LCOV - code coverage report
Current view: top level - contrib/file_fdw - file_fdw.c (source / functions) Hit Total Coverage
Test: PostgreSQL 18devel Lines: 312 347 89.9 %
Date: 2025-04-01 15:15:16 Functions: 22 22 100.0 %
Legend: Lines: hit not hit

          Line data    Source code
       1             : /*-------------------------------------------------------------------------
       2             :  *
       3             :  * file_fdw.c
       4             :  *        foreign-data wrapper for server-side flat files (or programs).
       5             :  *
       6             :  * Copyright (c) 2010-2025, PostgreSQL Global Development Group
       7             :  *
       8             :  * IDENTIFICATION
       9             :  *        contrib/file_fdw/file_fdw.c
      10             :  *
      11             :  *-------------------------------------------------------------------------
      12             :  */
      13             : #include "postgres.h"
      14             : 
      15             : #include <sys/stat.h>
      16             : #include <unistd.h>
      17             : 
      18             : #include "access/htup_details.h"
      19             : #include "access/reloptions.h"
      20             : #include "access/sysattr.h"
      21             : #include "access/table.h"
      22             : #include "catalog/pg_authid.h"
      23             : #include "catalog/pg_foreign_table.h"
      24             : #include "commands/copy.h"
      25             : #include "commands/copyfrom_internal.h"
      26             : #include "commands/defrem.h"
      27             : #include "commands/explain_format.h"
      28             : #include "commands/explain_state.h"
      29             : #include "commands/vacuum.h"
      30             : #include "executor/executor.h"
      31             : #include "foreign/fdwapi.h"
      32             : #include "foreign/foreign.h"
      33             : #include "miscadmin.h"
      34             : #include "nodes/makefuncs.h"
      35             : #include "optimizer/optimizer.h"
      36             : #include "optimizer/pathnode.h"
      37             : #include "optimizer/planmain.h"
      38             : #include "optimizer/restrictinfo.h"
      39             : #include "utils/acl.h"
      40             : #include "utils/memutils.h"
      41             : #include "utils/rel.h"
      42             : #include "utils/sampling.h"
      43             : #include "utils/varlena.h"
      44             : 
      45           2 : PG_MODULE_MAGIC_EXT(
      46             :                     .name = "file_fdw",
      47             :                     .version = PG_VERSION
      48             : );
      49             : 
      50             : /*
      51             :  * Describes the valid options for objects that use this wrapper.
      52             :  */
      53             : struct FileFdwOption
      54             : {
      55             :     const char *optname;
      56             :     Oid         optcontext;     /* Oid of catalog in which option may appear */
      57             : };
      58             : 
      59             : /*
      60             :  * Valid options for file_fdw.
      61             :  * These options are based on the options for the COPY FROM command.
      62             :  * But note that force_not_null and force_null are handled as boolean options
      63             :  * attached to a column, not as table options.
      64             :  *
      65             :  * Note: If you are adding new option for user mapping, you need to modify
      66             :  * fileGetOptions(), which currently doesn't bother to look at user mappings.
      67             :  */
      68             : static const struct FileFdwOption valid_options[] = {
      69             :     /* Data source options */
      70             :     {"filename", ForeignTableRelationId},
      71             :     {"program", ForeignTableRelationId},
      72             : 
      73             :     /* Format options */
      74             :     /* oids option is not supported */
      75             :     {"format", ForeignTableRelationId},
      76             :     {"header", ForeignTableRelationId},
      77             :     {"delimiter", ForeignTableRelationId},
      78             :     {"quote", ForeignTableRelationId},
      79             :     {"escape", ForeignTableRelationId},
      80             :     {"null", ForeignTableRelationId},
      81             :     {"default", ForeignTableRelationId},
      82             :     {"encoding", ForeignTableRelationId},
      83             :     {"on_error", ForeignTableRelationId},
      84             :     {"log_verbosity", ForeignTableRelationId},
      85             :     {"reject_limit", ForeignTableRelationId},
      86             :     {"force_not_null", AttributeRelationId},
      87             :     {"force_null", AttributeRelationId},
      88             : 
      89             :     /*
      90             :      * force_quote is not supported by file_fdw because it's for COPY TO.
      91             :      */
      92             : 
      93             :     /* Sentinel */
      94             :     {NULL, InvalidOid}
      95             : };
      96             : 
      97             : /*
      98             :  * FDW-specific information for RelOptInfo.fdw_private.
      99             :  */
     100             : typedef struct FileFdwPlanState
     101             : {
     102             :     char       *filename;       /* file or program to read from */
     103             :     bool        is_program;     /* true if filename represents an OS command */
     104             :     List       *options;        /* merged COPY options, excluding filename and
     105             :                                  * is_program */
     106             :     BlockNumber pages;          /* estimate of file's physical size */
     107             :     double      ntuples;        /* estimate of number of data rows */
     108             : } FileFdwPlanState;
     109             : 
     110             : /*
     111             :  * FDW-specific information for ForeignScanState.fdw_state.
     112             :  */
     113             : typedef struct FileFdwExecutionState
     114             : {
     115             :     char       *filename;       /* file or program to read from */
     116             :     bool        is_program;     /* true if filename represents an OS command */
     117             :     List       *options;        /* merged COPY options, excluding filename and
     118             :                                  * is_program */
     119             :     CopyFromState cstate;       /* COPY execution state */
     120             : } FileFdwExecutionState;
     121             : 
     122             : /*
     123             :  * SQL functions
     124             :  */
     125           4 : PG_FUNCTION_INFO_V1(file_fdw_handler);
     126           4 : PG_FUNCTION_INFO_V1(file_fdw_validator);
     127             : 
     128             : /*
     129             :  * FDW callback routines
     130             :  */
     131             : static void fileGetForeignRelSize(PlannerInfo *root,
     132             :                                   RelOptInfo *baserel,
     133             :                                   Oid foreigntableid);
     134             : static void fileGetForeignPaths(PlannerInfo *root,
     135             :                                 RelOptInfo *baserel,
     136             :                                 Oid foreigntableid);
     137             : static ForeignScan *fileGetForeignPlan(PlannerInfo *root,
     138             :                                        RelOptInfo *baserel,
     139             :                                        Oid foreigntableid,
     140             :                                        ForeignPath *best_path,
     141             :                                        List *tlist,
     142             :                                        List *scan_clauses,
     143             :                                        Plan *outer_plan);
     144             : static void fileExplainForeignScan(ForeignScanState *node, ExplainState *es);
     145             : static void fileBeginForeignScan(ForeignScanState *node, int eflags);
     146             : static TupleTableSlot *fileIterateForeignScan(ForeignScanState *node);
     147             : static void fileReScanForeignScan(ForeignScanState *node);
     148             : static void fileEndForeignScan(ForeignScanState *node);
     149             : static bool fileAnalyzeForeignTable(Relation relation,
     150             :                                     AcquireSampleRowsFunc *func,
     151             :                                     BlockNumber *totalpages);
     152             : static bool fileIsForeignScanParallelSafe(PlannerInfo *root, RelOptInfo *rel,
     153             :                                           RangeTblEntry *rte);
     154             : 
     155             : /*
     156             :  * Helper functions
     157             :  */
     158             : static bool is_valid_option(const char *option, Oid context);
     159             : static void fileGetOptions(Oid foreigntableid,
     160             :                            char **filename,
     161             :                            bool *is_program,
     162             :                            List **other_options);
     163             : static List *get_file_fdw_attribute_options(Oid relid);
     164             : static bool check_selective_binary_conversion(RelOptInfo *baserel,
     165             :                                               Oid foreigntableid,
     166             :                                               List **columns);
     167             : static void estimate_size(PlannerInfo *root, RelOptInfo *baserel,
     168             :                           FileFdwPlanState *fdw_private);
     169             : static void estimate_costs(PlannerInfo *root, RelOptInfo *baserel,
     170             :                            FileFdwPlanState *fdw_private,
     171             :                            Cost *startup_cost, Cost *total_cost);
     172             : static int  file_acquire_sample_rows(Relation onerel, int elevel,
     173             :                                      HeapTuple *rows, int targrows,
     174             :                                      double *totalrows, double *totaldeadrows);
     175             : 
     176             : 
     177             : /*
     178             :  * Foreign-data wrapper handler function: return a struct with pointers
     179             :  * to my callback routines.
