proxysql

Commit Graph

Author	SHA1	Message	Date
Rene Cannao	6d6015a203	fix: align MCP Bearer token test with case-insensitive implementation The test expected lowercase 'bearer' to be rejected, but the implementation uses strncasecmp() per RFC 7235 (auth-scheme is case-insensitive). Update test to expect true for lowercase prefix.	2 weeks ago
Rene Cannao	acaa100dc8	fix: address PR review feedback from coding agents Actionable fixes from CodeRabbit / agent reviews on PR #5549: 1. include/Stats_Tool_Handler.h: add #include <cstdint> — uint64_t used in parse_digest_filter() without direct include 2. lib/MCP_Endpoint.cpp: use strncasecmp() for Bearer token prefix check — RFC 7235 Section 2.1 says auth-scheme is case-insensitive. This is a real bug fix in production code. 3. test/tap/tests/unit/sqlite3db_unit-t.cpp: guard result->rows_count dereference with null check to avoid segfault if execute_statement returns null 4. test/tap/tests/unit/genai_mcp_endpoint_unit-t.cpp: remove memset() calls on MCP_Threads_Handler that leaked constructor's strdup("") allocations for auth fields 5. test/tap/groups/check_groups.py: use os.walk() instead of hardcoded directory list so new test subdirectories are discovered 6. test/tap/tests/unit/admin_disk_upgrade_unit-t.cpp: add comment documenting intentional calloc pattern (ProxySQL_Admin constructor has too many daemon deps for unit tests)	2 weeks ago
Rene Cannao	4fb7aac73f	fix: restore groups.json to compact one-line-per-entry format Previous commits inadvertently reformatted groups.json with Python's json.dump(indent=2), breaking the compact format required by test/tap/groups/README.md. The correct format is one entry per line with compact arrays so grep works for quick lookups.	2 weeks ago
Rene Cannao	e806b0219c	test: register 5 new unit tests in groups.json for CI	3 weeks ago
Rene Cannao	c8408d76b3	test: add 6 unit test files with 588 assertions (Round 5) New test files (test/tap/tests/unit/): Core: - pgsql_tokenizer_unit-t (65 tests): PgSQL SQL digest — dollar- quoted strings, ::type casts, ARRAY literals, boolean literal replacement, E/B/X/U& prefix strings, comment handling, grouping/IN-list collapsing, edge cases - admin_disk_upgrade_unit-t (60 tests): all 6 disk_upgrade_*() schema migration functions with in-memory SQLite — scheduler v1.2.0→current, mysql_users v1.3.0→current, rest_api_routes v2.0.15→current, mysql_servers v1.1.0→current (with weight cap and compression normalization bugs), pgsql_replication_hostgroups v3.0.1→current, mysql_query_rules v1.2.2→current - gtid_utils_unit-t (37 tests): addGtid interval merging (adjacent, non-consecutive, duplicate, bridge, reverse-order, within- existing), addGtidInterval (update, duplicate, non-overlapping), gtid_executed_to_string formatting - glovars_unit-t (78 tests): GloVars constructor defaults, ProxySQL_Checksum_Value (set/replace_zeros/version/epoch), generate_global_checksum determinism, command-line parsing (--initial/--reload/--no-version-check/-D/--bootstrap/-n/-f/-M) - query_processor_unit-t (+217 tests, 259 total): EXTENDED with PgSQL-specific tests — all-fields rule creation, client_addr wildcard, flagOUT attributes JSON, command counters, stats, rule text 35-column serialization, query_parser_command_type for 100+ PgSQL commands (DML/DDL/views/sequences/functions/ triggers/policies/LISTEN/NOTIFY/VACUUM/TEXT SEARCH/etc.), fast routing hashmap, firewall whitelist, query digests GenAI/MCP (PROXYSQLGENAI=1): - genai_mysql_catalog_unit-t (89 tests): MySQL_Catalog with in-memory SQLite — upsert/get/remove, schema isolation, list with pagination, FTS search, merge, multiple kinds, special character handling Production changes for testability: - include/proxysql_admin.h: add friend class TestDiskUpgrade - test/tap/tests/unit/Makefile: add DEBUG auto-detection from libproxysql.a symbols (fixes struct layout ODR violations when library is built with -DDEBUG) Running totals: 32 test files, 2553 assertions, all passing.	3 weeks ago
Rene Cannao	8df688dd8a	test: register 5 new unit tests in groups.json for CI	3 weeks ago
