path: root/doc/src/sgml/custom-scan.sgml

<!-- doc/src/sgml/custom-scan.sgml -->

<chapter id="custom-scan">
 <title>Writing a Custom Scan Provider</title>

 <indexterm zone="custom-scan">
  <primary>custom scan provider</primary>
  <secondary>handler for</secondary>
 </indexterm>

 <para>
  <productname>PostgreSQL</productname> supports a set of experimental facilities which
  are intended to allow extension modules to add new scan types to the system.
  Unlike a <link linkend="fdwhandler">foreign data wrapper</link>, which is only
  responsible for knowing how to scan its own foreign tables, a custom scan
  provider can provide an alternative method of scanning any relation in the
  system.  Typically, the motivation for writing a custom scan provider will
  be to allow the use of some optimization not supported by the core
  system, such as caching or some form of hardware acceleration.  This chapter
  outlines how to write a new custom scan provider.
 </para>

 <para>
  Implementing a new type of custom scan is a three-step process.  First,
  during planning, it is necessary to generate access paths representing a
  scan using the proposed strategy.  Second, if one of those access paths
  is selected by the planner as the optimal strategy for scanning a
  particular relation, the access path must be converted to a plan.
  Finally, it must be possible to execute the plan and generate the same
  results that would have been generated for any other access path targeting
  the same relation.
 </para>

 <sect1 id="custom-scan-path">
  <title>Creating Custom Scan Paths</title>

  <para>
    A custom scan provider will typically add paths for a base relation by
    setting the following hook, which is called after the core code has
    generated all the access paths it can for the relation (except for
    Gather paths, which are made after this call so that they can use
    partial paths added by the hook):
<programlisting>
typedef void (*set_rel_pathlist_hook_type) (PlannerInfo *root,
                                            RelOptInfo *rel,
                                            Index rti,
                                            RangeTblEntry *rte);
extern PGDLLIMPORT set_rel_pathlist_hook_type set_rel_pathlist_hook;
</programlisting>
  </para>

  <para>
    Although this hook function can be used to examine, modify, or remove
    paths generated by the core system, a custom scan provider will typically
    confine itself to generating <structname>CustomPath</structname> objects and adding
    them to <literal>rel</literal> using <function>add_path</function>.  The custom scan
    provider is responsible for initializing the <structname>CustomPath</structname>
    object, which is declared like this:
<programlisting>
typedef struct CustomPath
{
    Path      path;
    uint32    flags;
    List     *custom_paths;
    List     *custom_private;
    const CustomPathMethods *methods;
} CustomPath;
</programlisting>
  </para>

  <para>
    <structfield>path</structfield> must be initialized as for any other path, including
    the row-count estimate, start and total cost, and sort ordering provided
    by this path.  <structfield>flags</structfield> is a bit mask, which should include
    <literal>CUSTOMPATH_SUPPORT_BACKWARD_SCAN</literal> if the custom path can support
    a backward scan and <literal>CUSTOMPATH_SUPPORT_MARK_RESTORE</literal> if it
    can support mark and restore.  Both capabilities are optional.
    An optional <structfield>custom_paths</structfield> is a list of <structname>Path</structname>
    nodes used by this custom-path node; these will be transformed into
    <structname>Plan</structname> nodes by planner.
    <structfield>custom_private</structfield> can be used to store the custom path's
    private data.  Private data should be stored in a form that can be handled
    by <literal>nodeToString</literal>, so that debugging routines that attempt to
    print the custom path will work as designed.  <structfield>methods</structfield> must
    point to a (usually statically allocated) object implementing the required
    custom path methods, of which there is currently only one.
  </para>

  <para>
   A custom scan provider can also provide join paths.  Just as for base
   relations, such a path must produce the same output as would normally be
   produced by the join it replaces.  To do this, the join provider should
   set the following hook, and then within the hook function,
   create <structname>CustomPath</structname> path(s) for the join relation.
<programlisting>
typedef void (*set_join_pathlist_hook_type) (PlannerInfo *root,
                                             RelOptInfo *joinrel,
                                             RelOptInfo *outerrel,
                                             RelOptInfo *innerrel,
                                             JoinType jointype,
                                             JoinPathExtraData *extra);
extern PGDLLIMPORT set_join_pathlist_hook_type set_join_pathlist_hook;
</programlisting>

   This hook will be invoked repeatedly for the same join relation, with
   different combinations of inner and outer relations; it is the
   responsibility of the hook to minimize duplicated work.
  </para>

  <sect2 id="custom-scan-path-callbacks">
  <title>Custom Scan Path Callbacks</title>