     180             :  */
     181             : Datum
     182          38 : file_fdw_handler(PG_FUNCTION_ARGS)
     183             : {
     184          38 :     FdwRoutine *fdwroutine = makeNode(FdwRoutine);
     185             : 
     186          38 :     fdwroutine->GetForeignRelSize = fileGetForeignRelSize;
     187          38 :     fdwroutine->GetForeignPaths = fileGetForeignPaths;
     188          38 :     fdwroutine->GetForeignPlan = fileGetForeignPlan;
     189          38 :     fdwroutine->ExplainForeignScan = fileExplainForeignScan;
     190          38 :     fdwroutine->BeginForeignScan = fileBeginForeignScan;
     191          38 :     fdwroutine->IterateForeignScan = fileIterateForeignScan;
     192          38 :     fdwroutine->ReScanForeignScan = fileReScanForeignScan;
     193          38 :     fdwroutine->EndForeignScan = fileEndForeignScan;
     194          38 :     fdwroutine->AnalyzeForeignTable = fileAnalyzeForeignTable;
     195          38 :     fdwroutine->IsForeignScanParallelSafe = fileIsForeignScanParallelSafe;
     196             : 
     197          38 :     PG_RETURN_POINTER(fdwroutine);
     198             : }
     199             : 
     200             : /*
     201             :  * Validate the generic options given to a FOREIGN DATA WRAPPER, SERVER,
     202             :  * USER MAPPING or FOREIGN TABLE that uses file_fdw.
     203             :  *
     204             :  * Raise an ERROR if the option or its value is considered invalid.
     205             :  */
     206             : Datum
     207         122 : file_fdw_validator(PG_FUNCTION_ARGS)
     208             : {
     209         122 :     List       *options_list = untransformRelOptions(PG_GETARG_DATUM(0));
     210         122 :     Oid         catalog = PG_GETARG_OID(1);
     211         122 :     char       *filename = NULL;
     212         122 :     DefElem    *force_not_null = NULL;
     213         122 :     DefElem    *force_null = NULL;
     214         122 :     List       *other_options = NIL;
     215             :     ListCell   *cell;
     216             : 
     217             :     /*
     218             :      * Check that only options supported by file_fdw, and allowed for the
     219             :      * current object type, are given.
     220             :      */
     221         402 :     foreach(cell, options_list)
     222             :     {
     223         298 :         DefElem    *def = (DefElem *) lfirst(cell);
     224             : 
     225         298 :         if (!is_valid_option(def->defname, catalog))
     226             :         {
     227             :             const struct FileFdwOption *opt;
     228             :             const char *closest_match;
     229             :             ClosestMatchState match_state;
     230          16 :             bool        has_valid_options = false;
     231             : 
     232             :             /*
     233             :              * Unknown option specified, complain about it. Provide a hint
     234             :              * with a valid option that looks similar, if there is one.
     235             :              */
     236          16 :             initClosestMatch(&match_state, def->defname, 4);
     237         256 :             for (opt = valid_options; opt->optname; opt++)
     238             :             {
     239         240 :                 if (catalog == opt->optcontext)
     240             :                 {
     241          52 :                     has_valid_options = true;
     242          52 :                     updateClosestMatch(&match_state, opt->optname);
     243             :                 }
     244             :             }
     245             : 
     246          16 :             closest_match = getClosestMatch(&match_state);
     247          16 :             ereport(ERROR,
     248             :                     (errcode(ERRCODE_FDW_INVALID_OPTION_NAME),
     249             :                      errmsg("invalid option \"%s\"", def->defname),
     250             :                      has_valid_options ? closest_match ?
     251             :                      errhint("Perhaps you meant the option \"%s\".",
     252             :                              closest_match) : 0 :
     253             :                      errhint("There are no valid options in this context.")));
     254             :         }
     255             : 
     256             :         /*
     257             :          * Separate out filename, program, and column-specific options, since
     258             :          * ProcessCopyOptions won't accept them.
     259             :          */
     260         282 :         if (strcmp(def->defname, "filename") == 0 ||
     261         250 :             strcmp(def->defname, "program") == 0)
     262             :         {
     263          32 :             if (filename)
     264           0 :                 ereport(ERROR,
     265             :                         (errcode(ERRCODE_SYNTAX_ERROR),
     266             :                          errmsg("conflicting or redundant options")));
     267             : 
     268             :             /*
     269             :              * Check permissions for changing which file or program is used by
     270             :              * the file_fdw.
     271             :              *
     272             :              * Only members of the role 'pg_read_server_files' are allowed to
     273             :              * set the 'filename' option of a file_fdw foreign table, while
     274             :              * only members of the role 'pg_execute_server_program' are
     275             :              * allowed to set the 'program' option.  This is because we don't
     276             :              * want regular users to be able to control which file gets read
     277             :              * or which program gets executed.
     278             :              *
     279             :              * Putting this sort of permissions check in a validator is a bit
     280             :              * of a crock, but there doesn't seem to be any other place that
     281             :              * can enforce the check more cleanly.
     282             :              *
     283             :              * Note that the valid_options[] array disallows setting filename
     284             :              * and program at any options level other than foreign table ---
     285             :              * otherwise there'd still be a security hole.
