14 Commits (feature/pgsql-native-backend-protocol)

Author	SHA1	Message	Date
Rene Cannao	84cc0c997b	fix(test): wait for cluster admin ports before test_cluster1-t (#5782) ... Issue #5782: test_cluster1-t fails on CI-mariadb10-galera-g5 with `Can't connect to server on 'proxysql' (115)` (EINPROGRESS) deep inside create_connections() at line 153, after the first 6 local assertions pass. The cause is a startup-ordering race: not all 10 proxysql cluster instances (the primary on 6032 plus 9 satellite/ core nodes on 26001-26009) are reachable yet when the test starts. Add a reusable readiness probe `wait_for_proxysql_cluster()` to test/tap/tap/utils.{h,cpp}, plus a single-host convenience overload. The probe opens a plain (no SSL, no compression) admin connection to each endpoint with a 2s per-attempt connect timeout, round-robins across pending endpoints, retries every 2s, and gives up after a total deadline (default 60s) with a diag() listing every endpoint that did not respond plus its last mysql_error(). On success it emits a single diag() reporting how long the probe took, so future CI logs have timing visibility. Wire it into test_cluster1-t.cpp between the local admin checksum block and the create_connections() call. Both code paths (the hardcoded cluster_ports list and the cluster_nodes env override) go through the probe. The cryptic EINPROGRESS deep in create_connections() becomes either (a) a successful probe followed by a clean test, or (b) a clear, named-endpoint diagnostic at the top of the test output. Verified locally: test_cluster1-t runs to completion (208/208 ok, RC=0) under TAP group mariadb10-galera-g5 with the probe logging `all 10 endpoint(s) ready in 0.01s` (no race on a fast machine).	2 months ago
Rene Cannao	3264ebe34b	fix(test): check mysql_servers_v2 config checksum in cluster1 sync test ... The cluster sync mechanism syncs mysql_servers_v2 (config tables) between nodes, NOT mysql_servers (runtime). The runtime mysql_servers checksum is modified independently on each node by the monitor (GR discovery, read-only detection, etc.), so runtime checksums legitimately differ across nodes and will never converge. Checking the config checksum is the correct way to verify cluster sync with an active monitor. Root cause: Two groups of nodes formed with different runtime checksums (6032+6062 vs 6052+6042) but the same epoch. The cluster sync pulled config from peers and regenerated runtime, but the monitor immediately re-discovered GR servers and modified the runtime. In mysql_servers_v2 mode, the runtime is never re-synced, creating a permanent deadlock where diff_check increments forever but no pull is triggered.	2 months ago
Rene Cannao	a86a8d7026	Revert "fix(test): disable monitor during mysql_servers sync test" ... This reverts the mysql-monitor_enabled=0 change from test_cluster1-t. The monitor modifying mysql_servers across nodes is expected real-world behavior — the cluster sync must work correctly with an active monitor.	2 months ago
Rene Cannao	432365a460	fix(test): fix flaky test failures in mysql84-g8 and mysql84-gr-g5 ... - test_mysql_connect_retries-t: Reset server status to ONLINE between test phases to clear shunned state. MySQL_Thread's internal shunning logic (MySrvC::connect_error) triggers when enough connection errors accumulate within the same second, independent of the monitor. This caused the consistency check to get error 9001 (timeout) instead of 2002 (refused). - test_cluster1-t: Disable monitor during mysql_servers sync test. The monitor actively modifies mysql_servers on each node (discovering servers, updating statuses via GR topology), preventing checksum convergence across the 10-node cluster. - env.sh: Validate sourced path is under WORKSPACE (SonarCloud S3003). - pre-proxysql.bash: Use env vars for admin credentials (SonarCloud S2068).	2 months ago
Rene Cannao	d8e4a607de	fix(test/tap): real bugs from static analysis ... - fix uninitialized loop var (UB) in test_digest_umap_aux-t.cpp - fix dead stores and unused vars in test_binlog_reader-t.cpp - rename reserved identifiers (__timer, _mysql_hdr, th_args__in_t, etc.) - fix narrowing conversions (ssize_t->int, uint32_t->int, size_t->int) - fix implicit widening of multiplication (overflow risk) - consolidate duplicated conditional branch in reg_test_4556 - pass modname by const ref in test_cluster1-t.cpp - fix narrowing in test_default_conn_collation-t.cpp	3 months ago
