proxysql

Commit Graph

Author	SHA1	Message	Date
Wazir Ahmed	f815b77812	fix: make server_backoff_time atomic Signed-off-by: Wazir Ahmed <wazir@proxysql.com>	1 month ago
Wazir Ahmed	a76be53785	fix: skip session variable handling when tracking is disabled Signed-off-by: Wazir Ahmed <wazir@proxysql.com>	1 month ago
Wazir Ahmed	8d94919b22	fix: preserve EOF capability in enforced mode - Keep `CLIENT_DEPRECATE_EOF` enabled during frontend handshake when `session_track_variables=ENFORCED`. - Avoid unsafe mysql_error() calls in session-track TAP tests. Signed-off-by: Wazir Ahmed <wazir@proxysql.com>	1 month ago
Wazir Ahmed	99494b015d	chore(tap): Fix lint errors in groups.json Signed-off-by: Wazir Ahmed <wazir@proxysql.com>	1 month ago
Wazir Ahmed	baa5069eaf	CI: Add `mysql56-single` test infra Signed-off-by: Wazir Ahmed <wazir@proxysql.com>	1 month ago
Wazir Ahmed	885712c679	Merge branch 'v3.0' into session-track-system-variable	1 month ago
René Cannaò	872479aeba	Merge pull request #5650 from sysown/feat/cla-assistant-setup Add Contributor License Agreement (ICLA + CCLA)	1 month ago
Rene Cannao	615a620ee6	Add Contributor License Agreement (ICLA + CCLA) for CLA assistant Introduces .github/CLA.md with both Individual and Corporate variants, adapted from the Apache ICLA v2.2 and Software Grant / Corporate CLA (r190612) with the following changes: - "The Apache Software Foundation" replaced with "ProxySQL LLC" throughout; downstream references to "the Foundation" become "the Company". - Nonprofit-specific language ("not use Your Contributions in a way that is contrary to the public benefit or inconsistent with its nonprofit status and bylaws") removed — ProxySQL LLC is a commercial entity, so the reservation reduces to the simpler "Except for the license granted herein to the Company and recipients of software distributed by the Company, You reserve all right, title, and interest in and to Your Contributions." - Email-submission instructions and PDF form fields removed; signing is handled by the CLA assistant (https://cla-assistant.io) flow and identity is captured from the contributor's GitHub account. - Apache-specific references removed (Apache IDs, Apache mailing lists, Apache privacy policy). Substance of the license grants, definitions, and representations is preserved verbatim from the Apache templates. CONTRIBUTING.md gains a 'Contributor License Agreement' section linking to .github/CLA.md and describing the CLA assistant signing flow. Two browser steps still required to activate: 1. Sign in to cla-assistant.io as a sysown/proxysql admin 2. Point it at https://github.com/sysown/proxysql/blob/v3.0/.github/CLA.md (after this PR is merged so the raw URL resolves). No workflow file or webhook is needed; the bot operates via the installed GitHub App. Recommend legal review before enforcing on PRs.	1 month ago
René Cannaò	dbb40a30ba	Merge pull request #5647 from sysown/v3.0-slim-dbdeployer-images v3.0: slim dbdeployer images + test-harness and monitor fixes	1 month ago
Rene Cannao	0ce8693c29	ci: disable CI-taptests-groups auto-trigger (same 403 on jenkins-build-scripts) All four matrices (tests, tests/unit, tests_with_deps/*) of CI-taptests-groups fail at the 'Checkout jenkins_build_scripts' step with: fatal: unable to access 'https://github.com/proxysql/jenkins-build-scripts/': The requested URL returned error: 403 Same root cause as the CI-repltests / CI-shuntest disabling in the previous commit: GH_TOKEN_PROXYSQL no longer has access to the proxysql/jenkins-build-scripts repo. Comment out the workflow_run trigger so the workflow stops firing automatically on every push; workflow_dispatch is preserved for manual re-runs once the secret is refreshed.	1 month ago
Rene Cannao	feed404031	ci: disable repltests/shuntest auto-trigger and fix 'make unit_tests' Two test-infrastructure fixes on this branch. 1) CI-repltests / CI-shuntest: every run since the upstream proxysql/jenkins-build-scripts access broke has been failing at the Checkout step with HTTP 403. Comment out the workflow_run trigger so they no longer auto-fire on every push; keep workflow_dispatch so they can still be manually re-run once the token/repo-access is fixed. Same root cause affects CI-taptests-groups as well — that needs the GH_TOKEN_PROXYSQL secret refreshed, not a code change. 2) test/tap/tests/unit/Makefile: 'make unit_tests' (invoked from test/tap/Makefile via $(MAKECMDGOALS)) failed with make[1]: *** No rule to make target 'unit_tests' because the unit Makefile only exposed 'all' and 'debug'. Commit `16e94131f` intentionally forwards MAKECMDGOALS so 'make build_tap_test_debug' propagates 'debug' to the submake, which is what CI-unit-tests-asan-coverage relies on for -DDEBUG-guarded __thread stubs like mysql_thread___session_debug. Add a phony 'unit_tests: all' alias so the same pass-through also works when the goal name is 'unit_tests'. 'debug' is unchanged.	1 month ago