     286             :              */
     287          32 :             if (strcmp(def->defname, "filename") == 0 &&
     288          32 :                 !has_privs_of_role(GetUserId(), ROLE_PG_READ_SERVER_FILES))
     289           2 :                 ereport(ERROR,
     290             :                         (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
     291             :                          errmsg("permission denied to set the \"%s\" option of a file_fdw foreign table",
     292             :                                 "filename"),
     293             :                          errdetail("Only roles with privileges of the \"%s\" role may set this option.",
     294             :                                    "pg_read_server_files")));
     295             : 
     296          30 :             if (strcmp(def->defname, "program") == 0 &&
     297           0 :                 !has_privs_of_role(GetUserId(), ROLE_PG_EXECUTE_SERVER_PROGRAM))
     298           0 :                 ereport(ERROR,
     299             :                         (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
     300             :                          errmsg("permission denied to set the \"%s\" option of a file_fdw foreign table",
     301             :                                 "program"),
     302             :                          errdetail("Only roles with privileges of the \"%s\" role may set this option.",
     303             :                                    "pg_execute_server_program")));
     304             : 
     305          30 :             filename = defGetString(def);
     306             :         }
     307             : 
     308             :         /*
     309             :          * force_not_null is a boolean option; after validation we can discard
     310             :          * it - it will be retrieved later in get_file_fdw_attribute_options()
     311             :          */
     312         250 :         else if (strcmp(def->defname, "force_not_null") == 0)
     313             :         {
     314           8 :             if (force_not_null)
     315           0 :                 ereport(ERROR,
     316             :                         (errcode(ERRCODE_SYNTAX_ERROR),
     317             :                          errmsg("conflicting or redundant options"),
     318             :                          errhint("Option \"force_not_null\" supplied more than once for a column.")));
     319           8 :             force_not_null = def;
     320             :             /* Don't care what the value is, as long as it's a legal boolean */
     321           8 :             (void) defGetBoolean(def);
     322             :         }
     323             :         /* See comments for force_not_null above */
     324         242 :         else if (strcmp(def->defname, "force_null") == 0)
     325             :         {
     326           8 :             if (force_null)
     327           0 :                 ereport(ERROR,
     328             :                         (errcode(ERRCODE_SYNTAX_ERROR),
     329             :                          errmsg("conflicting or redundant options"),
     330             :                          errhint("Option \"force_null\" supplied more than once for a column.")));
     331           8 :             force_null = def;
     332           8 :             (void) defGetBoolean(def);
     333             :         }
     334             :         else
     335         234 :             other_options = lappend(other_options, def);
     336             :     }
     337             : 
     338             :     /*
     339             :      * Now apply the core COPY code's validation logic for more checks.
     340             :      */
     341         104 :     ProcessCopyOptions(NULL, NULL, true, other_options);
     342             : 
     343             :     /*
     344             :      * Either filename or program option is required for file_fdw foreign
     345             :      * tables.
     346             :      */
     347          54 :     if (catalog == ForeignTableRelationId && filename == NULL)
     348           2 :         ereport(ERROR,
     349             :                 (errcode(ERRCODE_FDW_DYNAMIC_PARAMETER_VALUE_NEEDED),
     350             :                  errmsg("either filename or program is required for file_fdw foreign tables")));
     351             : 
     352          52 :     PG_RETURN_VOID();
     353             : }
     354             : 
     355             : /*
     356             :  * Check if the provided option is one of the valid options.
     357             :  * context is the Oid of the catalog holding the object the option is for.
     358             :  */
     359             : static bool
     360         298 : is_valid_option(const char *option, Oid context)
     361             : {
     362             :     const struct FileFdwOption *opt;
     363             : 
     364        1876 :     for (opt = valid_options; opt->optname; opt++)
     365             :     {
     366        1860 :         if (context == opt->optcontext && strcmp(opt->optname, option) == 0)
     367         282 :             return true;
     368             :     }
     369          16 :     return false;
     370             : }
     371             : 
     372             : /*
     373             :  * Fetch the options for a file_fdw foreign table.
     374             :  *
     375             :  * We have to separate out filename/program from the other options because
     376             :  * those must not appear in the options list passed to the core COPY code.
     377             :  */
     378             : static void
     379         154 : fileGetOptions(Oid foreigntableid,
     380             :                char **filename, bool *is_program, List **other_options)
     381             : {
     382             :     ForeignTable *table;
     383             :     ForeignServer *server;
     384             :     ForeignDataWrapper *wrapper;
     385             :     List       *options;
     386             :     ListCell   *lc;
     387             : 
     388             :     /*
     389             :      * Extract options from FDW objects.  We ignore user mappings because
     390             :      * file_fdw doesn't have any options that can be specified there.
     391             :      *
     392             :      * (XXX Actually, given the current contents of valid_options[], there's
     393             :      * no point in examining anything except the foreign table's own options.
     394             :      * Simplify?)
     395             :      */
     396         154 :     table = GetForeignTable(foreigntableid);
     397         154 :     server = GetForeignServer(table->serverid);
     398         154 :     wrapper = GetForeignDataWrapper(server->fdwid);
     399             : 
     400         154 :     options = NIL;
     401         154 :     options = list_concat(options, wrapper->options);
     402         154 :     options = list_concat(options, server->options);
     403         154 :     options = list_concat(options, table->options);
     404         154 :     options = list_concat(options, get_file_fdw_attribute_options(foreigntableid));
     405             : 
     406             :     /*
     407             :      * Separate out the filename or program option (we assume there is only
     408             :      * one).
     409             :      */
     410         154 :     *filename = NULL;
     411         154 :     *is_program = false;
     412         308 :     foreach(lc, options)
     413             :     {
     414         308 :         DefElem    *def = (DefElem *) lfirst(lc);
     415             : 
     416         308 :         if (strcmp(def->defname, "filename") == 0)
     417             :         {
     418         154 :             *filename = defGetString(def);
     419         154 :             options = foreach_delete_current(options, lc);
     420         154 :             break;
     421             :         }
     422         154 :         else if (strcmp(def->defname, "program") == 0)
     423             :         {
     424           0 :             *filename = defGetString(def);
     425           0 :             *is_program = true;
     426           0 :             options = foreach_delete_current(options, lc);
     427           0 :             break;
     428             :         }
     429             :     }
     430             : 
     431             :     /*
     432             :      * The validator should have checked that filename or program was included
     433             :      * in the options, but check again, just in case.
     434             :      */
     435         154 :     if (*filename == NULL)
     436           0 :         elog(ERROR, "either filename or program is required for file_fdw foreign tables");
     437             : 
     438         154 :     *other_options = options;
     439         154 : }
     440             : 
     441             : /*
     442             :  * Retrieve per-column generic options from pg_attribute and construct a list
     443             :  * of DefElems representing them.
     444             :  *
     445             :  * At the moment we only have "force_not_null", and "force_null",
     446             :  * which should each be combined into a single DefElem listing all such
     447             :  * columns, since that's what COPY expects.
     448             :  */
     449             : static List *
     450         154 : get_file_fdw_attribute_options(Oid relid)
     451             : {
     452             :     Relation    rel;
     453             :     TupleDesc   tupleDesc;
     454             :     AttrNumber  natts;
     455             :     AttrNumber  attnum;
     456         154 :     List       *fnncolumns = NIL;
     457         154 :     List       *fncolumns = NIL;
     458             : 
     459         154 :     List       *options = NIL;
     460             : 
     461         154 :     rel = table_open(relid, AccessShareLock);
     462         154 :     tupleDesc = RelationGetDescr(rel);
     463         154 :     natts = tupleDesc->natts;
     464             : 
     465             :     /* Retrieve FDW options for all user-defined attributes. */
     466         486 :     for (attnum = 1; attnum <= natts; attnum++)
     467             :     {
     468         332 :         Form_pg_attribute attr = TupleDescAttr(tupleDesc, attnum - 1);
     469             :         List       *column_options;
     470             :         ListCell   *lc;
     471             : 
     472             :         /* Skip dropped attributes. */
     473         332 :         if (attr->attisdropped)
     474           0 :             continue;
     475             : 
     476         332 :         column_options = GetForeignColumnOptions(relid, attnum);
     477         364 :         foreach(lc, column_options)
     478             :         {
     479          32 :             DefElem    *def = (DefElem *) lfirst(lc);
     480             : 
     481          32 :             if (strcmp(def->defname, "force_not_null") == 0)
     482             :             {
     483          16 :                 if (defGetBoolean(def))
     484             :                 {
     485           8 :                     char       *attname = pstrdup(NameStr(attr->attname));
     486             : 
     487           8 :                     fnncolumns = lappend(fnncolumns, makeString(attname));
     488             :                 }
     489             :             }
     490          16 :             else if (strcmp(def->defname, "force_null") == 0)
     491             :             {
     492          16 :                 if (defGetBoolean(def))
     493             :                 {
     494           8 :                     char       *attname = pstrdup(NameStr(attr->attname));
     495             : 
     496           8 :                     fncolumns = lappend(fncolumns, makeString(attname));
     497             :                 }
     498             :             }
     499             :             /* maybe in future handle other column options here */
     500             :         }
     501             :     }
     502             : 
     503         154 :     table_close(rel, AccessShareLock);
     504             : 
     505             :     /*
     506             :      * Return DefElem only when some column(s) have force_not_null /
     507             :      * force_null options set
     508             :      */
     509         154 :     if (fnncolumns != NIL)
     510           8 :         options = lappend(options, makeDefElem("force_not_null", (Node *) fnncolumns, -1));
     511             : 
     512         154 :     if (fncolumns != NIL)
     513           8 :         options = lappend(options, makeDefElem("force_null", (Node *) fncolumns, -1));
     514             : 
     515         154 :     return options;
     516             : }
     517             : 
     518             : /*
     519             :  * fileGetForeignRelSize
     520             :  *      Obtain relation size estimates for a foreign table
     521             :  */
     522             : static void
     523          80 : fileGetForeignRelSize(PlannerInfo *root,
     524             :                       RelOptInfo *baserel,
     525             :                       Oid foreigntableid)
     526             : {
     527             :     FileFdwPlanState *fdw_private;
     528             : 
     529             :     /*
     530             :      * Fetch options.  We only need filename (or program) at this point, but
     531             :      * we might as well get everything and not need to re-fetch it later in
     532             :      * planning.