Rene Cannao	a36834f968	test: add 6 unit test files with 676 assertions (Round 4) New test files (test/tap/tests/unit/): Core: - c_tokenizer_unit-t (59 tests): tokenizer/tokenize/free_tokenizer, c_split_2, mysql_query_digest_and_first_comment (literal replacement, comment stripping, NULL handling, grouping_limit, IN-clause), mysql_query_digest_first_stage, strip_comments - config_write_unit-t (111 tests): all 7 Write_*_to_configfile() functions with in-memory SQLite — MySQL users/servers/query rules, scheduler, restapi, global variables, proxysql_servers; tests field formatting, NULL handling, double-quote escaping, prefix grouping - listen_validator_unit-t (112 tests): trim_copy, split_listener_list, is_ipv4/ipv6_wildcard, parse_listener (IPv4/IPv6/Unix socket/ invalid/edge cases), sets_intersect, listeners_conflict, validate_module_listener_conflicts - ezoption_parser_unit-t (110 tests): standalone ezOptionParser — flags, string/int/float options, defaults, required/unknown, aliases, multi-value, usage layout, export/import, validators, vector getters - pgsql_txn_state_unit-t (+58 tests, 112 total): EXTENDED with PgSQL_ExplicitTxnStateMgr state machine tests — BEGIN/COMMIT/ ROLLBACK lifecycle, ROLLBACK AND CHAIN, savepoint add/release/ rollback_to, reset_state, fill_internal_session JSON, case- insensitive savepoints, END/ABORT synonyms. Uses minimal PgSQL_Session mock with variable snapshots. GenAI/MCP (PROXYSQLGENAI=1): - genai_discovery_schema_unit-t (172 tests): Discovery_Schema with in-memory SQLite — run/agent_run management, schema/object/column/ index/FK insertion, profile/LLM summary/domain/metrics/notes, list_objects pagination+ordering, get_object with columns/indexes, get_relationships FK mappings, FTS rebuild+search, fingerprint_mcp_args, compute_mcp_digest determinism, MCP_Query_Digest_Stats timing, MCP_Query_Processor_Output, MCP_Query_Rule defaults Running totals: 27 test files, 1944+ assertions, all passing.	3 weeks ago
Rene Cannao	a8c5432fc5	fix: remove stale proxysql.db before launching cluster sync replicas test_cluster_sync_mysql_servers-t spawns replica ProxySQL instances with a specific config file setting admin port to 96061/96062. If a proxysql.db from a previous run exists in the workdir, ProxySQL ignores the config file and loads from the DB instead, using the old admin port. The test then fails to connect on the expected port. Fix: remove proxysql.db and proxysql_stats.db before launching the replica, not just after. The post-launch cleanup is kept as well for normal operation.	3 weeks ago
Rene Cannao	0f76a18329	fix: resolve CI failures in charset_find and statistics unit tests charset_find_unit-t: - Fix plan count mismatch: plan(63) but only 62 ok() calls. Changed to plan(62). statistics_unit-t: - Root cause: TSDB auto-detect in Makefile searched for symbol 'ProxySQL_TSDB' which doesn't exist. Changed to 'init_tsdb_variables'. Without TSDB detection, struct layout mismatch caused statsdb_disk to appear as NULL (member offset shifted by #ifdef PROXYSQLTSDB). - Added statsdb_ok guard + skip() for test_init_creates_tables to prevent segfault if construction fails (graceful degradation). - Use TMPDIR/tmp directly instead of GloVars.datadir for statsdb path.	3 weeks ago
Rene Cannao	99dcb5b160	fix: namespace stats_utils helpers to avoid symbol collision ProxySQL_Admin_Tests2.cpp defines its own parse_int_or_zero() in an anonymous namespace. Our declaration in Stats_Tool_Handler.h (included transitively via cpp.h) created an ambiguous overload error. Fix: wrap the 4 helper functions in namespace stats_utils in both the header and implementation. A using-directive within the .cpp keeps existing call sites working without qualification.	3 weeks ago
Rene Cannao	093693ba06	test: add check_groups.py to verify groups.json completeness Python tool that checks consistency between groups.json and compiled TAP test binaries: - WARNING (exit 0): tests in groups.json that aren't compiled yet - ERROR (exit 1): compiled tests missing from groups.json Scans test/tap/tests/, test/tap/tests/unit/, and test/tap/tests_with_deps/ for ELF binaries matching *-t pattern. Usage: python3 test/tap/groups/check_groups.py	3 weeks ago
Rene Cannao	fa668d646a	test: register 21 new unit tests in groups.json for CI Add all new unit test binaries to the CI test group configuration: - 13 core tests → unit-tests-g1 - 8 GenAI tests → ai-g1 This ensures CI pipelines discover and run all new tests.	3 weeks ago