Rene Cannao	04afb09886	ci: retrigger CI after MySQL APT keyring fix (#5649 )	1 month ago
Rene Cannao	deb1cbe93e	test_cluster_sync: widen sync timeout, fix own-vs-remote query mix-up Two test-only fixes for the cluster-sync flakes on legacy-g5 / mysql84-g5 / mysql90-g5 when groups run concurrently. 1) get_checksum_sync_timeout() used ((cluster_check_interval_ms/1000) * cluster_check_status_frequency) + 1 which integer-truncated the ms→s conversion (1001ms → 1s) and left only 1s of slack on top of a full status-collection cycle. A single cycle slipping under CI load — exactly what `bce08084b` exposed — was enough to time out. Compute in ms, round up, and enforce a 30s floor. Fast path still exits on the first synced poll, so healthy clusters see no additional latency. 2) The "Primary has detected checksum in the replica" and "Primary has detected its own new checksum" assertions formatted the same template with the same args (conn_opts = primary in this call site), so both polled the identical query for the primary's own row. Distinguish them: the replica-check filters on r_admin host/port, the own-check filters on admin host/port. The unused prim_own_sync_t[] duplicate template is dropped. The second wait_for_node_sync() now consumes wait_own_chksm_sync instead of wait_remote_chksm_syn. Builds cleanly; runtime verification deferred to CI.	1 month ago
Rene Cannao	ee6d4e506a	ci: retrigger CI after MySQL APT key fix	1 month ago
Rene Cannao	25c8fd1609	test/tap/tests: drain in-flight monitor cycle in pgsql-servers_ssl_params-t The test samples PgSQL_Monitor_ssl_connections_OK immediately after LOAD PGSQL SERVERS TO RUNTIME, then sleeps 5s and checks the counter did not advance (TLSv1 pin should have broken SSL). A monitor connect cycle scheduled with the OLD params can land AFTER ok_before is taken but BEFORE the sleep, succeeding and falsely bumping the counter. Poll for two consecutive equal 1s samples so ok_before is recorded after the new config has taken effect and the counter has plateaued.	1 month ago
Rene Cannao	96c686dafa	MySQL_Monitor: GR thread ignores cached ping state on first iteration Why: mysql_server_ping_log can carry stale "unpingable" entries from before monitor_GR_thread_HG was (re)started (e.g. a previous test left those hostnames marked bad). With the cache filter applied on the very first cycle, find_resp_srvs() returns empty, the writer HG stays empty for a full healthcheck_interval, and GR-based tests flake on warm-up. Probe all configured hosts on the first iteration so subsequent cycles see a fresh ping_log. The filter resumes from iteration two onward.	1 month ago
Rene Cannao	96bd55784d	test/tap/tests: wire mysql-zstd_compression_level_libmysql-t into tests target Commit `174c5bda2` added the pattern rule for the _libmysql variant of the zstd_compression_level test but did not add it to the 'tests:' target list, so 'make build_tap_tests' silently skipped building it. Add the missing entry so the libmysql variant is built alongside its siblings.	1 month ago
Rene Cannao	bce08084b8	test_cluster_sync: wait_for_node_sync must treat "row absent" as not-synced wait_for_node_sync() in test_cluster_sync-t.cpp defaults row_value to -1, updates it from the result row only when both myrow and myrow[0] are non-null, then checks: if (row_value == 0) { not_synced = true; ... } This catches "row present but empty" (row_value == 0) but silently misses "row absent" (row_value stays at the -1 default when mysql_fetch_row returns NULL). The common predicate passed through this helper is: SELECT LENGTH(checksum) FROM stats_proxysql_servers_checksums WHERE hostname='%s' AND port='%d' AND name='%s' That row does not exist on a freshly-spawned replica until it has first synchronised checksums with the peer whose hostname/port the test is polling. Under heavy CI load the replica takes longer to reach that point; the loop exits as "synced" before the