     533             :      */
     534          80 :     fdw_private = (FileFdwPlanState *) palloc(sizeof(FileFdwPlanState));
     535          80 :     fileGetOptions(foreigntableid,
     536             :                    &fdw_private->filename,
     537             :                    &fdw_private->is_program,
     538             :                    &fdw_private->options);
     539          80 :     baserel->fdw_private = fdw_private;
     540             : 
     541             :     /* Estimate relation size */
     542          80 :     estimate_size(root, baserel, fdw_private);
     543          80 : }
     544             : 
     545             : /*
     546             :  * fileGetForeignPaths
     547             :  *      Create possible access paths for a scan on the foreign table
     548             :  *
     549             :  *      Currently we don't support any push-down feature, so there is only one
     550             :  *      possible access path, which simply returns all records in the order in
     551             :  *      the data file.
     552             :  */
     553             : static void
     554          80 : fileGetForeignPaths(PlannerInfo *root,
     555             :                     RelOptInfo *baserel,
     556             :                     Oid foreigntableid)
     557             : {
     558          80 :     FileFdwPlanState *fdw_private = (FileFdwPlanState *) baserel->fdw_private;
     559             :     Cost        startup_cost;
     560             :     Cost        total_cost;
     561             :     List       *columns;
     562          80 :     List       *coptions = NIL;
     563             : 
     564             :     /* Decide whether to selectively perform binary conversion */
     565          80 :     if (check_selective_binary_conversion(baserel,
     566             :                                           foreigntableid,
     567             :                                           &columns))
     568           8 :         coptions = list_make1(makeDefElem("convert_selectively",
     569             :                                           (Node *) columns, -1));
     570             : 
     571             :     /* Estimate costs */
     572          80 :     estimate_costs(root, baserel, fdw_private,
     573             :                    &startup_cost, &total_cost);
     574             : 
     575             :     /*
     576             :      * Create a ForeignPath node and add it as only possible path.  We use the
     577             :      * fdw_private list of the path to carry the convert_selectively option;
     578             :      * it will be propagated into the fdw_private list of the Plan node.
     579             :      *
     580             :      * We don't support pushing join clauses into the quals of this path, but
     581             :      * it could still have required parameterization due to LATERAL refs in
     582             :      * its tlist.
     583             :      */
     584          80 :     add_path(baserel, (Path *)
     585          80 :              create_foreignscan_path(root, baserel,
     586             :                                      NULL,  /* default pathtarget */
     587             :                                      baserel->rows,
     588             :                                      0,
     589             :                                      startup_cost,
     590             :                                      total_cost,
     591             :                                      NIL,   /* no pathkeys */
     592             :                                      baserel->lateral_relids,
     593             :                                      NULL,  /* no extra plan */
     594             :                                      NIL,   /* no fdw_restrictinfo list */
     595             :                                      coptions));
     596             : 
     597             :     /*
     598             :      * If data file was sorted, and we knew it somehow, we could insert
     599             :      * appropriate pathkeys into the ForeignPath node to tell the planner
     600             :      * that.
     601             :      */
     602          80 : }
     603             : 
     604             : /*
     605             :  * fileGetForeignPlan
     606             :  *      Create a ForeignScan plan node for scanning the foreign table
     607             :  */
     608             : static ForeignScan *
     609          80 : fileGetForeignPlan(PlannerInfo *root,
     610             :                    RelOptInfo *baserel,
     611             :                    Oid foreigntableid,
     612             :                    ForeignPath *best_path,
     613             :                    List *tlist,
     614             :                    List *scan_clauses,
     615             :                    Plan *outer_plan)
     616             : {
     617          80 :     Index       scan_relid = baserel->relid;
     618             : 
     619             :     /*
     620             :      * We have no native ability to evaluate restriction clauses, so we just
     621             :      * put all the scan_clauses into the plan node's qual list for the
     622             :      * executor to check.  So all we have to do here is strip RestrictInfo
     623             :      * nodes from the clauses and ignore pseudoconstants (which will be
     624             :      * handled elsewhere).
     625             :      */
     626          80 :     scan_clauses = extract_actual_clauses(scan_clauses, false);
     627             : 
     628             :     /* Create the ForeignScan node */
     629          80 :     return make_foreignscan(tlist,
     630             :                             scan_clauses,
     631             :                             scan_relid,
     632             :                             NIL,    /* no expressions to evaluate */
     633             :                             best_path->fdw_private,
     634             :                             NIL,    /* no custom tlist */
     635             :                             NIL,    /* no remote quals */
     636             :                             outer_plan);
     637             : }
     638             : 
     639             : /*
     640             :  * fileExplainForeignScan
     641             :  *      Produce extra output for EXPLAIN
     642             :  */
     643             : static void
     644           6 : fileExplainForeignScan(ForeignScanState *node, ExplainState *es)
     645             : {
     646             :     char       *filename;
     647             :     bool        is_program;
     648             :     List       *options;
     649             : 
     650             :     /* Fetch options --- we only need filename and is_program at this point */
     651           6 :     fileGetOptions(RelationGetRelid(node->ss.ss_currentRelation),
     652             :                    &filename, &is_program, &options);
     653             : 
     654           6 :     if (is_program)
     655           0 :         ExplainPropertyText("Foreign Program", filename, es);
     656             :     else
     657           6 :         ExplainPropertyText("Foreign File", filename, es);
     658             : 
     659             :     /* Suppress file size if we're not showing cost details */
     660           6 :     if (es->costs)
     661             :     {
     662             :         struct stat stat_buf;
     663             : 
     664           0 :         if (!is_program &&
     665           0 :             stat(filename, &stat_buf) == 0)
     666           0 :             ExplainPropertyInteger("Foreign File Size", "b",
     667           0 :                                    (int64) stat_buf.st_size, es);
     668             :     }
     669           6 : }
     670             : 
     671             : /*
     672             :  * fileBeginForeignScan
     673             :  *      Initiate access to the file by creating CopyState
     674             :  */
     675             : static void
     676          70 : fileBeginForeignScan(ForeignScanState *node, int eflags)
     677             : {
     678          70 :     ForeignScan *plan = (ForeignScan *) node->ss.ps.plan;
     679             :     char       *filename;
     680             :     bool        is_program;
     681             :     List       *options;
     682             :     CopyFromState cstate;
     683             :     FileFdwExecutionState *festate;
     684             : 
     685             :     /*
     686             :      * Do nothing in EXPLAIN (no ANALYZE) case.  node->fdw_state stays NULL.
