SPI_EXECUTE(3)

📦 version 13.23, in opensuse-leap 16.0

💅 NAME 💅

okay so bestie, welcome to SPI_execute — it executes a command. that's it, that's the whole vibe ✨

(this is SPI_EXECUTE(3) from the PostgreSQL 13.23 documentation, and yes i'm gonna walk you through every single bit of it, no cap 🫶)

🌸 SYNOPSIS 🌸

int SPI_execute(const char * command, bool read_only, long count)

like, three args, one int back. iconic signature fr fr 💅

✨ DESCRIPTION ✨

so SPI_execute runs the SQL command you hand it, for count rows. if read_only is true, the command has to be read-only — and honestly the execution overhead gets somewhat reduced, so that's a little win 🎀

big caveat energy: this function can only be called from a connected C function. not connected = not invited to the party, sorry 😭

🦋 the count situation 🦋

if count is zero, the command runs for all the rows it applies to. if count is greater than zero, then no more than count rows get retrieved — execution just stops once it hits the count, much like adding a LIMIT clause to the query. for example:

SPI_execute("SELECT * FROM foo", true, 5);

will retrieve at most 5 rows from the table. cute, right? but like — note that this limit is only effective when the command actually returns rows. for example:

SPI_execute("INSERT INTO foo SELECT * FROM bar", false, 5);

inserts all rows from bar, straight up ignoring the count parameter 💀. however, with

SPI_execute("INSERT INTO foo SELECT * FROM bar RETURNING *", false, 5);

at most 5 rows would be inserted, since execution would stop after the fifth RETURNING result row is retrieved. see how that works? the RETURNING is doing the heavy lifting 👑

💜 multiple commands in one string 💜

you can pass multiple commands in one string, bestie — SPI_execute returns the result for the command executed last. the count limit applies to each command separately (even though only the last result will actually be returned, which is giving main character energy only for the final one). and the limit is not applied to any hidden commands generated by rules. iykyk.

🌈 read_only = false (the read/write mode) 🌈

when read_only is false, SPI_execute increments the command counter and computes a new snapshot before executing each command in the string.

the snapshot does not actually change if the current transaction isolation level is SERIALIZABLE or REPEATABLE READ, but in READ COMMITTED mode the snapshot update allows each command to see the results of newly committed transactions from other sessions. this is essential for consistent behavior when the commands are modifying the database — like, genuinely essential, don't skip it 🔥

🧚 read_only = true (the read-only mode) 🧚

when read_only is true, SPI_execute does not update either the snapshot or the command counter, and it allows only plain SELECT commands to appear in the command string. the commands are executed using the snapshot previously established for the surrounding query.

this execution mode is somewhat faster than the read/write mode, due to eliminating per-command overhead. it also allows genuinely stable functions to be built: since successive executions will all use the same snapshot, there will be no change in the results. stable? we love her 🎀

⚠️ the mixing warning ⚠️

it is generally unwise to mix read-only and read-write commands within a single function using SPI; that could result in very confusing behavior, since the read-only queries would not see the results of any database updates done by the read-write queries.

like, don't do it. i'm begging. 🙈

🍓 results: SPI_processed and SPI_tuptable 🍓

the actual number of rows for which the (last) command was executed is returned in the global variable SPI_processed.

if the return value of the function is SPI_OK_SELECT, SPI_OK_INSERT_RETURNING, SPI_OK_DELETE_RETURNING, or SPI_OK_UPDATE_RETURNING, then you can use the global pointer SPITupleTable *SPI_tuptable to access the result rows. some utility commands (such as EXPLAIN) also return row sets, and SPI_tuptable will contain the result in these cases too.

some utility commands (COPY, CREATE TABLE AS) don't return a row set, so SPI_tuptable is NULL, but they still return the number of rows processed in SPI_processed. so like, not a total loss 💖

🌟 the struct, in all her glory 🌟

the structure SPITupleTable is defined thus:

typedef struct SPITupleTable
{
    /* Public members */
    TupleDesc   tupdesc;        /* tuple descriptor */
    HeapTuple  *vals;           /* array of tuples */
    uint64      numvals;        /* number of valid tuples */

    /* Private members, not intended for external callers */
    uint64      alloced;        /* allocated length of vals array */
    MemoryContext tuptabcxt;    /* memory context of result table */
    slist_node  next;           /* link for internal bookkeeping */
    SubTransactionId subid;     /* subxact in which tuptable was created */
} SPITupleTable;

the fields tupdesc, vals, and numvals can be used by SPI callers; the remaining fields are internal. vals is an array of pointers to rows. the number of rows is given by numvals (for somewhat historical reasons, this count is also returned in SPI_processed). tupdesc is a row descriptor which you can pass to SPI functions dealing with rows.

🎀 cleanup 🎀

SPI_finish frees all SPITupleTables allocated during the current C function. you can free a particular result table earlier, if you are done with it, by calling SPI_freetuptable. so tidy! so organized! 💅

🫶 ARGUMENTS 🫶

  • const char * command — string containing command to execute (this is the whole point bestie)
  • bool read_only — true for read-only execution
  • long count — maximum number of rows to return, or 0 for no limit

👑 RETURN VALUE 👑

if the execution of the command was successful then one of the following (nonnegative) values will be returned:

  • SPI_OK_SELECT — if a SELECT (but not SELECT INTO) was executed
  • SPI_OK_SELINTO — if a SELECT INTO was executed
  • SPI_OK_INSERT — if an INSERT was executed
  • SPI_OK_DELETE — if a DELETE was executed
  • SPI_OK_UPDATE — if an UPDATE was executed
  • SPI_OK_INSERT_RETURNING — if an INSERT RETURNING was executed
  • SPI_OK_DELETE_RETURNING — if a DELETE RETURNING was executed
  • SPI_OK_UPDATE_RETURNING — if an UPDATE RETURNING was executed
  • SPI_OK_UTILITY — if a utility command (e.g., CREATE TABLE) was executed
  • SPI_OK_REWRITTEN — if the command was rewritten into another kind of command (e.g., UPDATE became an INSERT) by a rule.

on error, one of the following negative values is returned (booo 😭):

  • SPI_ERROR_ARGUMENT — if command is NULL or count is less than 0
  • SPI_ERROR_COPY — if COPY TO stdout or COPY FROM stdin was attempted
  • SPI_ERROR_TRANSACTION — if a transaction manipulation command was attempted (BEGIN, COMMIT, ROLLBACK, SAVEPOINT, PREPARE TRANSACTION, COMMIT PREPARED, ROLLBACK PREPARED, or any variant thereof)
  • SPI_ERROR_OPUNKNOWN — if the command type is unknown (shouldn't happen 👀)
  • SPI_ERROR_UNCONNECTED — if called from an unconnected C function

💜 NOTES 💜

all SPI query-execution functions set both SPI_processed and SPI_tuptable (just the pointer, not the contents of the structure). so: save these two global variables into local C function variables if you need to access the result table of SPI_execute or another query-execution function across later calls. do not skip this step, i'm serious, the globals will move on without you 💀

that's the whole page bestie, you ate that 🫶✨

this has been SPI_EXECUTE(3) from PostgreSQL 13.23, dated 2025 — go forth and execute commands responsibly 💖🎀🦋

📦 where this page comes from