row exists; the caller's fetch_remote_checksum() then correctly returns empty and aborts the test with "Failed to fetch current checksum for module 'mysql_servers_v2'". That is exactly the failure observed in legacy-g5 / mysql84-g5 / mysql90-g5 during the multi-group CI run. mysql95-g5 passed because it started roughly two hours later, after concurrent groups had finished. Fix: treat row_value <= 0 as not-synced, so the "row absent" case is polled for the full CHECKSUM_SYNC_TIMEOUT instead of being silently skipped. Positive values still mean "synced" and exit the loop immediately, so there is no latency cost in the fast path. Verified: test_cluster_sync-t now passes 399/399 in legacy-g5; the 'wait_for_node_sync' timeouts that now appear in the log are the disabled-sync half of paired enabled/disabled assertions, where the timeout is the expected outcome.	1 month ago
Rene Cannao	2b1a4262b1	tap/utils: add wait_for_log_match() to fix log-scrape races check_logs_for_command() in pgsql-copy_from_test-t did a single-shot call to get_matching_lines() right after the SQL that was expected to cause the "Switching to Fast Forward" / "Switching back to Normal" log line. Because ProxySQL writes that line asynchronously with respect to the query's reply, the scan occasionally reached EOF before the producer had flushed the line, turning the assertion into a flake. The failure that originally surfaced this was legacy-g1, test 85/327 of pgsql-copy_from_test-t: 1 of 236 assertions (not ok 126 - "Switching back to Normal mode"). It never reproduced under three targeted reruns, which is consistent with a timing race rather than a logic bug. Two changes: 1. Add wait_for_log_match() in test/tap/tap/utils.{h,cpp}: retry get_matching_lines() up to a timeout, clearing both eofbit and failbit between iterations so successive getline() calls pick up lines appended since the last scan. (Clearing only failbit, as get_matching_lines does internally, is not enough — once a stream has hit EOF its sticky eofbit keeps getline() from reading new bytes.) Defaults: 2000 ms total budget, 100 ms between polls — matching the idiom already used in admin_set_credentials_logging-t. 2. Rewrite check_logs_for_command() in pgsql-copy_from_test-t.cpp to call the new helper. Positive assertions still return as soon as the line appears (no added latency in the common case). Negative assertions (check_logs_for_command(...) == false) now wait up to the full timeout to confirm absence, closing the symmetrical race where a line would have arrived a few hundred ms later than the single-shot scan. Verified: 3/3 reruns of pgsql-copy_from_test-t in legacy-g1 post-fix all pass 236/236, runtime ~40 s per run (single-shot version was slightly faster because it short-circuited on negative assertions without waiting).	1 month ago
Rene Cannao	6322d58376	sqlite3: CACHING_SHA2_PASSWORD() accepts optional rounds arg (closes #5640 ) The ProxySQL admin SQLite3 extension CACHING_SHA2_PASSWORD() previously always emitted hashes with 5000 rounds (hardcoded "\$A\$005\$" prefix and "\$5\$rounds=5000\$" crypt salt). MySQL's caching_sha2_password_digest_rounds is configurable from 5000 to 4095000 in steps of 1000, so a backend that uses any non-default value produces hashes the extension cannot reproduce — breaking the byte-for-byte hash equivalence test in test_sqlite3_pass_exts-t on e.g. the mysql95 backend where rounds=10000. Extend the SQLite function to accept an optional third integer argument carrying the desired rounds value: CACHING_SHA2_PASSWORD(pass) -> random salt, 5000 rounds CACHING_SHA2_PASSWORD(pass, salt) -> given salt, 5000 rounds CACHING_SHA2_PASSWORD(pass, salt, rounds) -> given salt, given rounds rounds is validated: integer, multiple of 1000, in [5000, 4095000] — matching MySQL's caching_sha2_password_digest_rounds bounds. The "\$5\$rounds=<N>\$" salt prefix and "\$A\$<RRR>\$" hash prefix are now built at runtime from the resolved value, with <RRR> emitted as 3-char zero-padded uppercase hex of (rounds/1000) per MySQL's sql/auth/sha2_password.cc::digest_round_separator(). Update test_sqlite3_pass_exts-t to parse the rounds field from the MySQL-generated hash (chars 3..5 of \$A\$<RRR>\$..., base-16, ×1000) and pass it to CACHING_SHA2_PASSWORD() — so the hash-equivalence assertion matches any backend rounds. Update the INV_INPUTS fixture: 3-arg is now valid, so reject 4-arg ("wrong number"), reject 3-arg with non-integer rounds ("Invalid argument type"), reject out-of-range or non-multiple-of-1000 rounds ("Invalid rounds: ..."). Verified against mysql95 backend (rounds=10000): test_sqlite3_pass_exts-t passes 2117/2117, test_auth_methods-t passes 10774/10774 (previously failing with 50+122 not-ok respectively before the base-16 parse fix in MySQL_Protocol.cpp).	1 month ago