     687             :      */
     688          70 :     if (eflags & EXEC_FLAG_EXPLAIN_ONLY)
     689           6 :         return;
     690             : 
     691             :     /* Fetch options of foreign table */
     692          64 :     fileGetOptions(RelationGetRelid(node->ss.ss_currentRelation),
     693             :                    &filename, &is_program, &options);
     694             : 
     695             :     /* Add any options from the plan (currently only convert_selectively) */
     696          64 :     options = list_concat(options, plan->fdw_private);
     697             : 
     698             :     /*
     699             :      * Create CopyState from FDW options.  We always acquire all columns, so
     700             :      * as to match the expected ScanTupleSlot signature.
     701             :      */
     702          64 :     cstate = BeginCopyFrom(NULL,
     703             :                            node->ss.ss_currentRelation,
     704             :                            NULL,
     705             :                            filename,
     706             :                            is_program,
     707             :                            NULL,
     708             :                            NIL,
     709             :                            options);
     710             : 
     711             :     /*
     712             :      * Save state in node->fdw_state.  We must save enough information to call
     713             :      * BeginCopyFrom() again.
     714             :      */
     715          62 :     festate = (FileFdwExecutionState *) palloc(sizeof(FileFdwExecutionState));
     716          62 :     festate->filename = filename;
     717          62 :     festate->is_program = is_program;
     718          62 :     festate->options = options;
     719          62 :     festate->cstate = cstate;
     720             : 
     721          62 :     node->fdw_state = festate;
     722             : }
     723             : 
     724             : /*
     725             :  * fileIterateForeignScan
     726             :  *      Read next record from the data file and store it into the
     727             :  *      ScanTupleSlot as a virtual tuple
     728             :  */
     729             : static TupleTableSlot *
     730         246 : fileIterateForeignScan(ForeignScanState *node)
     731             : {
     732         246 :     FileFdwExecutionState *festate = (FileFdwExecutionState *) node->fdw_state;
     733         246 :     EState     *estate = CreateExecutorState();
     734             :     ExprContext *econtext;
     735         246 :     MemoryContext oldcontext = CurrentMemoryContext;
     736         246 :     TupleTableSlot *slot = node->ss.ss_ScanTupleSlot;
     737         246 :     CopyFromState cstate = festate->cstate;
     738             :     ErrorContextCallback errcallback;
     739             : 
     740             :     /* Set up callback to identify error line number. */
     741         246 :     errcallback.callback = CopyFromErrorCallback;
     742         246 :     errcallback.arg = cstate;
     743         246 :     errcallback.previous = error_context_stack;
     744         246 :     error_context_stack = &errcallback;
     745             : 
     746             :     /*
     747             :      * We pass ExprContext because there might be a use of the DEFAULT option
     748             :      * in COPY FROM, so we may need to evaluate default expressions.
     749             :      */
     750         246 :     econtext = GetPerTupleExprContext(estate);
     751             : 
     752         260 : retry:
     753             : 
     754             :     /*
     755             :      * DEFAULT expressions need to be evaluated in a per-tuple context, so
     756             :      * switch in case we are doing that.
     757             :      */
     758         260 :     MemoryContextSwitchTo(GetPerTupleMemoryContext(estate));
     759             : 
     760             :     /*
     761             :      * The protocol for loading a virtual tuple into a slot is first
     762             :      * ExecClearTuple, then fill the values/isnull arrays, then
     763             :      * ExecStoreVirtualTuple.  If we don't find another row in the file, we
     764             :      * just skip the last step, leaving the slot empty as required.
     765             :      *
     766             :      */
     767         260 :     ExecClearTuple(slot);
     768             : 
     769         260 :     if (NextCopyFrom(cstate, econtext, slot->tts_values, slot->tts_isnull))
     770             :     {
     771         196 :         if (cstate->opts.on_error == COPY_ON_ERROR_IGNORE &&
     772          32 :             cstate->escontext->error_occurred)
     773             :         {
     774             :             /*
     775             :              * Soft error occurred, skip this tuple and just make
     776             :              * ErrorSaveContext ready for the next NextCopyFrom. Since we
     777             :              * don't set details_wanted and error_data is not to be filled,
     778             :              * just resetting error_occurred is enough.
     779             :              */
     780          16 :             cstate->escontext->error_occurred = false;
     781             : 
     782             :             /* Switch back to original memory context */
     783          16 :             MemoryContextSwitchTo(oldcontext);
     784             : 
     785             :             /*
     786             :              * Make sure we are interruptible while repeatedly calling
     787             :              * NextCopyFrom() until no soft error occurs.
     788             :              */
     789          16 :             CHECK_FOR_INTERRUPTS();
     790             : 
     791             :             /*
     792             :              * Reset the per-tuple exprcontext, to clean-up after expression
     793             :              * evaluations etc.