Rene Cannao	ece648fc16	test: add 6 more unit test files with 550 assertions (Tier 2+3) Expand coverage for firewall whitelist, statistics, config validation, logging, and GenAI thread variable management. New test files (test/tap/tests/unit/): Core: - query_processor_firewall_unit-t (44 tests): load_firewall() users/rules/sqli fingerprints, find_firewall_whitelist_user/rule, whitelisted_sqli_fingerprint, reload/clear, memory accounting, PgSQL parity, multiple digest bsearch - statistics_unit-t (20 tests, 95 with TSDB): timer functions (timetoget for MySQL threads, query cache, CPU, logger, digest), TSDB set/get/has variable with range validation, insert/query metrics, backend health, downsampling, retention - config_validation_unit-t (54 tests): addField() formatting and escaping, validate_backend_users() duplicate detection for MySQL/PgSQL, validate_backend_servers() with field aliases and default ports, validate_proxysql_servers() - log_utils_unit-t (42 tests): LogBuffer append/write/operator<< overloads, data() accessor, reset(), should_log() rate limiting, GetLogBufferThreadContext() lifecycle GenAI/MCP (PROXYSQLGENAI=1): - genai_mcp_thread_unit-t (197 tests): MCP_Threads_Handler constructor defaults (26 vars), boolean/port/timeout/string variable set/get with range validation, has_variable, get_variables_list, null safety, target auth map - genai_thread_unit-t (193 tests): GenAI_Threads_Handler constructor defaults (31 vars), integer range validation for 15+ variables, double/boolean/string round-trips, has_variable, get_variables_list, overwrite and rejection persistence Note: found potential bug — llm_cache_enabled has no get/set handler despite being listed in variables array Minimal production change: - include/proxysql_config.h: add friend class for test access to private validate_backend_users/servers/proxysql_servers methods Running totals: 21 test files, 1268 assertions, all passing.	3 weeks ago
Rene Cannao	1d2f16b506	test: add 4 more unit test files with 198 assertions (Tier 1) Expand coverage for charset lookups, GenAI Query/LLM/MCP components. New test files (test/tap/tests/unit/): Core: - charset_find_unit-t (62 tests): all 4 proxysql_find_charset functions — edge cases for NULL returns, utf8mb3→utf8 translation, case-insensitive matching, collation preference GenAI/MCP (PROXYSQLGENAI=1): - genai_query_handler_unit-t (65 tests): validate_sql_identifier, escape_string_literal, strip_leading_comments, json_string/ json_int/json_double safe extraction, is_pgsql_protocol, is_runtime_online_status, uppercase_or_unknown - genai_llm_clients_unit-t (35 tests): is_retryable_error HTTP status codes (408/429/5xx) and CURL error codes, plus LLM_WriteCallback data accumulation - genai_mcp_endpoint_unit-t (36 tests): validate_bearer_token extraction/comparison, create_jsonrpc_response/error JSON-RPC 2.0 formatting Minimal production changes for testability: - include/Query_Tool_Handler.h: friend class + extern declarations for 8 pure helper functions - lib/Query_Tool_Handler.cpp: remove static from 8 file-local pure functions (no behavioral change) - include/MCP_Endpoint.h: friend class + new static validate_bearer_token() extracted from authenticate_request() - lib/MCP_Endpoint.cpp: extract token validation into static method - lib/LLM_Clients.cpp: remove static from is_retryable_error() and rename WriteCallback to LLM_WriteCallback (avoid collisions) Running totals: 15 test files, 718 assertions, all passing.	3 weeks ago
Rene Cannao	accc68c9dc	test: add 11 new unit test files with 520 assertions for lib/ coverage Add targeted unit tests for pure functions and isolated components across lib/, increasing test coverage for both core utilities and GenAI/MCP components. New test files (test/tap/tests/unit/): Core utilities: - pgsql_variables_validator_unit-t (75 tests): bool, float, string, memory size, search_path, and client_encoding validators - gen_utils_unit-t (28 tests): mywildcmp wildcard matching, escape_string_single_quotes, Proxy_file_regular - mysql_encode_unit-t (23 tests): mysql_encode_length, CPY3/CPY8, proxy_my_crypt XOR, write_encoded_length - proxysql_utils_unit-t (23 tests): hex/unhex, sql_escape, split_str, replace_str, generate_multi_rows_query, checksums - proxy_protocol_unit-t (22 tests): PROXY v1 header parsing, subnet validation, IPv4/IPv6 network matching - pgsql_error_helper_unit-t (139 tests): SQLSTATE code mapping, all 38 error class branches, severity classification, fill/reset - pgsql_txn_state_unit-t (54 tests): TxnCmdParser for BEGIN, COMMIT, ROLLBACK, SAVEPOINT, RELEASE, AND CHAIN variants - sqlite3db_unit-t (21 tests): execute, return_one_int, check_table_structure, build_table, row ops, locking GenAI/MCP (PROXYSQLGENAI=1): - genai_stats_parsing_unit-t (79 tests): parse_server_filter, parse_digest_filter, parse_ll_or_zero, get_interval_config, calculate_percentile - genai_fts_string_unit-t (32 tests): MySQL_FTS sanitize_name, escape_identifier, escape_sql, table name construction - genai_anomaly_unit-t (24 tests): normalize_query SQL normalization, check_sql_injection pattern detection Minimal production changes for GenAI testability: - include/Anomaly_Detector.h: add friend class for test access - include/Stats_Tool_Handler.h: declare 4 parsing helpers, move 2 pure computation methods (get_interval_config, calculate_percentile) to public - lib/Stats_Tool_Handler.cpp: remove static from 4 free helper functions (no behavioral change, only linkage visibility)	3 weeks ago
Rene Cannao	1ebe4de9b7	Fix mysql84 pre-proxysql hook, basictests setup, and move flaky tests - mysql84 pre-proxysql.bash: use docker exec for all ProxySQL admin commands (was using mysql -h127.0.0.1 which can't reach the container) - basictests setup-infras.bash: use INSERT OR REPLACE when remapping hostgroups 1300/1301 -> 0/1 to avoid UNIQUE constraint violation when target hostgroups already have entries - proxysql-tester.py: skip per-stage coverage generation when MULTI_GROUP=1 (was spawning 32 gcov processes on 35k .gcno files) - Move test_cluster_sync-t and test_read_only_actions_offline_hard_servers-t to legacy-g5 pending investigation of cluster sync algorithm issues. Move reg_test_3765_ssl_pollout-t from legacy-g5 to legacy-g4.	3 weeks ago
Rene Cannao	f8c64142cb	Fix cluster CI: remove duplicate primary INSERT and add sync_algorithm=3 - Remove redundant INSERT of primary into proxysql_servers on nodes. Commit `b02147b92` added the primary to the shared PROXYSQL_SERVERS_SQL but didn't remove the pre-existing per-node INSERT, causing a UNIQUE constraint violation that broke all multi-node CI groups. - Add cluster_mysql_servers_sync_algorithm=3 to test_cluster_sync config files and update hardcoded datadir paths to match current workspace.	3 weeks ago
Rene Cannao	72a1c5dd4a	Set cluster_mysql_servers_sync_algorithm=3 in test_cluster_sync replica config The replica used the default algorithm (1) which syncs runtime mysql_servers — including monitor-shunned status. The test checks the config table (mysql_servers) for status='ONLINE', but the runtime sync overwrites the config table with SHUNNED status. With auto_select (3) and monitor enabled (-M flag), the replica uses mysql_servers_v2 sync which preserves the configured status.	3 weeks ago
Rene Cannao	0fa1b53709	Fix test_cluster_sync-t: connect to spawned replica via 127.0.0.1 The test spawns its own ProxySQL replica (port 16062) inside the test-runner container, but connected to it via cl.host (TAP_ADMINHOST) which resolves to the main ProxySQL container in the Docker setup. Since port 16062 doesn't exist on the main container, the connection hung indefinitely, causing the test to stall. Now uses R_HOST="127.0.0.1" for all replica connections (R_PORT), since the replica runs locally inside the same container as the test.	3 weeks ago
Rene Cannao	5f7fb254a9	Fix pgsql transaction test and auth_methods crash-on-skip - pgsql-transaction_variable_state_tracking-t: replace superuser-only variable allow_in_place_tablespaces with user-settable enable_mergejoin. Fixes #5536. - test_auth_methods-t: call stop_internal_noise_threads() before early return on MySQL 5.7, preventing 'terminate called without an active exception' crash from joinable std::thread destructors. Fixes #5537.	3 weeks ago
Rene Cannao	7739b86745	Move test_auth_methods-t from legacy-g2 to mysql84-g2 only Test requires MySQL 8.0+ (detects 5.7 and skips, but crashes with 'terminate called without an active exception' during noise thread cleanup). The crash-on-skip is a separate test bug.	3 weeks ago
Rene Cannao	5c9121ef40	Add hostgroup debug diagnostics and cache test tolerance - Add dump_hostgroup_debug() utility: dumps runtime_mysql_servers and stats_mysql_connection_pool for a given hostgroup via diag() output. Helps diagnose "hostgroup has no servers available" CI failures. - test_throttle_max_bytes_per_second_to_client-t: dump hostgroup 1300, 0, and 1 when create_table_test_sbtest1() fails. - test_warnings-t: dump target hostgroup state before running tests. - test_unsupported_queries-t: dump target hostgroup state before running tests. - test_query_cache_soft_ttl_pct-t: add tolerance of +1 to all three assertions. The soft TTL refresh race allows two threads to both see "needs refresh" simultaneously under load, producing one extra backend hit. This is a known property of the soft TTL mechanism.	3 weeks ago