Rene Cannao	4e0efd7536	scripts/lint: pass explicit target to bear-wrapped make generate-compile-commands.sh wrapped the top-level build with bear -- make -j\$(nproc) with no explicit target, so the inner make resolved its default goal dynamically. Combined with the previous Makefile bug that left lint-generate-cdb as the default goal, invoking `make` with no arguments produced an exponential fork bomb: lint-generate-cdb → this script → bear -- make -jN → lint-generate-cdb → this script → …, with each level forking up to \$(nproc) concurrent sub-makes under Bear's LD_PRELOAD exec intercept. Spawned hundreds of `bear -- make -jN` processes before grinding the system to a halt. The Makefile default-goal fix in `7a1d639a4` closes the trigger, but the script itself should also be robust against a future re-introduction of the same hazard. Pass `build_src` explicitly as the target so the inner make never re-enters lint-generate-cdb regardless of what the default goal happens to be.	1 month ago
Rene Cannao	7a1d639a4b	build: set default goal to build_src, not the first target The Makefile had `.DEFAULT: default` intending to make the `default` target (which builds the project) the fallback when `make` is invoked with no arguments. `.DEFAULT` is a different special target — it provides a fallback recipe for targets with no rule, and has no effect on default goal selection. GNU make picks the first real target as the default goal unless `.DEFAULT_GOAL` is set. The first real target in this Makefile is `lint-generate-cdb` (line 19), so `make` with no arguments ran `scripts/lint/generate-compile-commands.sh`, which does `bear -- make -j\$(nproc)` — a full rebuild of the project under Bear's compile-intercept wrappers. Each C++ compile under Bear spawned a wrapper process, producing hundreds of bear processes per plain `make` invocation and burying the intended default behaviour. Replace `.DEFAULT: default` with `.DEFAULT_GOAL := default` so `make` with no arguments runs `default: build_src` as originally intended.	1 month ago
Rene Cannao	4e39f3bdc4	fix: parse caching_sha2_password rounds as hex (auth fails on rounds >= 10000) MySQL stores the rounds field of caching_sha2_password hashes ($A$<RRR>$<salt><hash>) as 3-char zero-padded uppercase hex of (rounds/1000). See sql/auth/sha2_password.cc: sprintf(rounds_str, "%03X", m_stored_digest_rounds) ProxySQL was parsing this field with stol(s, 10) (default base-10), which silently truncates at the first hex digit (A-F): 005 → 5 (5000 rounds, MySQL 8.0 default) ✓ matches base-16 009 → 9 (9000 rounds) ✓ matches base-16 00A → 0 (10000 rounds, stops at 'A') ✗ should be 10 010 → 10 (16000 rounds, decimal!) ✗ should be 16 01A → 1 (26000 rounds, stops at 'A') ✗ should be 26 The wrong rounds value is then fed into sha256_crypt_r() via "$5$rounds=<n*1000>$<salt>", producing a digest that does not match the stored hash, so verification fails with Access denied. Trigger: any MySQL backend with caching_sha2_password_digest_rounds >= 10000 (or any value whose hex encoding contains A-F). Surfaces in CI on the mysql95 dbdeployer image, where backend-generated hashes start with $A$00A$. The bug is latent on MySQL 8.0/8.4/9.0 with default rounds=5000 because "005" parses identically in base-10 and base-16. Fix: pass base=16 to stol() in both PPHR_verify_sha2() and PPHR_sha2full().	1 month ago
Rene Cannao	ef351a6a67	test/groups: drop reg_test_3992 from mysql8x/95 groups The reg_test_3992_fast_forward_malformed_packet-{t,pymysql,mysqlsh} tests hard-code MariaDB-specific users (mariadbuser, mariadbuserff) that are provisioned only by infra-{mariadb10,dbdeployer-mariadb10}. The mysql84-g2, mysql90-g2 and mysql95-g2 groups don't include a MariaDB infra, so the test fails immediately with "Access denied for user 'mariadbuserff'". Restrict the three variants to legacy-g2 and the mysql-* knob-variant g2 groups (which inherit the legacy infra layout including MariaDB).	1 month ago