     794             :              */
     795          16 :             ResetPerTupleExprContext(estate);
     796             : 
     797          16 :             if (cstate->opts.reject_limit > 0 &&
     798           8 :                 cstate->num_errors > cstate->opts.reject_limit)
     799           2 :                 ereport(ERROR,
     800             :                         (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
     801             :                          errmsg("skipped more than REJECT_LIMIT (%" PRId64 ") rows due to data type incompatibility",
     802             :                                 cstate->opts.reject_limit)));
     803             : 
     804             :             /* Repeat NextCopyFrom() until no soft error occurs */
     805          14 :             goto retry;
     806             :         }
     807             : 
     808         180 :         ExecStoreVirtualTuple(slot);
     809             :     }
     810             : 
     811             :     /* Switch back to original memory context */
     812         240 :     MemoryContextSwitchTo(oldcontext);
     813             : 
     814             :     /* Remove error callback. */
     815         240 :     error_context_stack = errcallback.previous;
     816             : 
     817         240 :     return slot;
     818             : }
     819             : 
     820             : /*
     821             :  * fileReScanForeignScan
     822             :  *      Rescan table, possibly with new parameters
     823             :  */
     824             : static void
     825           6 : fileReScanForeignScan(ForeignScanState *node)
     826             : {
     827           6 :     FileFdwExecutionState *festate = (FileFdwExecutionState *) node->fdw_state;
     828             : 
     829           6 :     EndCopyFrom(festate->cstate);
     830             : 
     831          12 :     festate->cstate = BeginCopyFrom(NULL,
     832             :                                     node->ss.ss_currentRelation,
     833             :                                     NULL,
     834           6 :                                     festate->filename,
     835           6 :                                     festate->is_program,
     836             :                                     NULL,
     837             :                                     NIL,
     838             :                                     festate->options);
     839           6 : }
     840             : 
     841             : /*
     842             :  * fileEndForeignScan
     843             :  *      Finish scanning foreign table and dispose objects used for this scan
     844             :  */
     845             : static void
     846          62 : fileEndForeignScan(ForeignScanState *node)
     847             : {
     848          62 :     FileFdwExecutionState *festate = (FileFdwExecutionState *) node->fdw_state;
     849             : 
     850             :     /* if festate is NULL, we are in EXPLAIN; nothing to do */
     851          62 :     if (!festate)
     852           6 :         return;
     853             : 
     854          56 :     if (festate->cstate->opts.on_error == COPY_ON_ERROR_IGNORE &&
     855           6 :         festate->cstate->num_errors > 0 &&
     856           6 :         festate->cstate->opts.log_verbosity >= COPY_LOG_VERBOSITY_DEFAULT)
     857           2 :         ereport(NOTICE,
     858             :                 errmsg_plural("%" PRIu64 " row was skipped due to data type incompatibility",
     859             :                               "%" PRIu64 " rows were skipped due to data type incompatibility",
     860             :                               festate->cstate->num_errors,
     861             :                               festate->cstate->num_errors));
     862             : 
     863          56 :     EndCopyFrom(festate->cstate);
     864             : }
     865             : 
     866             : /*
     867             :  * fileAnalyzeForeignTable
     868             :  *      Test whether analyzing this foreign table is supported
     869             :  */
     870             : static bool
     871           2 : fileAnalyzeForeignTable(Relation relation,
     872             :                         AcquireSampleRowsFunc *func,
     873             :                         BlockNumber *totalpages)
     874             : {
     875             :     char       *filename;
     876             :     bool        is_program;
     877             :     List       *options;
     878             :     struct stat stat_buf;
     879             : 
     880             :     /* Fetch options of foreign table */
     881           2 :     fileGetOptions(RelationGetRelid(relation), &filename, &is_program, &options);
     882             : 
     883             :     /*
     884             :      * If this is a program instead of a file, just return false to skip
     885             :      * analyzing the table.  We could run the program and collect stats on
     886             :      * whatever it currently returns, but it seems likely that in such cases
     887             :      * the output would be too volatile for the stats to be useful.  Maybe
     888             :      * there should be an option to enable doing this?
     889             :      */
     890           2 :     if (is_program)
     891           0 :         return false;
     892             : 
     893             :     /*
     894             :      * Get size of the file.  (XXX if we fail here, would it be better to just
     895             :      * return false to skip analyzing the table?)
     896             :      */
     897           2 :     if (stat(filename, &stat_buf) < 0)
     898           0 :         ereport(ERROR,
     899             :                 (errcode_for_file_access(),
     900             :                  errmsg("could not stat file \"%s\": %m",
     901             :                         filename)));
     902             : 
     903             :     /*
     904             :      * Convert size to pages.  Must return at least 1 so that we can tell
     905             :      * later on that pg_class.relpages is not default.
     906             :      */
     907           2 :     *totalpages = (stat_buf.st_size + (BLCKSZ - 1)) / BLCKSZ;
     908           2 :     if (*totalpages < 1)
     909           0 :         *totalpages = 1;
     910             : 
     911           2 :     *func = file_acquire_sample_rows;
     912             : 
     913           2 :     return true;
     914             : }
     915             : 
     916             : /*
     917             :  * fileIsForeignScanParallelSafe
     918             :  *      Reading a file, or external program, in a parallel worker should work
     919             :  *      just the same as reading it in the leader, so mark scans safe.
     920             :  */
     921             : static bool
     922          72 : fileIsForeignScanParallelSafe(PlannerInfo *root, RelOptInfo *rel,
     923             :                               RangeTblEntry *rte)
     924             : {
     925          72 :     return true;
     926             : }
     927             : 
     928             : /*
     929             :  * check_selective_binary_conversion
     930             :  *
     931             :  * Check to see if it's useful to convert only a subset of the file's columns
     932             :  * to binary.  If so, construct a list of the column names to be converted,
     933             :  * return that at *columns, and return true.  (Note that it's possible to
     934             :  * determine that no columns need be converted, for instance with a COUNT(*)
     935             :  * query.  So we can't use returning a NIL list to indicate failure.)
     936             :  */
     937             : static bool
     938          80 : check_selective_binary_conversion(RelOptInfo *baserel,
     939             :                                   Oid foreigntableid,
     940             :                                   List **columns)
     941             : {
     942             :     ForeignTable *table;
     943             :     ListCell   *lc;
     944             :     Relation    rel;
     945             :     TupleDesc   tupleDesc;
     946             :     int         attidx;
     947          80 :     Bitmapset  *attrs_used = NULL;
     948          80 :     bool        has_wholerow = false;
     949             :     int         numattrs;
     950             :     int         i;
     951             : 
     952          80 :     *columns = NIL;             /* default result */
     953             : 
     954             :     /*
     955             :      * Check format of the file.  If binary format, this is irrelevant.
     956             :      */
     957          80 :     table = GetForeignTable(foreigntableid);
     958          80 :     foreach(lc, table->options)
     959             :     {
     960          80 :         DefElem    *def = (DefElem *) lfirst(lc);
     961             : 
     962          80 :         if (strcmp(def->defname, "format") == 0)
     963             :         {
     964          80 :             char       *format = defGetString(def);
     965             : 
     966          80 :             if (strcmp(format, "binary") == 0)
     967           0 :                 return false;
     968          80 :             break;
     969             :         }
     970             :     }
     971             : 
     972             :     /* Collect all the attributes needed for joins or final output. */
     973          80 :     pull_varattnos((Node *) baserel->reltarget->exprs, baserel->relid,
     974             :                    &attrs_used);
     975             : 
     976             :     /* Add all the attributes used by restriction clauses. */
     977          96 :     foreach(lc, baserel->baserestrictinfo)
     978             :     {
     979          16 :         RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
     980             : 
     981          16 :         pull_varattnos((Node *) rinfo->clause, baserel->relid,
     982             :                        &attrs_used);
     983             :     }
     984             : 
     985             :     /* Convert attribute numbers to column names. */
     986          80 :     rel = table_open(foreigntableid, AccessShareLock);
     987          80 :     tupleDesc = RelationGetDescr(rel);
     988             : 
     989          80 :     attidx = -1;
     990         254 :     while ((attidx = bms_next_member(attrs_used, attidx)) >= 0)
     991             :     {
     992             :         /* attidx is zero-based, attnum is the normal attribute number */
     993         182 :         AttrNumber  attnum = attidx + FirstLowInvalidHeapAttributeNumber;
     994             : 
     995         182 :         if (attnum == 0)
     996             :         {
     997           8 :             has_wholerow = true;
     998           8 :             break;
     999             :         }
    1000             : 
    1001             :         /* Ignore system attributes. */
    1002         174 :         if (attnum < 0)
    1003          26 :             continue;
    1004             : 
    1005             :         /* Get user attributes. */
    1006         148 :         if (attnum > 0)
    1007             :         {
    1008         148 :             Form_pg_attribute attr = TupleDescAttr(tupleDesc, attnum - 1);
    1009         148 :             char       *attname = NameStr(attr->attname);
    1010             : 
    1011             :             /* Skip dropped attributes (probably shouldn't see any here). */
    1012         148 :             if (attr->attisdropped)
    1013           0 :                 continue;
    1014             : 
    1015             :             /*
    1016             :              * Skip generated columns (COPY won't accept them in the column
    1017             :              * list)
    1018             :              */
    1019         148 :             if (attr->attgenerated)
    1020           2 :                 continue;
    1021         146 :             *columns = lappend(*columns, makeString(pstrdup(attname)));
    1022             :         }
    1023             :     }
    1024             : 
    1025             :     /* Count non-dropped user attributes while we have the tupdesc. */
    1026          80 :     numattrs = 0;
    1027         252 :     for (i = 0; i < tupleDesc->natts; i++)
    1028             :     {
    1029         172 :         Form_pg_attribute attr = TupleDescAttr(tupleDesc, i);
    1030             : 
    1031         172 :         if (attr->attisdropped)
    1032           0 :             continue;
    1033         172 :         numattrs++;
    1034             :     }
    1035             : 
    1036          80 :     table_close(rel, AccessShareLock);
    1037             : 
    1038             :     /* If there's a whole-row reference, fail: we need all the columns. */
    1039          80 :     if (has_wholerow)
    1040             :     {
    1041           8 :         *columns = NIL;
    1042           8 :         return false;
    1043             :     }
    1044             : 
    1045             :     /* If all the user attributes are needed, fail. */
    1046          72 :     if (numattrs == list_length(*columns))
    1047             :     {
    1048          64 :         *columns = NIL;
    1049          64 :         return false;
    1050             :     }
    1051             : 
    1052           8 :     return true;
    1053             : }
    1054             : 
    1055             : /*
    1056             :  * Estimate size of a foreign table.
