Merge pull request #5513 from sysown/v3.0-5497

Phase 3.9: PgSQL monitor health decisions + unit tests

include/PgSQLMonitorDecision.h

@@ -0,0 +1,44 @@
+/**
+ * @file PgSQLMonitorDecision.h
+ * @brief Pure decision functions for PgSQL monitor health state.
+ *
+ * PgSQL monitor is simpler than MySQL — it uses ping failure
+ * threshold directly with shun_and_killall() (always aggressive).
+ * Unshunning follows the same time-based recovery as MySQL
+ * (already covered by MonitorHealthDecision.h can_unshun_server).
+ *
+ * @see Phase 3.9 (GitHub issue #5497)
+ */
+
+#ifndef PGSQL_MONITOR_DECISION_H
+#define PGSQL_MONITOR_DECISION_H
+
+/**
+ * @brief Determine if a PgSQL server should be shunned based on ping failures.
+ *
+ * PgSQL monitor shuns servers when they miss N consecutive heartbeats.
+ * Unlike MySQL, PgSQL always uses aggressive shunning (kill all connections).
+ *
+ * @param missed_heartbeats      Number of consecutive missed pings.
+ * @param max_failures_threshold Config: pgsql-monitor_ping_max_failures.
+ * @return true if the server should be shunned.
+ */
+bool pgsql_should_shun_on_ping_failure(
+	unsigned int missed_heartbeats,
+	unsigned int max_failures_threshold
+);
+
+/**
+ * @brief Determine if a PgSQL server's read-only status indicates it should
+ *        be taken offline for a writer hostgroup.
+ *
+ * @param is_read_only  Whether the server reports read_only=true.
+ * @param is_writer_hg  Whether this is a writer hostgroup.
+ * @return true if the server should be set to OFFLINE_SOFT.
+ */
+bool pgsql_should_offline_for_readonly(
+	bool is_read_only,
+	bool is_writer_hg
+);
+
+#endif // PGSQL_MONITOR_DECISION_H

lib/Makefile

@@ -109,6 +109,7 @@ _OBJ_CXX := ProxySQL_GloVars.oo network.oo debug.oo configfile.oo Query_Cache.oo
 	ServerSelection.oo \
 	TransactionState.oo \
 	HostgroupRouting.oo \
+	PgSQLMonitorDecision.oo \
 	MySQLErrorClassifier.oo \
 	BackendSyncDecision.oo \
 	proxy_sqlite3_symbols.oo

lib/PgSQLMonitorDecision.cpp

@@ -0,0 +1,27 @@
+/**
+ * @file PgSQLMonitorDecision.cpp
+ * @brief Implementation of PgSQL monitor health decisions.
+ *
+ * @see PgSQLMonitorDecision.h
+ * @see Phase 3.9 (GitHub issue #5497)
+ */
+
+#include "PgSQLMonitorDecision.h"
+
+bool pgsql_should_shun_on_ping_failure(
+	unsigned int missed_heartbeats,
+	unsigned int max_failures_threshold)
+{
+	if (max_failures_threshold == 0) {
+		return false;  // shunning disabled
+	}
+	return (missed_heartbeats >= max_failures_threshold);
+}
+
+bool pgsql_should_offline_for_readonly(
+	bool is_read_only,
+	bool is_writer_hg)
+{
+	// A read-only server in a writer hostgroup should go OFFLINE_SOFT
+	return (is_read_only && is_writer_hg);
+}

test/tap/tests/unit/Makefile

@@ -237,6 +237,7 @@ UNIT_TESTS := smoke_test-t query_cache_unit-t query_processor_unit-t \
 	server_selection_unit-t
   hostgroup_routing_unit-t \
 	transaction_state_unit-t \
+	pgsql_monitor_unit-t \
 	mysql_error_classifier_unit-t \
 	backend_sync_unit-t
 

test/tap/tests/unit/pgsql_monitor_unit-t.cpp

@@ -0,0 +1,51 @@
+/**
+ * @file pgsql_monitor_unit-t.cpp
+ * @brief Unit tests for PgSQL monitor health decisions.
+ *
+ * @see Phase 3.9 (GitHub issue #5497)
+ */
+
+#include "tap.h"
+#include "test_globals.h"
+#include "test_init.h"
+#include "proxysql.h"
+#include "PgSQLMonitorDecision.h"
+
+static void test_ping_shunning() {
+	ok(pgsql_should_shun_on_ping_failure(3, 3) == true,
+		"shun: failures=3 meets threshold=3");
+	ok(pgsql_should_shun_on_ping_failure(5, 3) == true,
+		"shun: failures=5 exceeds threshold=3");
+	ok(pgsql_should_shun_on_ping_failure(2, 3) == false,
+		"no shun: failures=2 below threshold=3");
+	ok(pgsql_should_shun_on_ping_failure(0, 3) == false,
+		"no shun: zero failures");
+	ok(pgsql_should_shun_on_ping_failure(1, 1) == true,
+		"shun: threshold=1, single failure");
+	ok(pgsql_should_shun_on_ping_failure(10, 0) == false,
+		"no shun: threshold=0 (disabled)");
+}
+
+static void test_readonly_offline() {
+	ok(pgsql_should_offline_for_readonly(true, true) == true,
+		"offline: read_only + writer HG");
+	ok(pgsql_should_offline_for_readonly(true, false) == false,
+		"not offline: read_only + reader HG");
+	ok(pgsql_should_offline_for_readonly(false, true) == false,
+		"not offline: not read_only + writer HG");
+	ok(pgsql_should_offline_for_readonly(false, false) == false,
+		"not offline: not read_only + reader HG");
+}
+
+int main() {
+	plan(11);
+	int rc = test_init_minimal();
+	ok(rc == 0, "test_init_minimal() succeeds");
+
+	test_ping_shunning();       // 6
+	test_readonly_offline();    // 4
+	// Total: 1+6+4 = 11
+
+	test_cleanup_minimal();
+	return exit_status();
+}