Rene Cannao	b860662c17	test/infra: slim dbdeployer images (~70-77% smaller) Every dbdeployer-* CI image was carrying dead weight from the MySQL/MariaDB unpack step. Cleanup is folded into the same RUN layer as the unpack so the removals actually reclaim space instead of leaving it in a prior layer: - remove the tarball that 'dbdeployer downloads get-unpack' left at / (~1GB) - remove bin/mysqld-debug (debug server binary, unused at runtime) - for MySQL 5.7 only: also remove lib/libmysqld.a and lib/libmysqld-debug.a (embedded-server static libs, deprecated in 5.7 and removed in 8.0) - strip all ELF binaries under bin/ and shared libs under lib/ using binutils installed temporarily and purged within the same layer Result: 2.8 GB -> 653 MB for mysql95, with similar drops across every tier. Existing behavior is preserved (libmysqlclient.a kept, no runtime changes).	1 month ago
René Cannaò	82fadbc1f4	Merge pull request #5623 from sysown/v3.0-issue5621 Fix regression: restore upper-word capabilities dropped from greeting	1 month ago
René Cannaò	17a54193cc	Merge pull request #5637 from sysown/v3.0_new_zstd feat: add mysql-zstd_compression_level variable	1 month ago
Rene Cannao	4d251c80ec	Revert 'enable_client_session_tracking' variable; keep greeting fix Issue #5621 mis-diagnosed "CLIENT_SESSION_TRACKING missing from the greeting" as a long-standing configuration gap and proposed a new 'mysql-enable_client_session_tracking' variable analogous to 'mysql-enable_client_deprecate_eof'. On closer inspection this was in fact a regression introduced by commit `8c6a6444d` (zstd refactor), which dropped the hard-coded upper-word bits from the greeting. The right fix is simply to restore those hard-coded bits — commits `536c9d916` and `6b311b701` already do that via the 'extended_capabilities' local in 'generate_pkt_initial_handshake'. The new variable adds no value on top of that: the hard-coded bits cannot be disabled through 'mysql-server_capabilities' (which is OR-combined with the greeting), and there is no protocol hazard motivating a per-instance off-switch for CLIENT_SESSION_TRACKING, unlike CLIENT_DEPRECATE_EOF. Revert the variable (MySQL_Thread.h / proxysql_structs.h / MySQL_Thread.cpp / MySQL_Protocol.cpp) and restore CLIENT_SESSION_TRACKING to the 'extended_capabilities' literal, matching v3.0.6 exactly. Keep the test's doxygen / diag / baseline-check rewrite from the investigation — it's strictly better coverage and will catch any future regression of the hard-coded bits early. Test plan reduced to 2 scenarios (deprecate_eof on/off) since session_tracking is now unconditionally advertised.	1 month ago
Rene Cannao	174c5bda20	test: expand zstd_compression_level test with MySQL 8.4 C API and timing benchmark The existing test only verified admin variable behavior (default, range, independence). It did not prove that ZSTD compression was actually active on the wire. Two new test variants address this: MariaDB connector path (mysql-zstd_compression_level-t): - Tests 1-9: unchanged (variable defaults, range validation, independence) - Tests 10-11: use mysql CLI via execvp with --compression-algorithms=zstd and --zstd-compression-level={3,19}. Each call spawns a new process (= new connection). Verifies CONNECTION_ID() retrieval and large resultset transfer. Skipped if mysql CLI lacks zstd support. MySQL 8.4 connector path (mysql-zstd_compression_level_libmysql-t): - Tests 1-9: same as MariaDB path - Test 10: establishes a ZSTD connection using mysql_options( MYSQL_OPT_COMPRESSION_ALGORITHMS, "zstd") and prints mysql_thread_id() without running any query. - Test 11: timing benchmark proving compression is active. Creates a query rule (rule_id=1, cache_ttl=60000) to cache all SELECTs, then runs a 576-row resultset query 1000 times on a plain connection and 1000 times on a ZSTD level 22 connection. ZSTD level 22 is extremely CPU-intensive: observed 40122ms vs 858ms (46.7x ratio). A ratio >= 3x proves compression is really running on the client<->ProxySQL link. Other changes: - MySQL_Protocol.cpp: simplify redundant clamping to a simple cast (the variable is already validated at SET time) - Makefile: add build rule for _libmysql variant linking against MySQL 8.4 connector - groups.json: register _libmysql variant for mysql84/90/95-g1 - CLAUDE.md: update testing section with CI script usage and DO NOT list	1 month ago