Rene Cannao	42f4507e72	Fix test_tls_stats plan count and move caching_sha2 to mysql84 - test_tls_stats-t: plan(29) -> plan(27). The test executes 27 ok() calls on the happy path; plan was miscounted during authoring. - reg_test_4935-caching_sha2-t: remove from legacy-g4 (MySQL 5.7 doesn't support caching_sha2_password). Keep in mysql84-g4.	3 weeks ago
René Cannaò	d1c86b141f	Merge pull request #5525 from sysown/v3.0-ci260322 Migrate CI-basictests from jenkins-build-scripts to in-repo test/infra	3 weeks ago
Rene Cannao	ba309e9c89	Fix groups.json: restore compact one-line-per-entry format	3 weeks ago
Rene Cannao	e299de51cf	Add step-by-step diagnostics to reg_test__ssl_client_busy_wait-t Test now prints a description of what it does and numbered steps: 1. Connect to admin 2. Disable cluster scheduler 3. Measure idle CPU baseline 4. Trigger BUSY_LOOP (disconnect during query) 5. Measure CPU after BUSY_LOOP 6. Wait for cleanup 7. Verify CPU returned to idle 8. Trigger INF_LOOP (disconnect after response) 9. Measure CPU after INF_LOOP 10. Recover cluster scheduler	3 weeks ago
Rene Cannao	cc9a0414d3	Add connection details to noise thread failure logs All noise thread failure messages now include host, port, user, and the actual error message. Previously they only logged generic "connection failure" with no details, making it impossible to diagnose why a noise thread failed.	3 weeks ago
Rene Cannao	801a44399a	Split clickhouse tests into separate legacy-clickhouse group Move 3 clickhouse TAP tests (clickhouse_php_conn-t, test_clickhouse_server-t, test_clickhouse_server_libmysql-t) from legacy-g1/g3 to legacy-clickhouse-g1. Remove infra-clickhouse23 from legacy/infras.lst. This eliminates the clickhouse container from all legacy-g1..g4 runs, reducing container count per group.	3 weeks ago
Rene Cannao	6769ea4845	Make legacy/pre-proxysql.bash a no-op	3 weeks ago
Rene Cannao	6fcb8fd631	Run ProxySQL cluster nodes inside a single container Instead of spawning 1+9 Docker containers for the ProxySQL cluster, run all instances as background processes inside the primary container. Each node gets its own port pair (admin=6032+i10, mysql=6033+i10) and data directory. This eliminates 9 Docker containers per group, avoiding Docker networking/iptables contention that caused "no route to host" errors under parallel test execution. Changes: - start-proxysql-isolated.bash: starts cluster nodes as background processes inside the container, then initializes the cluster and installs the scheduler — all in one script - check_all_nodes.bash: polls 127.0.0.1 on different ports instead of container hostnames - env-isolated.bash: TAP_CLUSTER_NODES uses proxysql:PORT format - default/pre-proxysql.bash, legacy/pre-proxysql.bash: no longer call cluster_start.bash/cluster_init.bash (handled by start-proxysql) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	09e9207da9	Move reg_test_3765_ssl_pollout-t from legacy-g2 to legacy-g5 This test does 122-second CPU measurement waits, making it too slow for the legacy-g2 group which needs to complete in a reasonable time.	3 weeks ago
Rene Cannao	f2db55f0c8	Fix basictests hostgroup hook: move instead of copy to avoid monitor conflicts Previously the hook copied servers from 1300/1301 into 0/1, leaving both pairs active. The replication hostgroup monitor managed both pairs simultaneously, causing race conditions that removed servers from 0/1. Now moves servers in-place (UPDATE) and deletes the original 1300/1301 replication entry, so only one pair (0/1) exists for the monitor. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	16e94131fa	Pass MAKECMDGOALS to unit test submake Without this, `make build_tap_test_debug` calls `make` (default target) in test/tap/tests/unit/ instead of `make debug`. The debug target adds -DDEBUG which is needed to define __thread stubs like mysql_thread___session_debug that are guarded by #ifdef DEBUG. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	577341103e	Fix unit test Makefile: auto-detect feature flags from libproxysql.a Unit tests failed to link because feature flags (PROXYSQLCLICKHOUSE, PROXYSQLFFTO, etc.) were only set from environment variables that may not be present during standalone `make` in test/tap/tests/unit/. Now auto-detects which features libproxysql.a was built with by checking for characteristic symbols (ClickHouse_Server, MySQLFFTO, GenAI_Thread, ProxySQL_TSDB). Also fixes ClickHouse→LZ4 link order. All 18 unit test binaries now compile and link successfully. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	573a3fd256	Address AI code review feedback on PgSQL FFTO tests (#5517 ) Fixes from Copilot and CodeRabbit reviews on PR #5526: 1. Buffer overflow risk in mysql_real_escape_string: All 7 PgSQL test files used fixed 256-byte buffers for escaped credentials. Changed to VLAs sized from strlen() (2*len+1), matching the safe pattern used in the existing test_ffto_pgsql-t.cpp. 2. NULL dereference in polling loops: mysql_store_result() can return NULL, and calling mysql_fetch_row(NULL) is undefined behavior. Added NULL checks in all verify_pg_digest(), poll_pg_digest_count(), verify_digest(), and poll_digest_count() polling loops across both PgSQL and MySQL test files. In loop contexts, uses continue; in non-loop contexts, uses conditional fetch (res ? fetch(res) : NULL). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	fd7ac876e2	Fix build issues in PgSQL FFTO E2E tests (#5517 ) Two build issues fixed across all 7 PgSQL FFTO test files: 1. PROTOCOL_VERSION macro conflict: pg_lite_client.h defines PROTOCOL_VERSION as a static const int, but mariadb_version.h (included via mysql.h) defines it as a preprocessor macro (#define PROTOCOL_VERSION 10). Fixed by including pg_lite_client.h before mysql.h in the two tests that use it. 2. MYSQL_QUERY in void function: The clear_pg_stats() helper is a void function, but MYSQL_QUERY macro expands to "return EXIT_FAILURE" on error, which doesn't compile in a void context. Fixed by using mysql_query() directly instead of the MYSQL_QUERY macro in all clear_pg_stats() definitions. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	d1d10188cc	Register PgSQL FFTO tests in Makefile and groups.json (#5517 ) Add Makefile rules for the two pg_lite_client-based tests: - test_ffto_pgsql_pipeline-t (links pg_lite_client.cpp) - test_ffto_pgsql_stmt_portal-t (links pg_lite_client.cpp) The other 5 PgSQL FFTO tests use the default wildcard rule (%-t: %-t.cpp) and don't need explicit Makefile entries. Register all 7 new PgSQL FFTO tests in groups.json under legacy-g4 and mysql84-g4 (same groups as test_ffto_pgsql-t): - test_ffto_pgsql_command_types-t - test_ffto_pgsql_concurrent-t - test_ffto_pgsql_errors-t - test_ffto_pgsql_large_resultsets-t - test_ffto_pgsql_mixed_protocol-t - test_ffto_pgsql_pipeline-t - test_ffto_pgsql_stmt_portal-t Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	c19d049cc9	FFTO E2E TAP test: PostgreSQL statement/portal management (#5517 ) Add test_ffto_pgsql_stmt_portal-t.cpp covering FFTO's statement and portal tracking via m_statements and m_portals maps. Uses pg_lite_client for raw protocol access. Part of issue #5517. Test scenarios (10 TAP assertions): 1. Named statement "my_sel" executed 20 times — verifies count_star=20 and that all executions share one digest via the statement name lookup chain (portal -> statement -> SQL text). 2. Close statement then re-prepare with same name but different SQL — verifies FFTO erases the old m_statements entry on Close ('C') and stores new SQL on the subsequent Parse ('P'). 3. Unnamed statement (empty string name) — verifies FFTO correctly handles the PostgreSQL unnamed statement/portal convention. Requires Makefile rule to link pg_lite_client.cpp. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	9f29dc9ccf	FFTO E2E TAP test: PostgreSQL extended query pipelining (#5517 ) Add test_ffto_pgsql_pipeline-t.cpp covering FFTO behavior with pipelined extended query protocol. Uses pg_lite_client for raw protocol access since libpq doesn't expose true pipelining. Part of issue #5517. Test scenarios (13 TAP assertions): 1. 3 different queries (SELECT, INSERT, UPDATE) pipelined before a single Sync — verifies FFTO's m_pending_queries deque correctly queues and finalizes each query on its CommandComplete response. 2. Same prepared statement executed 10 times in pipeline (all Bind+Execute before Sync) — verifies count_star=10 aggregation. Requires Makefile rule to link pg_lite_client.cpp. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	8c9d54aa15	FFTO E2E TAP test: concurrent PostgreSQL sessions (#5517 ) Add test_ffto_pgsql_concurrent-t.cpp covering FFTO behavior under concurrent PgSQL load with 10 threads. Part of issue #5517. Test scenarios (13 TAP assertions): 1. Table setup with 100 seed rows via generate_series. 