Skip to content

34000 — invalid_cursor_name

PostgreSQL SQLSTATE 34000: cursor lifecycle, session affinity, and recovery.

34000 — invalid_cursor_name

At a glance

34000 means that a cursor or portal name cannot be resolved in the current backend session. A cursor that was closed, declared on another pooled connection, or ended with its transaction is not the same resource as a cursor with a bad position (24000).

Field Value
SQLSTATE 34000
Condition invalid_cursor_name
Status active
Known present by 7.4
Locked snapshots 9.0.23, 9.1.24, 9.2.24, 9.3.25, 9.4.26, 9.5.25, 9.6.24, 10.23, 11.22, 12.22, 13.23, 14.24, 15.19, 16.15, 17.11, 18.6, 19beta3
Macros ERRCODE_INVALID_CURSOR_NAME, ERRCODE_UNDEFINED_CURSOR
Aliases ERRCODE_UNDEFINED_CURSOR

Meaning

The executor’s portal path reports cursor "%s" does not exist when FETCH cannot find the named portal. Cursor names are tied to a backend session, and ordinary cursors are tied to a transaction unless declared with an appropriate hold option. A pool checkout therefore must keep declaration and use on the same connection and within the intended transaction boundary.

Diagnosis

Record SQLSTATE, the dynamic cursor name, backend PID, and transaction status. Check the application path for an earlier CLOSE, an implicit commit, a connection return to the pool, or a different session. pg_cursors and the server-side session identity can help when the diagnosis runs on the same backend; a new connection cannot inspect or fetch a cursor owned by the old one.

Response

For an explicit block, rollback after the failed FETCH before issuing more commands. Re-declare the cursor on the same session, or deliberately redesign the workflow to materialize keys/results when work must cross pool checkouts. Use a holdable cursor only when its commit-survival semantics are intended; it does not make a cursor portable between sessions.

Messages

The selected core path uses cursor "%s" does not exist; protocol and PL/pgSQL paths can use portal "%s" or the same cursor wording. The name is dynamic, so use the SQLSTATE and diagnostic fields rather than a full-message equality check across versions and callers.

Representative case

The shared registry in verify/cases/34000/snippets.json (SHA-256 ad5b519e8ca30fd2636b1d0bace32dc18558cccc7b9d360081c27b12a1405f08) declares and closes a cursor, proves the missing-FETCH error, rolls back, and recreates it in the same session. See the public case export and structured evidence.

BEGIN;
DECLARE cursor_name CURSOR FOR SELECT 1;
CLOSE cursor_name;
FETCH cursor_name;
ROLLBACK;
BEGIN;
DECLARE cursor_name CURSOR FOR SELECT 1;
FETCH cursor_name;
CLOSE cursor_name;
COMMIT;

The runner substitutes a unique cursor name. The first FETCH is intentionally after CLOSE; the second FETCH occurs after a new BEGIN and declaration, so the successful row proves lifecycle repair rather than a client-side mock.

The selected case observed 34000 on PostgreSQL 18.6 and 10.21 after closing a named cursor and attempting FETCH. The failed block reached INERROR; rollback followed by a new declaration, fetch, close, and commit returned the session to IDLE.

Versions

The locked catalogue records this condition across the formal snapshots. 18.6 and 10.23 source scans both contain the portal and PL/pgSQL paths; the selected runtime case observes the same SQLSTATE on 18.6 and 10.21, while exact source line and caller wording vary.

Compare 24000 invalid cursor state for a cursor that exists but is not positioned, and 26000 invalid prepared statement name for another session-local resource.

Sources

  • src.errcodes.18.6 (SHA-256 6e8de346643ba84aa3c9c6a73360acfc7b2dfb89162c06c08ce9bf5bcd5bbcba)
  • src.portalcmds.18.6 (SHA-256 e71c5bdb2da67771fb5f42f18b823e8af3ee6c31d5314e97a47ad18f7788bf35)
  • DECLARE documentation · local call scan src.calls.REL_18_6 (SHA-256 9ee8a0e81d8f0825c5c1ae45583439859a26e602bdd4ce2f2a62aa278867ccbf)