Pgpool closes the frontend after multiple `ParameterStatus` messages in an extended pipeline

[apkipa]

Pgpool-II version

4.8devel (1a873669), backend_clustering_mode = raw, connection_cache = on, one configured PostgreSQL node.

Description

A query that changes two reportable session parameters, followed by another extended-protocol query in the same pipeline segment, succeeds on direct PostgreSQL but closes the Pgpool frontend.

Before waiting for the next client query, PostgreSQL reports each changed GUC_REPORT variable as a backend ParameterStatus message (guc.c, application_name, DateStyle). When handling the next Parse while the backend is idle, Pgpool sends an internal Sync, consumes at most one ParameterStatus, and then requires ReadyForQuery (pool_proto_modules.c). The second message triggers invalid read kind "S" and closes the frontend.

Reproduce

Start PostgreSQL 16 and Pgpool-II in the configuration above, install psycopg, and run:

import psycopg


def run(label, dsn):
    conn = psycopg.connect(dsn, autocommit=True)
    try:
        with conn.pipeline():
            conn.execute(
                "SELECT set_config(%s, %s, false), "
                "set_config(%s, %s, false)",
                ("application_name", "business", "DateStyle", "ISO, DMY"),
            )
            result = conn.execute("SELECT 99")
        print(f"{label}: {result.fetchone()[0]}")
    except Exception as exc:
        print(f"{label}: {type(exc).__name__}: {str(exc).splitlines()[0]}; closed={conn.closed}")
    finally:
        conn.close()


run("direct", DIRECT_DSN)
run("pgpool", PGPOOL_DSN)

Expected behavior

Both endpoints should complete the pipeline, remain usable, and print:

direct: 99
pgpool: 99

Actual behavior

The reproducer prints:

direct: 99
pgpool: OperationalError: consuming input failed: server closed the connection unexpectedly; closed=True