Rene Cannao	6b311b7014	Restore greeting upper-word caps via local 'extended_capabilities' Previous commit (`536c9d916`) OR-ed the dropped upper-word bits into 'mysql_thread___server_capabilities', which made them leak into '(myds)->myconn->options.server_capabilities' — a behaviour delta vs. pre-8c6a6444d. Revert that and reinstate the original pattern: a local 'extended_capabilities' whose only consumer is 'upper_word'. Seed it with the hardcoded baseline (CLIENT_MULTI_, CLIENT_PLUGIN_AUTH, CLIENT_REMEMBER_OPTIONS) and fold in any other upper-word bits currently set in 'server_capabilities' (CLIENT_DEPRECATE_EOF, CLIENT_SESSION_TRACKING, CLIENT_ZSTD_COMPRESSION, ...) so the greeting stays in sync with the per-session / per-toggle state — preserving 8c6a6444d's zstd advertisement while restoring the #4023 baseline.	1 month ago
Rene Cannao	536c9d9169	Restore greeting upper-word capabilities dropped in `8c6a6444d` The zstd refactor in `8c6a6444d` replaced the per-greeting 'extended_capabilities' local with 'server_capabilities' derived from 'mysql_thread___server_capabilities'. As a side effect, the four upper-word bits the greeting used to hardcode were no longer advertised: CLIENT_MULTI_STATEMENTS CLIENT_MULTI_RESULTS CLIENT_PS_MULTI_RESULTS CLIENT_REMEMBER_OPTIONS OR these bits back into 'mysql_thread___server_capabilities' before the upper-word memcpy, matching what real MySQL 8.x advertises (see #4023). Also rewrite test_greeting_capabilities-t with doxygen documentation and step-by-step diag() tracing that decodes capability bitmasks to names, and reinstate the baseline-capability assertion that silently degraded when commit `6e1483f1f` tightened the caps_match overwrite into '&='. In 'PPHR_2', fold the CLIENT_SESSION_TRACKING enforcement into the shared two-step comment block for clarity.	1 month ago
Rene Cannao	f9f6f74ecc	fix: address code review issues in zstd compression level test - Remove mysql_error(NULL) UB on failed admin connection (use static error string instead). - Correct Test 10 comment and label: it exercises zlib (CLIENT_COMPRESS), not zstd compression, verifying the new variable doesn't break existing compressed connections.	1 month ago
Rene Cannao	4631b1f466	fix: clean groups.json diff — only add mysql-zstd_compression_level-t entry	1 month ago
Rene Cannao	6dae8d125f	test: add TAP tests and housekeeping for mysql-zstd_compression_level - New test mysql-zstd_compression_level-t.cpp: validates default value, range enforcement (1-22), rejection of 0 and 23, independence from mysql-protocol_compression_level, and functional compressed connection. - Register mysql-zstd_compression_level-t in groups.json (same groups as mysql-protocol_compression_level-t). - Add variable visibility check to reg_test_2793-compression-t.cpp. - Add unit test binary pattern to .gitignore.	1 month ago
Rene Cannao	1c3057ecf2	feat: use mysql-zstd_compression_level for zstd compression Replace mysql_thread___protocol_compression_level with mysql_thread___zstd_compression_level in both the handshake (MySQL_Protocol.cpp) and data stream (mysql_data_stream.cpp) zstd level selection paths, so the two compression algorithms can be configured independently.	1 month ago
Rene Cannao	aa24529f39	feat: register mysql-zstd_compression_level variable in MySQL thread	1 month ago
Rene Cannao	266a10617f	feat: declare mysql_thread___zstd_compression_level thread-local and struct member	1 month ago
René Cannaò	aecac96c84	Merge pull request #5638 from sysown/fix/lint-groups-json-format fix: reformat groups.json and add --fix to linter	1 month ago
Rene Cannao	a0f685b998	fix: reformat groups.json to one-line-per-entry and add --fix to linter groups.json was reformatted with multi-line arrays by a recent merge, breaking the one-line-per-test convention documented in README.md. Changes: - Reformat groups.json to canonical format (one line per test, compact arrays, sorted keys, space before colon). - Add --fix mode to lint_groups_json.py that auto-reformats the file in-place, preventing this class of issue in the future.	1 month ago
René Cannaò	13ff9d7671	Merge pull request #5625 from sysown/v3.0-dbdeployer-mysql84-gr infra: migrate mysql84-gr to dbdeployer and cap pgsql_servers max_connections	1 month ago
Rene Cannao	7c6f42bcd8	fix: MySQL 9.x charset handling and log_last_insert_id test race - validate_charset: compare server_version as a numeric major version instead of testing the first character, so MySQL 9.x and later are treated like 8.x for collations with id >= 255 - test_log_last_insert_id-t: re-enable PROXYSQL FLUSH LOGS; with buffered query logging, eventslog_flush_timeout=0 alone still races the periodic flush thread	1 month ago