    1057             :  *
    1058             :  * The main result is returned in baserel->rows.  We also set
    1059             :  * fdw_private->pages and fdw_private->ntuples for later use in the cost
    1060             :  * calculation.
    1061             :  */
    1062             : static void
    1063          80 : estimate_size(PlannerInfo *root, RelOptInfo *baserel,
    1064             :               FileFdwPlanState *fdw_private)
    1065             : {
    1066             :     struct stat stat_buf;
    1067             :     BlockNumber pages;
    1068             :     double      ntuples;
    1069             :     double      nrows;
    1070             : 
    1071             :     /*
    1072             :      * Get size of the file.  It might not be there at plan time, though, in
    1073             :      * which case we have to use a default estimate.  We also have to fall
    1074             :      * back to the default if using a program as the input.
    1075             :      */
    1076          80 :     if (fdw_private->is_program || stat(fdw_private->filename, &stat_buf) < 0)
    1077           0 :         stat_buf.st_size = 10 * BLCKSZ;
    1078             : 
    1079             :     /*
    1080             :      * Convert size to pages for use in I/O cost estimate later.
    1081             :      */
    1082          80 :     pages = (stat_buf.st_size + (BLCKSZ - 1)) / BLCKSZ;
    1083          80 :     if (pages < 1)
    1084           0 :         pages = 1;
    1085          80 :     fdw_private->pages = pages;
    1086             : 
    1087             :     /*
    1088             :      * Estimate the number of tuples in the file.
    1089             :      */
    1090          80 :     if (baserel->tuples >= 0 && baserel->pages > 0)
    1091           0 :     {
    1092             :         /*
    1093             :          * We have # of pages and # of tuples from pg_class (that is, from a
    1094             :          * previous ANALYZE), so compute a tuples-per-page estimate and scale
    1095             :          * that by the current file size.
    1096             :          */
    1097             :         double      density;
    1098             : 
    1099           0 :         density = baserel->tuples / (double) baserel->pages;
    1100           0 :         ntuples = clamp_row_est(density * (double) pages);
    1101             :     }
    1102             :     else
    1103             :     {
    1104             :         /*
    1105             :          * Otherwise we have to fake it.  We back into this estimate using the
    1106             :          * planner's idea of the relation width; which is bogus if not all
    1107             :          * columns are being read, not to mention that the text representation
    1108             :          * of a row probably isn't the same size as its internal
    1109             :          * representation.  Possibly we could do something better, but the
    1110             :          * real answer to anyone who complains is "ANALYZE" ...
    1111             :          */
    1112             :         int         tuple_width;
    1113             : 
    1114          80 :         tuple_width = MAXALIGN(baserel->reltarget->width) +
    1115             :             MAXALIGN(SizeofHeapTupleHeader);
    1116          80 :         ntuples = clamp_row_est((double) stat_buf.st_size /
    1117          80 :                                 (double) tuple_width);
    1118             :     }
    1119          80 :     fdw_private->ntuples = ntuples;
    1120             : 
    1121             :     /*
    1122             :      * Now estimate the number of rows returned by the scan after applying the
    1123             :      * baserestrictinfo quals.
    1124             :      */
    1125          80 :     nrows = ntuples *
    1126          80 :         clauselist_selectivity(root,
    1127             :                                baserel->baserestrictinfo,
    1128             :                                0,
    1129             :                                JOIN_INNER,
    1130             :                                NULL);
    1131             : 
    1132          80 :     nrows = clamp_row_est(nrows);
    1133             : 
    1134             :     /* Save the output-rows estimate for the planner */
    1135          80 :     baserel->rows = nrows;
    1136          80 : }
    1137             : 
    1138             : /*
    1139             :  * Estimate costs of scanning a foreign table.
    1140             :  *
    1141             :  * Results are returned in *startup_cost and *total_cost.
    1142             :  */
    1143             : static void
    1144          80 : estimate_costs(PlannerInfo *root, RelOptInfo *baserel,
    1145             :                FileFdwPlanState *fdw_private,
    1146             :                Cost *startup_cost, Cost *total_cost)
    1147             : {
    1148          80 :     BlockNumber pages = fdw_private->pages;
    1149          80 :     double      ntuples = fdw_private->ntuples;
    1150          80 :     Cost        run_cost = 0;
    1151             :     Cost        cpu_per_tuple;
    1152             : 
    1153             :     /*
    1154             :      * We estimate costs almost the same way as cost_seqscan(), thus assuming
    1155             :      * that I/O costs are equivalent to a regular table file of the same size.
    1156             :      * However, we take per-tuple CPU costs as 10x of a seqscan, to account
    1157             :      * for the cost of parsing records.
    1158             :      *
    1159             :      * In the case of a program source, this calculation is even more divorced
    1160             :      * from reality, but we have no good alternative; and it's not clear that
    1161             :      * the numbers we produce here matter much anyway, since there's only one
    1162             :      * access path for the rel.
    1163             :      */
    1164          80 :     run_cost += seq_page_cost * pages;
    1165             : 
    1166          80 :     *startup_cost = baserel->baserestrictcost.startup;
    1167          80 :     cpu_per_tuple = cpu_tuple_cost * 10 + baserel->baserestrictcost.per_tuple;
    1168          80 :     run_cost += cpu_per_tuple * ntuples;
    1169          80 :     *total_cost = *startup_cost + run_cost;
    1170          80 : }
    1171             : 
    1172             : /*
    1173             :  * file_acquire_sample_rows -- acquire a random sample of rows from the table
    1174             :  *
    1175             :  * Selected rows are returned in the caller-allocated array rows[],
    1176             :  * which must have at least targrows entries.
    1177             :  * The actual number of rows selected is returned as the function result.
    1178             :  * We also count the total number of rows in the file and return it into
    1179             :  * *totalrows.  Rows skipped due to on_error = 'ignore' are not included
    1180             :  * in this count.  Note that *totaldeadrows is always set to 0.