Rene Cannao	c2d2628a58	fix(test/tap): memory leaks and NULL derefs in three TAP tests ... test_stats_proxysql_message_metrics-t.cpp: - Fix UB: mysql_error() called on NULL handle when mysql_init() fails. - Fix proxy/proxy_admin leaks on every exit path (MYSQL_QUERY macro silently returns EXIT_FAILURE from the enclosing function, bypassing any end-of-scope cleanup); use std::unique_ptr with mysql_close deleter. - Fix result-set leaks: second mysql_store_result() assignment dropped the previous MYSQL_RES*, and neither was ever freed. Switch to a unique_ptr<MYSQL_RES> and reset() at reassignment. - Guard out-of-bounds extract_mysql_rows(...)[0] with an empty-result check; previously only the first call was guarded by num_rows==1. - Fix latent error-propagation bug: if (rc == EXIT_FAILURE) only caught rc == 1; mysql_query() can return other non-zero values. Use rc != EXIT_SUCCESS instead. test_cluster1-t.cpp: - Fix three NULL-deref UB sites: mysql_error() on NULL after mysql_init(). - Fix leaks in create_connections(): on mysql_real_connect() failure the just-initialized MYSQL* was never closed. - Fix leak of the global conns[] vector: connections were never closed on exit. Added close_all_connections() called on all exit paths. - Fix leak of proxysql_admin (never closed); wrapped in unique_ptr with mysql_close deleter so destructor runs even on MYSQL_QUERY early-return. - Remove dead variable MYSQL_RES* proxy_res (declared, never used). - Fix two signed/unsigned comparisons (int loop counter vs size_t size). reg_test_4556-ssl_error_queue-t.cpp (selective fixes — file has more issues left for follow-up): - create_server_conn: add mysql_init NULL check; close handle on mysql_real_connect failure (was leaked). - create_test_database: close handle on query failure (was leaked). - create_ssl_conn_inv_cert: add mysql_init NULL check; free the tempnam path (malloc'd, never freed); close myconn on all paths. - create_ssl_conn_missing_cert: same as above, plus check fopen return, always fclose tmp_file (was only on success path), unlink temp file, and use explicit "%s" format instead of passing the data buffer as the format string. - clean_conn_pool: fix three MYSQL_RES leaks (mysql_store_result with no matching mysql_free_result). - warmup_conn_pool: on any error, the just-opened MYSQL* AND all previously-pushed conns leaked (the function returned a fresh empty vector, stranding everything already allocated). Added a cleanup_and_fail lambda that closes everything before returning. - Fix signed int loop counter vs uint32_t CONNS_TOTAL. No behavioral changes to test assertions; only leak/UB fixes. Verified with -fsyntax-only on all three files.	3 months ago
Rene Cannao	582c22333f	test: Migration to new testing infrastructure and cleanup ... - Update CommandLine to support TAP_CLUSTER_NODES, TAP_WORKDIR, and other env vars. - Update default admin credentials to 'radmin' in CommandLine to match CI standards. - Enforce connections via ProxySQL in fast-forward and binlog tests. - Replace hardcoded credentials and hosts with CommandLine variables in cluster tests. - Improve portability of tests by using dynamic host/port and TEST_DEPS. - Ensure strict equality checks for query counts in binlog reader tests.	4 months ago
Wazir Ahmed	8ebd7e48ae	TAP: Move init_mysql_conn() and run_q() to utils ... Signed-off-by: Wazir Ahmed <wazir@proxysql.com>	10 months ago
Miro Stauder	c78819bda9	use inludes from our deps intead of system	3 years ago
Javier Jaramago Fernández	a999eeddca	Fix 'test_cluster1-t' checksum fetching logic	4 years ago
Javier Jaramago Fernández	f12839acc1	Revert "Fix test 'test_cluster1' synchronization for 'mysql_servers'" ... This reverts commit 13a7d7fab0.	4 years ago
Javier Jaramago Fernández	13a7d7fab0	Fix test 'test_cluster1' synchronization for 'mysql_servers'	4 years ago
René Cannaò	bc6d4f2aba	Improve error verbosity in ProxySQL Cluster	5 years ago
René Cannaò	068ae1dc55	Adding a TAP test to verify various aspects of Cluster ... This TAP test assumes that there are a cluster of 10 proxysql nodes	5 years ago