  <para>
<programlisting>
Plan *(*PlanCustomPath) (PlannerInfo *root,
                         RelOptInfo *rel,
                         CustomPath *best_path,
                         List *tlist,
                         List *clauses,
                         List *custom_plans);
</programlisting>
    Convert a custom path to a finished plan.  The return value will generally
    be a <literal>CustomScan</literal> object, which the callback must allocate and
    initialize.  See <xref linkend="custom-scan-plan"/> for more details.
   </para>
  </sect2>
 </sect1>

 <sect1 id="custom-scan-plan">
  <title>Creating Custom Scan Plans</title>

  <para>
    A custom scan is represented in a finished plan tree using the following
    structure:
<programlisting>
typedef struct CustomScan
{
    Scan      scan;
    uint32    flags;
    List     *custom_plans;
    List     *custom_exprs;
    List     *custom_private;
    List     *custom_scan_tlist;
    Bitmapset *custom_relids;
    const CustomScanMethods *methods;
} CustomScan;
</programlisting>
  </para>

  <para>
    <structfield>scan</structfield> must be initialized as for any other scan, including
    estimated costs, target lists, qualifications, and so on.
    <structfield>flags</structfield> is a bit mask with the same meaning as in
    <structname>CustomPath</structname>.
    <structfield>custom_plans</structfield> can be used to store child
    <structname>Plan</structname> nodes.
    <structfield>custom_exprs</structfield> should be used to
    store expression trees that will need to be fixed up by
    <filename>setrefs.c</filename> and <filename>subselect.c</filename>, while
    <structfield>custom_private</structfield> should be used to store other private data
    that is only used by the custom scan provider itself.
    <structfield>custom_scan_tlist</structfield> can be NIL when scanning a base
    relation, indicating that the custom scan returns scan tuples that match
    the base relation's row type.  Otherwise it is a target list describing
    the actual scan tuples.  <structfield>custom_scan_tlist</structfield> must be
    provided for joins, and could be provided for scans if the custom scan
    provider can compute some non-Var expressions.
    <structfield>custom_relids</structfield> is set by the core code to the set of
    relations (range table indexes) that this scan node handles; except when
    this scan is replacing a join, it will have only one member.
    <structfield>methods</structfield> must point to a (usually statically allocated)
    object implementing the required custom scan methods, which are further
    detailed below.
  </para>

  <para>
   When a <structname>CustomScan</structname> scans a single relation,
   <structfield>scan.scanrelid</structfield> must be the range table index of the table
   to be scanned.  When it replaces a join, <structfield>scan.scanrelid</structfield>
   should be zero.
  </para>

  <para>
   Plan trees must be able to be duplicated using <function>copyObject</function>,
   so all the data stored within the <quote>custom</quote> fields must consist of
   nodes that that function can handle.  Furthermore, custom scan providers
   cannot substitute a larger structure that embeds
   a <structname>CustomScan</structname> for the structure itself, as would be possible
   for a <structname>CustomPath</structname> or <structname>CustomScanState</structname>.
  </para>

  <sect2 id="custom-scan-plan-callbacks">
   <title>Custom Scan Plan Callbacks</title>
   <para>
<programlisting>
Node *(*CreateCustomScanState) (CustomScan *cscan);
</programlisting>
    Allocate a <structname>CustomScanState</structname> for this
    <structname>CustomScan</structname>.  The actual allocation will often be larger than
    required for an ordinary <structname>CustomScanState</structname>, because many
    providers will wish to embed that as the first field of a larger structure.
    The value returned must have the node tag and <structfield>methods</structfield>
    set appropriately, but other fields should be left as zeroes at this
    stage; after <function>ExecInitCustomScan</function> performs basic initialization,
    the <function>BeginCustomScan</function> callback will be invoked to give the
    custom scan provider a chance to do whatever else is needed.
   </para>
  </sect2>
 </sect1>

 <sect1 id="custom-scan-execution">
  <title>Executing Custom Scans</title>

  <para>
   When a <structfield>CustomScan</structfield> is executed, its execution state is
   represented by a <structfield>CustomScanState</structfield>, which is declared as
   follows:
<programlisting>
typedef struct CustomScanState
{
    ScanState ss;
    uint32    flags;
    const CustomExecMethods *methods;
} CustomScanState;
</programlisting>
  </para>

  <para>
   <structfield>ss</structfield> is initialized as for any other scan state,
   except that if the scan is for a join rather than a base relation,
   <literal>ss.ss_currentRelation</literal> is left NULL.
   <structfield>flags</structfield> is a bit mask with the same meaning as in
   <structname>CustomPath</structname> and <structname>CustomScan</structname>.
   <structfield>methods</structfield> must point to a (usually statically allocated)
   object implementing the required custom scan state methods, which are
   further detailed below.  Typically, a <structname>CustomScanState</structname>, which
   need not support <function>copyObject</function>, will actually be a larger
   structure embedding the above as its first member.
  </para>

  <sect2 id="custom-scan-execution-callbacks">
   <title>Custom Scan Execution Callbacks</title>