Rene Cannao	d212102ce5	fix(tests): drop pgsql-* tests from mysql{84,90,95}-g[1-5] Follow-up to `4534e5b4d` (merge legacy-g[1-5] into mysql{84,90,95}-g[1-5]). The previous merge was over-inclusive: legacy-g* contains PostgreSQL tests (named 'pgsql-' or 'pgsql_' or '_pgsql_') that exercise ProxySQL's PostgreSQL front-end against the docker-pgsql16-single backend. Those tests don't care which MySQL version is also present, and the mysql{84,90,95} groups' infras.lst only lists infra-dbdeployer-mysqlXX — no PostgreSQL backend is brought up. So in myrun3 the merged pgsql tests failed (no backend) or hung until the per-group timeout (test_ffto_pgsql_concurrent-t). Removed 153 group memberships (pgsql-containing test names) from mysql{84,90,95}-g[1-5]. Safety check confirmed: no pgsql-containing test was in mysql-g before the merge, so this revert is scoped exactly to memberships the merge added. New sizes: legacy mysql84 mysql90 mysql95 g1 87 78 72 78 g2 43 45 45 45 g3 40 40 40 40 g4 94 61 61 61 g5 5 6 6 6 Net result vs pre-merge (pre-4534e5b4d): - mysql84-g1 back to 78 (same as pre-merge) - mysql90-g1 up from 16 to 72 (the non-pgsql legacy tests; +56) - mysql95-g1 back to 78 (same as pre-merge) - g2 / g3 / g5 unchanged from pre-merge for 84/95, mysql90-g* gets a few non-pgsql additions - g4 keeps the non-pgsql portion of the legacy merge (61 vs 60) Legacy-g[1-5] are untouched — pgsql tests remain where they belong. check_groups.py reports 385/385 registered, no orphans.	1 month ago
Rene Cannao	4534e5b4dc	feat(tests): merge legacy-g[1-5] test sets into mysql{84,90,95}-g[1-5] Every TAP test in legacy-gN is now also a member of mysql84-gN, mysql90-gN, and mysql95-gN (N=1..5). No tests removed from any group; legacy-g[1-5] keeps its existing membership intact. Rationale: the legacy group covers the MySQL 5.7 / MariaDB 10 / PgSQL / ClickHouse matrix, but many of its TAP tests are not inherently tied to MySQL 5.7 — they exercise ProxySQL behavior that should also hold against modern MySQL backends (8.4, 9.0, 9.5). Running them against those newer servers widens the matrix and catches 9.x-specific regressions in tests that weren't previously in a 9.x group. Mechanical merge via a walk over groups.json: for each test: for N in 1..5: if 'legacy-gN' in memberships: add 'mysql84-gN', 'mysql90-gN', 'mysql95-gN' if absent 224 new memberships added. Per-bucket union size (i.e. now a superset of legacy-gN): g1: legacy=87 mysql84=93 mysql90=87 mysql95=93 g2: legacy=43 mysql84=45 mysql90=45 mysql95=45 g3: legacy=40 mysql84=40 mysql90=40 mysql95=40 g4: legacy=94 mysql84=96 mysql90=96 mysql95=96 g5: legacy=5 mysql84=7 mysql90=7 mysql95=7 Verified as supersets (every test in legacy-gN is present in mysql{84,90,95}-gN). check_groups.py reports 385/385 tests registered, no orphans. Expect new 9.x-specific failures to surface when the mysql{90,95}-g* groups run these newly-inherited legacy tests — they're likely to hit the same family of mysql_native_password / caching_sha2_password issues we've already been working through in the sibling PR threads. Those will be triaged as they appear.	1 month ago
Rene Cannao	d5b796cdea	fix(test/tap): halve expected backend conn count in test_sqlite3_pass_exts-t on 9.x Follow-up to `cbbd10ba7` (skip native_password backend fixtures on MySQL 9.x). The Phase-3 end-to-end loop runs RAND_USERS_GEN (100) iterations on 8.4 — 50 with mysql_native_password + 50 with caching_sha2_password — and each successful login creates one backend connection. On 9.x the native half is skipped, so only 50 actual connections are made. The post-loop assertion "Number of backend conns created should match conn attempts" was hardcoded to compare against RAND_USERS_GEN. When native was skipped, actual=50 but expected=100, producing: not ok 2113 - Number of backend conns created should match conn attempts exp:'100', act:'50' Fix: compute expected_conns based on g_mysql_supports_native_password (the same flag that gates the Phase-3 loop range) and use it both for the wait_for_cond() condition string and the assertion. Verified on a live mysql90 backend: test now reports ok 2113 - ... exp:'50', act:'50' and exits 0 (2113/2113 assertions pass in 3.4s).	1 month ago