2. 10 pthreads, each with own PGconn (fast_forward=1). Each thread executes 50 SELECTs + 50 INSERTs with per-thread ID ranges. 3. Aggregate stats: SUM(count_star) >= 1000, SUM(rows_affected) >= 500. Thread safety: std::mutex serializes TAP ok() calls. Each thread gets its own PGconn -> PgSQL_Session -> PgSQLFFTO instance. Connection retries (3 attempts) handle transient load. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	db14e0bba7	FFTO E2E TAP test: PostgreSQL mixed simple + extended protocol (#5517 ) Add test_ffto_pgsql_mixed_protocol-t.cpp covering FFTO behavior when alternating between simple query ('Q') and extended query protocol (Parse+Bind+Execute) within the same PgSQL session. Part of #5517. Test scenarios (16 TAP assertions): 1. Simple SELECT then extended SELECT — simple uses ? normalization, extended uses $1 parameters. Both produce separate digest entries. 2. Extended INSERT then simple UPDATE — validates clean state transition from extended back to simple protocol. 3. Extended SELECT executed 20 times — validates count_star=20 and sum_rows_sent=20 across repeated prepared statement execution. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	fc8fbc463a	FFTO E2E TAP test: PostgreSQL error handling (#5517 ) Add test_ffto_pgsql_errors-t.cpp covering FFTO recovery after PostgreSQL ErrorResponse messages. Part of issue #5517. Test scenarios (10 TAP assertions): 1. Syntax error recovery — sends invalid SQL, then valid SELECT. Verifies FFTO resets state after ErrorResponse ('E') and correctly tracks the subsequent query. 2. Constraint violation — duplicate key INSERT fails, followed by successful INSERT. Verifies rows_affected only counts the successful operation. 3. Transaction error — BEGIN, failing INSERT, ROLLBACK. Verifies ROLLBACK is tracked as a successful query despite the preceding error putting the transaction in aborted state. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	d54ef905ab	FFTO E2E TAP test: PostgreSQL CommandComplete tag varieties (#5517 ) Add test_ffto_pgsql_command_types-t.cpp covering FFTO behavior with different PostgreSQL CommandComplete tag formats. Part of the comprehensive PgSQL E2E test suite for issue #5517. Test scenarios (17 TAP assertions): 1. Standard DML types (INSERT, SELECT, UPDATE, DELETE) — verifies correct row count extraction from tags like "INSERT 0 3", "SELECT 2", "UPDATE 1", "DELETE 1". 2. DDL commands (CREATE TABLE, ALTER TABLE, DROP TABLE) — verifies digest recording for commands with no row count in tag. 3. 100 rapid-fire SELECTs — validates count_star accuracy under rapid execution. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	f02f671076	FFTO E2E TAP test: PostgreSQL large result sets (#5517 ) Add test_ffto_pgsql_large_resultsets-t.cpp covering FFTO behavior with large PostgreSQL result sets. Part of the comprehensive PgSQL E2E test suite for issue #5517. Test scenarios (13 TAP assertions): 1. SELECT returning 10,000 rows via generate_series — verifies sum_rows_sent accuracy through PgSQLFFTO's CommandComplete tag parsing ("SELECT 10000"). 2. SELECT with ~1 KB TEXT columns (50 rows) — validates row counting with large per-row payloads under FFTO buffer management. 3. Result set under ffto_max_buffer_size (64 KB) — confirms FFTO stays active and records digests when within threshold. 4. Bulk INSERT of 500 rows via generate_series — validates sum_rows_affected tracking from "INSERT 0 500" tag. Uses the PgSQL FFTO test pattern: MySQL admin connection for config and stats queries, libpq PGconn for data operations, stats cleared via stats_pgsql_query_digest_reset, verify_pg_digest() with polling. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	eb487560ab	Fix mysql84+ pre-proxysql.bash: remove infra-default, fix infra paths - Remove references to non-existent infra-default/docker-compose-destroy.bash - Fix paths from test/infra/control/${INFRA}/ to test/infra/${INFRA}/ (infra directories live under test/infra/, not test/infra/control/) Affects: mysql84, mysql84-gr, mysql90-mysql93, mysql90-gr-mysql93-gr Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	87d9e70387	Address AI code review feedback on FFTO E2E tests (#5516 ) Fixes from Gemini, Copilot, and CodeRabbit reviews on PR #5524: 1. bypass_recovery: Updated scenario 4 description from "boundary test (pkt_len == buffer)" to "query under buffer limit" to match the actual test behavior. The exact boundary test was intentionally replaced with an under-limit sanity check due to session-level buffer management subtleties. 