SSMBase.h

Data buckets containing the fixed length data (scalars and direct arrays of data type Int, Float, Bool, etc.). For variable shaped data (strings and indirect arrays) they contain references to the actual data position in the string buckets or in an external file.
String buckets containing strings and array of strings.
Index buckets containing the index info for the data buckets.

Bucket access is handled by class BucketCache. It also keeps a list of free buckets. A bucket is freed when it is not needed anymore (e.g. all data from it are deleted).

Data buckets form the main part of the SSM. The data can be viewed as a few streams of buckets, where each stream contains the data of a given number of columns. Each stream has an SSMIndex object describing the number of rows stored in each data bucket of the stream. The SSM starts with a single bucket stream (holding all columns), but when columns are added, new bucket streams might be created.

For example, we have an SSM with a bucket size of 100 bytes. There are 5 Int columns (A,B,C,D,E) each taking 4 bytes per row. Column A, B, C, and D are stored in bucket stream 1, while column E is stored in bucket stream 2. So in stream 1 each bucket can hold 6 rows, while in stream 2 each bucket can hold 25 rows. For a 100 row table it will result in 17+4 data buckets.

A few classes collaborate to make it work:

Each bucket stream has an SSMIndex object to map row number to bucket number. Note that in principle each bucket in a stream contains the same number of rows. However, when a row is deleted it is removed from its bucket shifting the remainder to the left. Data in the next buckets is not shifted, so that bucket has now one row less.
For each column SSMBase knows to which bucket stream it belongs and at which offset the column starts in a bucket. Note that column data in a bucket are adjacent, which is done to make it easier to use the ColumnCache object in SSMColumn and to be able to efficiently store Bool values as bits.
Each column has an SSMColumn object knowing how many bits each data cell takes in a bucket. The SSMColumn objects handle all access to data in the columns (using SSMBase and SSMIndex).

String buckets are used by class SSMStringHandler to store scalar strings and fixed and variable shaped arrays of strings. The bucketnr, offset, and length of such string (arrays) are stored in the data buckets.
Indirect arrays of other data types are also stored indirectly and their offset is stored in the data buckets. Such arrays are handled by class StIndArray which uses an extra file to store the arrays.

Index buckets are used by SSMBase to make the SSMIndex data persistent. It uses alternately 2 sets of index buckets. In that way there is always an index availanle in case the system crashes. If possible 2 halfs of a single bucket are used alternately, otherwise separate buckets are used.

Motivation

The public interface of SSMBase is quite large, because the other internal SSM classes need these functions. To have a class with a minimal interface for the normal user, class StandardStMan is derived from it.
StandardStMan needs an isA- instead of hasA-relation to be able to bind columns to it in class SetupNewTable.

To Do

Remove AipsIO argument from open and close.
When only 1 bucket in use addcolumn can check if there's enough room to fit the new column (so rearange the bucket) in the free row space.

Member Description

explicit SSMBase (Int aBucketSize=0, uInt aCacheSize=1)

Create a Standard storage manager with default name SSM.

explicit SSMBase (const String& aDataManName, Int aBucketSize=0, uInt aCacheSize=1)

Create a Standard storage manager with the given name.

SSMBase (const String& aDataManName, const Record& spec)

Create a Standard storage manager with the given name. The specifications are part of the record (as created by dataManagerSpec).

~SSMBase()

virtual DataManager* clone() const

Clone this object. It does not clone SSMColumn objects possibly used. The caller has to delete the newly created object.

virtual String dataManagerType() const

Get the type name of the data manager (i.e. StandardStMan).

virtual String dataManagerName() const

Get the name given to the storage manager (in the constructor).

virtual Record dataManagerSpec() const

Record a record containing data manager specifications.

uInt getVersion() const

Get the version of the class.

void setCacheSize (uInt aCacheSize)

Set the cache size (in buckets).

uInt getCacheSize() const

Get the current cache size (in buckets).

void clearCache()

Clear the cache used by this storage manager. It will flush the cache as needed and remove all buckets from it.

void showCacheStatistics (ostream& anOs) const

Show the statistics of all caches used.

void showIndexStatistics (ostream & anOs) const

Show Statistics of all indices used.

void showBaseStatistics (ostream & anOs) const

Show Statistics of the Base offsets/index etc.

uInt getBucketSize() const

Get the bucket size.

uInt getNRow() const

Get the number of rows in this storage manager.

virtual Bool canAddRow() const

The storage manager can add rows.

virtual Bool canRemoveRow() const

The storage manager can delete rows.

virtual Bool canAddColumn() const

The storage manager can add columns.

virtual Bool canRemoveColumn() const

The storage manager can delete columns.

static DataManager* makeObject (const String& aDataManType, const Record& spec)

Make the object from the type name string. This function gets registered in the DataManager "constructor" map. The caller has to delete the object.

