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

Author	SHA1	Message	Date
Rene Cannao	13a43e2ffe	fix(test/tap): revert two semantic regressions from cleanup commits ... Two recent "cleanup" commits on this branch silently changed test behavior and caused the tests to fail under myrun2. 1) kill_connection2-t.cpp (regressed in 3f0cd6a0c): The "cleanup - replace NULL with nullptr, convert loops to range-based" commit flipped one assertion from ok(rc == 0, ...) to ok(rc != 0, ...) on the first DO 1 batch after a 12s sleep. With mysql-max_transaction_time=17000 (17s), the connection is still alive at 12s by design — ProxySQL only kills it at ~18s. The original test correctly expected rc == 0 there; the second DO 1 batch (after +12s = 24s total) continues to correctly expect rc != 0. 2) max_connections_ff-t.cpp (regressed in da655099e): The "static analysis narrowing conversions" commit introduced an integer overflow by wrapping a chrono::nanoseconds count in static_cast<int>(). For the 8000ms test the elapsed time is ~8e9 ns, which overflows int32 (max ~2.147e9) and wraps negative, producing "Waited: -0.589502" and a failed assertion. The 2000ms test happened to fit under 2.147e9 ns and kept passing, hiding the bug. Use static_cast<double> for the nanoseconds count, and drop the unnecessary int casts on connect_timeout / poll_timeout — those are already long, so long / 1000.0 is a valid double.	3 months ago
Rene Cannao	3f0cd6a0c6	fix(test/tap): cleanup kill_connection2-t.cpp - replace NULL with nullptr, convert loops to range-based	3 months ago
Rene Cannao	b66a14a39f	fix(test/tap): leaks and uninit read in kill_connection{2,3}-t.cpp ... Both kill_connection2-t.cpp and kill_connection3-t.cpp share the same structure and had the same bugs: - create_connections(): fix NULL-deref UB when mysql_init returns NULL (mysql_error was called on the NULL handle) and fix leaks on both mysql_init failure and mysql_real_connect failure. On any error the function now closes the in-flight handle and any previously-pushed conns[] slots before returning. - find_tids(): - Initialize 'tid' to 0: it was declared uninitialized and read on the ok() line below, which would read garbage if mysql_fetch_row returned NULL on the first iteration. - Check mysql_store_result for NULL before passing to mysql_fetch_row (was UB on NULL). - NULL-guard row[0] before atoll to avoid crashes on unexpected NULL columns. - main(): fix leaks at all error-return paths: - NULL-deref UB when mysql_init(proxysql_admin) returned NULL. - proxysql_admin leaked on mysql_real_connect failure. - conns[] (5 or 35 connections) leaked across every error-return. - Most importantly: after the 'wait_timeout' / 'kill' cycle kills the first batch of conns server-side, main() called create_ connections() a SECOND time, overwriting every conns[i] slot without closing the now-dead client-side MYSQL* structs. That dropped all NUM_CONNS MYSQL handles on the floor. Added an explicit close_all_conns() call between the two batches. - Happy-path exit now also runs close_all_conns() + mysql_close (proxysql_admin); previously everything leaked at normal exit. A new static helper close_all_conns() centralizes the cleanup; conns[] is file-scope and zero-initialized, so the NULL check in close_all_conns() is correct at entry and after overwrite. Compile-verified with -fsyntax-only.	3 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
René Cannaò	8afeb616ed	Added test for connections timeout ... Added test that verifies that client connections are dropped because of: - mysql-wait_timeout - mysql-max_transaction_time IT also helped finding a bug in how mysql-max_transaction_time was handled	5 years ago