2. large_queries: Removed redundant stats_mysql_query_digest_reset call at the start of scenario 1 (the one right before the INSERT is the one that matters). Fixed indentation of the inner reset block. Updated scenario 5 description from "10 distinct queries" to "10 queries with different literals — normalization" since they all normalize to the same digest. 3. large_resultsets: Updated file header to remove references to the removed over-threshold bypass scenario. 4. mixed_protocol: Updated scenario 1 description from "separate digests" to "normalized to same digest" since text and prepared SELECT both normalize identically. Added result check for exec_prepared_select() in scenario 2 to catch silent failures. 5. concurrent: Added mysql_field_count() guard before incrementing total_ok in the SELECT loop — prevents counting success when mysql_store_result() returns NULL for a result-set query. 6. transactions: Fixed SAVEPOINT/RELEASE SAVEPOINT disambiguation. The LIKE '%SAVEPOINT sp1%' pattern matched both digests. Now uses NOT LIKE '%RELEASE%' to isolate the plain SAVEPOINT digest. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	29fec9d9ff	Add hostgroup aliasing hook for basictests group The in-repo infra uses hostgroup pairs like 1300/1301, but the legacy proxysql-tester.py benchmark validates against hostgroups 0 and 1. This setup hook copies the first hostgroup pair into 0/1 and remaps users and query rules accordingly. Benchmark and change_user tests now pass locally. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	b241745145	Fix FFTO E2E tests to pass in TAP infrastructure (#5516 ) Fixes discovered by running tests in the Docker-based TAP test infrastructure (ensure-infras.bash + run-tests-isolated.bash): 1. Digest normalization mismatch: The ProxySQL query digest normalizer strips spaces after commas in column lists, producing "SELECT id,val" not "SELECT id, val". Fixed all verify_digest() LIKE patterns to match the normalized form (no space after commas). 2. Stats table clearing: Replaced DELETE FROM stats_mysql_query_digest and TRUNCATE TABLE with SELECT * FROM stats_mysql_query_digest_reset, which atomically reads and clears the stats. The DELETE/TRUNCATE approach left stale digests that accumulated across test scenarios within the same test file, causing count_star and rows_affected mismatches. 3. Mixed protocol test (scenario 1): Removed verify of the seed INSERT that was executed before stats were cleared — it could never be found in the digest table. Separated text INSERT and prepared INSERT verification in scenario 2 since they produce different digest texts (VALUES without column list vs with column list). 4. Large resultsets test: Removed the over-threshold scenario (result set exceeding ffto_max_buffer_size) because the server-side bypass behavior is non-deterministic — FFTO may still record the digest with complete rows_sent depending on buffer flushing timing. Bypass behavior is already thoroughly tested in bypass_recovery and test_ffto_bypass-t. 5. Bypass recovery test: Changed boundary test from exact pkt_len == buffer_size (which was unreliable due to session-level buffer management) to a simple under-limit test that verifies FFTO records digests for queries well under the buffer threshold. All 6 tests now pass in the TAP infrastructure: bypass_recovery: PASS (9/9 assertions) concurrent: PASS (13/13 assertions) large_queries: PASS (13/13 assertions) large_resultsets: PASS (13/13 assertions) mixed_protocol: PASS (22/22 assertions) transactions: PASS (34/34 assertions) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	0c58d1c645	Fix REPO_ROOT derivation: 4 levels up from test/tap/groups/<name>/ Scripts are at test/tap/groups/<name>/pre-proxysql.bash, need 4 parent traversals to reach repo root, not 3. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago
Rene Cannao	75d958f29d	Fix hardcoded /home/rene/proxysql paths in pre-proxysql.bash scripts Replace hardcoded absolute paths with dynamically derived REPO_ROOT so these scripts work on any machine (CI runners, other developers). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	3 weeks ago

1 2 3 4 5 ...

1884 Commits (6d6015a2034dbee2ea4eb50d2e84381df008fa46)