child transaction of a readOnly nested user transaction will be compatible with the locks of the parent transaction. The locks in the child transaction of a non-readOnly nested user transaction will NOT be compatible with those of the parent transaction - this is necessary for correct recovery behavior.

    A commit in the child transaction will release locks associated with the child transaction only, work can continue in the parent transaction at this point.

    Any abort of the child transaction will result in an abort of both the child transaction and parent transaction, either initiated by an explict abort() call or by an exception that results in an abort.

    A TransactionController.destroy() call should be made on the child transaction once all child work is done, and the caller wishes to continue work in the parent transaction.

    AccessFactory.getTransaction() will always return the "parent" transaction, never the child transaction. Thus clients using nested user transactions must keep track of the transaction, as there is no interface to query the storage system to get the current child transaction. The idea is that a nested user transaction should be used to for a limited amount of work, committed, and then work continues in the parent transaction.

    Nested User transactions are meant to be used to implement system work necessary to commit as part of implementing a user's request, but where holding the lock for the duration of the user transaction is not acceptable. 2 examples of this are system catalog read locks accumulated while compiling a plan, and auto-increment.

    Once the first write of a non-readOnly nested transaction is done, then the nested user transaction must be committed or aborted before any write operation is attempted in the parent transaction.