    1181             :  *
    1182             :  * Note that the returned list of rows is not always in order by physical
    1183             :  * position in the file.  Therefore, correlation estimates derived later
    1184             :  * may be meaningless, but it's OK because we don't use the estimates
    1185             :  * currently (the planner only pays attention to correlation for indexscans).
    1186             :  */
    1187             : static int
    1188           2 : file_acquire_sample_rows(Relation onerel, int elevel,
    1189             :                          HeapTuple *rows, int targrows,
    1190             :                          double *totalrows, double *totaldeadrows)
    1191             : {
    1192           2 :     int         numrows = 0;
    1193           2 :     double      rowstoskip = -1;    /* -1 means not set yet */
    1194             :     ReservoirStateData rstate;
    1195             :     TupleDesc   tupDesc;
    1196             :     Datum      *values;
    1197             :     bool       *nulls;
    1198             :     bool        found;
    1199             :     char       *filename;
    1200             :     bool        is_program;
    1201             :     List       *options;
    1202             :     CopyFromState cstate;
    1203             :     ErrorContextCallback errcallback;
    1204           2 :     MemoryContext oldcontext = CurrentMemoryContext;
    1205             :     MemoryContext tupcontext;
    1206             : 
    1207             :     Assert(onerel);
    1208             :     Assert(targrows > 0);
    1209             : 
    1210           2 :     tupDesc = RelationGetDescr(onerel);
    1211           2 :     values = (Datum *) palloc(tupDesc->natts * sizeof(Datum));
    1212           2 :     nulls = (bool *) palloc(tupDesc->natts * sizeof(bool));
    1213             : 
    1214             :     /* Fetch options of foreign table */
    1215           2 :     fileGetOptions(RelationGetRelid(onerel), &filename, &is_program, &options);
    1216             : 
    1217             :     /*
    1218             :      * Create CopyState from FDW options.
    1219             :      */
    1220           2 :     cstate = BeginCopyFrom(NULL, onerel, NULL, filename, is_program, NULL, NIL,
    1221             :                            options);
    1222             : 
    1223             :     /*
    1224             :      * Use per-tuple memory context to prevent leak of memory used to read
    1225             :      * rows from the file with Copy routines.
    1226             :      */
    1227           2 :     tupcontext = AllocSetContextCreate(CurrentMemoryContext,
    1228             :                                        "file_fdw temporary context",
    1229             :                                        ALLOCSET_DEFAULT_SIZES);
    1230             : 
    1231             :     /* Prepare for sampling rows */
    1232           2 :     reservoir_init_selection_state(&rstate, targrows);
    1233             : 
    1234             :     /* Set up callback to identify error line number. */
    1235           2 :     errcallback.callback = CopyFromErrorCallback;
    1236           2 :     errcallback.arg = cstate;
    1237           2 :     errcallback.previous = error_context_stack;
    1238           2 :     error_context_stack = &errcallback;
    1239             : 
    1240           2 :     *totalrows = 0;
    1241           2 :     *totaldeadrows = 0;
    1242             :     for (;;)
    1243             :     {
    1244             :         /* Check for user-requested abort or sleep */
    1245          10 :         vacuum_delay_point(true);
    1246             : 
    1247             :         /* Fetch next row */
    1248          10 :         MemoryContextReset(tupcontext);
    1249          10 :         MemoryContextSwitchTo(tupcontext);
    1250             : 
    1251          10 :         found = NextCopyFrom(cstate, NULL, values, nulls);
    1252             : 
    1253          10 :         MemoryContextSwitchTo(oldcontext);
    1254             : 
    1255          10 :         if (!found)
    1256           2 :             break;
    1257             : 
    1258           8 :         if (cstate->opts.on_error == COPY_ON_ERROR_IGNORE &&
    1259           8 :             cstate->escontext->error_occurred)
    1260             :         {
    1261             :             /*
    1262             :              * Soft error occurred, skip this tuple and just make
    1263             :              * ErrorSaveContext ready for the next NextCopyFrom. Since we
    1264             :              * don't set details_wanted and error_data is not to be filled,
    1265             :              * just resetting error_occurred is enough.
    1266             :              */
    1267           4 :             cstate->escontext->error_occurred = false;
    1268             : 
    1269             :             /* Repeat NextCopyFrom() until no soft error occurs */
    1270           4 :             continue;
    1271             :         }
    1272             : 
    1273             :         /*
    1274             :          * The first targrows sample rows are simply copied into the
    1275             :          * reservoir.  Then we start replacing tuples in the sample until we
    1276             :          * reach the end of the relation. This algorithm is from Jeff Vitter's
    1277             :          * paper (see more info in commands/analyze.c).
    1278             :          */
    1279           4 :         if (numrows < targrows)
    1280             :         {
    1281           4 :             rows[numrows++] = heap_form_tuple(tupDesc, values, nulls);
    1282             :         }
    1283             :         else
    1284             :         {
    1285             :             /*
    1286             :              * t in Vitter's paper is the number of records already processed.
    1287             :              * If we need to compute a new S value, we must use the
    1288             :              * not-yet-incremented value of totalrows as t.
    1289             :              */
    1290           0 :             if (rowstoskip < 0)
    1291           0 :                 rowstoskip = reservoir_get_next_S(&rstate, *totalrows, targrows);
    1292             : 
    1293           0 :             if (rowstoskip <= 0)
    1294             :             {
    1295             :                 /*
    1296             :                  * Found a suitable tuple, so save it, replacing one old tuple
    1297             :                  * at random
    1298             :                  */
    1299           0 :                 int         k = (int) (targrows * sampler_random_fract(&rstate.randstate));
    1300             : 
    1301             :                 Assert(k >= 0 && k < targrows);
    1302           0 :                 heap_freetuple(rows[k]);
    1303           0 :                 rows[k] = heap_form_tuple(tupDesc, values, nulls);
    1304             :             }
    1305             : 
    1306           0 :             rowstoskip -= 1;
    1307             :         }
    1308             : 
    1309           4 :         *totalrows += 1;
    1310             :     }
    1311             : 
    1312             :     /* Remove error callback. */
    1313           2 :     error_context_stack = errcallback.previous;
    1314             : 
    1315             :     /* Clean up. */
    1316           2 :     MemoryContextDelete(tupcontext);
    1317             : 
    1318           2 :     if (cstate->opts.on_error == COPY_ON_ERROR_IGNORE &&
    1319           2 :         cstate->num_errors > 0 &&
    1320           2 :         cstate->opts.log_verbosity >= COPY_LOG_VERBOSITY_DEFAULT)
    1321           0 :         ereport(NOTICE,
    1322             :                 errmsg_plural("%" PRIu64 " row was skipped due to data type incompatibility",
    1323             :                               "%" PRIu64 " rows were skipped due to data type incompatibility",
    1324             :                               cstate->num_errors,
    1325             :                               cstate->num_errors));
    1326             : 
    1327           2 :     EndCopyFrom(cstate);
    1328             : 
    1329           2 :     pfree(values);
    1330           2 :     pfree(nulls);
    1331             : 
    1332             :     /*
    1333             :      * Emit some interesting relation info
    1334             :      */
    1335           2 :     ereport(elevel,
    1336             :             (errmsg("\"%s\": file contains %.0f rows; "
    1337             :                     "%d rows in sample",
    1338             :                     RelationGetRelationName(onerel),
    1339             :                     *totalrows, numrows)));
    1340             : 
    1341           2 :     return numrows;
    1342             : }

Generated by: LCOV version 1.14