   <para>
<programlisting>
void (*BeginCustomScan) (CustomScanState *node,
                         EState *estate,
                         int eflags);
</programlisting>
    Complete initialization of the supplied <structname>CustomScanState</structname>.
    Standard fields have been initialized by <function>ExecInitCustomScan</function>,
    but any private fields should be initialized here.
   </para>

   <para>
<programlisting>
TupleTableSlot *(*ExecCustomScan) (CustomScanState *node);
</programlisting>
    Fetch the next scan tuple.  If any tuples remain, it should fill
    <literal>ps_ResultTupleSlot</literal> with the next tuple in the current scan
    direction, and then return the tuple slot.  If not,
    <literal>NULL</literal> or an empty slot should be returned.
   </para>

   <para>
<programlisting>
void (*EndCustomScan) (CustomScanState *node);
</programlisting>
    Clean up any private data associated with the <literal>CustomScanState</literal>.
    This method is required, but it does not need to do anything if there is
    no associated data or it will be cleaned up automatically.
   </para>

   <para>
<programlisting>
void (*ReScanCustomScan) (CustomScanState *node);
</programlisting>
    Rewind the current scan to the beginning and prepare to rescan the
    relation.
   </para>

   <para>
<programlisting>
void (*MarkPosCustomScan) (CustomScanState *node);
</programlisting>
    Save the current scan position so that it can subsequently be restored
    by the <function>RestrPosCustomScan</function> callback.  This callback is
    optional, and need only be supplied if the
    <literal>CUSTOMPATH_SUPPORT_MARK_RESTORE</literal> flag is set.
   </para>

   <para>
<programlisting>
void (*RestrPosCustomScan) (CustomScanState *node);
</programlisting>
    Restore the previous scan position as saved by the
    <function>MarkPosCustomScan</function> callback.  This callback is optional,
    and need only be supplied if the
    <literal>CUSTOMPATH_SUPPORT_MARK_RESTORE</literal> flag is set.
   </para>

   <para>
<programlisting>
Size (*EstimateDSMCustomScan) (CustomScanState *node,
                               ParallelContext *pcxt);
</programlisting>
    Estimate the amount of dynamic shared memory that will be required
    for parallel operation.  This may be higher than the amount that will
    actually be used, but it must not be lower.  The return value is in bytes.
    This callback is optional, and need only be supplied if this custom
    scan provider supports parallel execution.
   </para>

   <para>
<programlisting>
void (*InitializeDSMCustomScan) (CustomScanState *node,
                                 ParallelContext *pcxt,
                                 void *coordinate);
</programlisting>
    Initialize the dynamic shared memory that will be required for parallel
    operation.  <literal>coordinate</literal> points to a shared memory area of
    size equal to the return value of <function>EstimateDSMCustomScan</function>.
    This callback is optional, and need only be supplied if this custom
    scan provider supports parallel execution.
   </para>

   <para>
<programlisting>
void (*ReInitializeDSMCustomScan) (CustomScanState *node,
                                   ParallelContext *pcxt,
                                   void *coordinate);
</programlisting>
    Re-initialize the dynamic shared memory required for parallel operation
    when the custom-scan plan node is about to be re-scanned.
    This callback is optional, and need only be supplied if this custom
    scan provider supports parallel execution.
    Recommended practice is that this callback reset only shared state,
    while the <function>ReScanCustomScan</function> callback resets only local
    state.  Currently, this callback will be called
    before <function>ReScanCustomScan</function>, but it's best not to rely on
    that ordering.
   </para>

   <para>
<programlisting>
void (*InitializeWorkerCustomScan) (CustomScanState *node,
                                    shm_toc *toc,
                                    void *coordinate);
</programlisting>
    Initialize a parallel worker's local state based on the shared state
    set up by the leader during <function>InitializeDSMCustomScan</function>.
    This callback is optional, and need only be supplied if this custom
    scan provider supports parallel execution.
   </para>

   <para>
<programlisting>
void (*ShutdownCustomScan) (CustomScanState *node);
</programlisting>
    Release resources when it is anticipated the node will not be executed
    to completion.  This is not called in all cases; sometimes,
    <literal>EndCustomScan</literal> may be called without this function having
    been called first.  Since the DSM segment used by parallel query is
    destroyed just after this callback is invoked, custom scan providers that
    wish to take some action before the DSM segment goes away should implement
    this method.
   </para>

   <para>
<programlisting>
void (*ExplainCustomScan) (CustomScanState *node,
                           List *ancestors,
                           ExplainState *es);
</programlisting>
    Output additional information for <command>EXPLAIN</command> of a custom-scan
    plan node.  This callback is optional.  Common data stored in the
    <structname>ScanState</structname>, such as the target list and scan relation, will
    be shown even without this callback, but the callback allows the display
    of additional, private state.
   </para>
  </sect2>
 </sect1>
</chapter>