SSMColumn& getColumn (uInt aColNr)

Get access to the given column.

SSMIndex& getIndex (uInt anIdxNr)

Get access to the given Index.

void setBucketDirty()

Make the current bucket in the cache dirty (i.e. something has been changed in it and it needs to be written when removed from the cache). (used by SSMColumn::putValue).

StManArrayFile* openArrayFile (ByteIO::OpenOption anOpt)

Open (if needed) the file for indirect arrays with the given mode. Return a pointer to the object.

char* find (uInt aRowNr, uInt aColNr, uInt& aStartRow, uInt& anEndRow)

Find the bucket containing the column and row and return the pointer to the beginning of the column data in that bucket. It also fills in the start and end row for the column data.

uInt getNewBucket()

Add a new bucket and get its bucket number.

char* getBucket (uInt aBucketNr)

Read the bucket (if needed) and return the pointer to it.

void removeBucket (uInt aBucketNr)

Remove a bucket from the bucket cache.

uInt getRowsPerBucket (uInt aColumn) const

Get rows per bucket for the given column.

SSMStringHandler* getStringHandler()

Return a pointer to the (one and only) StringHandler object.

static char* readCallBack (void* anOwner, const char* aBucketStorage)

Callbacks for BucketCache access.

static void writeCallBack (void* anOwner, char* aBucketStorage, const char* aBucket)
static void deleteCallBack (void, char aBucket)
static char* initCallBack (void* anOwner)

SSMBase (const SSMBase& that)

Copy constructor (only meant for clone function).

SSMBase& operator= (const SSMBase& that)

Assignment cannot be used.

void recreate()

(Re)create the index, file, and cache object. It is used when all rows are deleted from the table.

virtual Bool flush (AipsIO&, Bool doFsync)

Flush and optionally fsync the data. It returns a True status if it had to flush (i.e. if data have changed).

virtual void create (uInt aNrRows)

Let the storage manager create files as needed for a new table. This allows a column with an indirect array to create its file.

virtual void open (uInt aRowNr, AipsIO&)

Open the storage manager file for an existing table, read in the data, and let the SSMColumn objects read their data.

virtual void resync (uInt aRowNr)

Resync the storage manager with the new file contents. This is done by clearing the cache.

virtual void reopenRW()

Reopen the storage manager files for read/write.

virtual void deleteManager()

The data manager will be deleted (because all its columns are requested to be deleted). So clean up the things needed (e.g. delete files).

void init()

Let the storage manager initialize itself (upon creation). It determines the bucket size and fills the index.

uInt setBucketSize()

Determine and set the bucket size. It returns the number of rows per bucket.

uInt getNrIndices() const

Get the number of indices in use.

virtual void addRow (uInt aNrRows)

Add rows to the storage manager. Per column it extends number of rows.

virtual void removeRow (uInt aRowNr)

Delete a row from all columns.

virtual void addColumn (DataManagerColumn*)

Do the final addition of a column.

virtual void removeColumn (DataManagerColumn*)

Remove a column from the data file.

virtual DataManagerColumn* makeScalarColumn (const String& aName, int aDataType, const String& aDataTypeID)

Create a column in the storage manager on behalf of a table column. The caller has to delete the newly created object.

Create a scalar column.

virtual DataManagerColumn* makeDirArrColumn (const String& aName, int aDataType, const String& aDataTypeID)

Create a column in the storage manager on behalf of a table column. The caller has to delete the newly created object.

Create a direct array column.

virtual DataManagerColumn* makeIndArrColumn (const String& aName, int aDataType, const String& aDataTypeID)

Create a column in the storage manager on behalf of a table column. The caller has to delete the newly created object.

Create an indirect array column.

BucketCache& getCache()

Get the cache object. This will construct the cache object if not present yet. The cache object will be deleted by the destructor.

void makeCache()

Construct the cache object (if not constructed yet).

void readHeader()

Read the header..

void readIndexBuckets()

Read the index from its buckets.

void writeIndex()

Write the header and the indices.