Rene Cannao	cbbd10ba74	fix(test/tap): skip native_password backend fixtures on MySQL 9.x in test_sqlite3_pass_exts-t Fixes #5631. The test has three logical phases: Phase 1 Admin SQLite3 extension validation for MYSQL_NATIVE_PASSWORD() and CACHING_SHA2_PASSWORD() — pure hash computation inside ProxySQL, does not touch the backend. Phase 2 MySQL/Admin hash compatibility — creates USER_GEN_COUNT (100) users on the backend, half with mysql_native_password and half with caching_sha2_password, then compares the backend's authentication_string with the hash that ProxySQL's Admin SQLite3 function produces. Phase 3 End-to-end connection test — creates RAND_USERS_GEN (100) users with the same 50/50 split and attempts to connect through ProxySQL. On MySQL 9.0+ the mysql_native_password plugin is not loadable, so the first CREATE USER IDENTIFIED WITH 'mysql_native_password' in Phase 2 fails with ER_PLUGIN_IS_NOT_LOADED and the test bails after only 10 assertions (all of Phase 1 passes — ProxySQL's internal hash extensions work fine on 9.x; this was a good green signal). Fix: - Add g_mysql_supports_native_password; set to false on server_version >= 9.0. - Adjust actual_test_count so plan() matches reality: if (has_native) { count += USER_GEN_COUNT + RAND_USERS_GEN; } else { count += USER_GEN_COUNT / 2 + RAND_USERS_GEN / 2; } - Wrap the Phase 2 native_password loop in 'if (has_native)'. - Phase 3 uses 'for (i = 0 .. RAND_USERS_GEN)' with 'i < 50 ? native : sha2'. Skip the native half by starting the loop at RAND_USERS_GEN/2 when has_native is false. Phase 1 coverage is fully preserved — the ProxySQL Admin SQLite3 hash functions (including MYSQL_NATIVE_PASSWORD()) still work on 9.x; only the backend-side CREATE USER path is skipped.	1 month ago
Rene Cannao	5bb19a99b1	fix(test/tap): skip reg_test_4935-caching_sha2-t on MySQL 9.x Fixes #5629. This regression test for issue #4935 ("proxysql ignores database in DSN at first login") uses a hardcoded plan(864) built around a 2×2 fixture table (2 mysql_native_password users + 2 caching_sha2_password users) crossed with 2 def_auths × 3 req_auths × 2×2×2 bool matrices. On MySQL 9.0+ the mysql_native_password plugin is unavailable and any CREATE USER IDENTIFIED WITH 'mysql_native_password' hits ER_PLUGIN_IS_NOT_LOADED, killing the test during fixture setup. Filtering the fixture matrix in-place to remove the native half would require re-deriving plan(N) from scratch — the current plan(864) encodes the expected-pass count of a non-trivial loop that has a per-iteration 'exp_first_login_failure()' skip. Getting that math right and keeping it right under future changes is fragile. Pragmatic fix: detect 9.x and skip the whole test with plan(0). The #4935 regression (database-in-DSN on first login for caching_sha2) is already fully exercised against the mysql57 and mysql84 infras in legacy-g* / mysql84-g*. Losing the 9.x-specific coverage for this one regression test is an acceptable trade-off — the ProxySQL code path under test is server-version-agnostic, so additional 9.x coverage adds little beyond what 8.4 already provides. Mirrors the skip pattern the test suite already uses for MySQL < 8.0.	1 month ago
Rene Cannao	5f50724845	fix(test/tap): skip mysql_native_password fixtures on MySQL 9.x in test_auth_methods-t Fixes #5626. The test's fixture table (backend_users) has 18 users split 9/9 across mysql_native_password and caching_sha2_password. On MySQL 9.0+ the native-password plugin is unavailable, and the test bailed on creating the first native-password fixture (dualpass2). Following the existing version-gating pattern in the same test (which already branches on server_version < 80000 for MySQL 5.x and on is_mariadb for MariaDB), add a third branch: has_native_password = true if server_version >= 90000 && !is_mariadb: has_native_password = false Then thread has_native_password through: - the backend_users fixture list (filter out native entries before passing to config_mysql_backend_users) - def_auths (the "ProxySQL default authentication plugin" matrix) - req_auths (the "client-requested authentication plugin" matrix) plan() adjusts automatically because the plan count is computed from scs_stats/rnd_scs_stats which are derived from the actual created users and config_combs, not hardcoded. Coverage on 5.7 and 8.4 is unchanged: both plugins still exercised. On 9.x, only the caching_sha2_password half runs — which is the only half the backend can support anyway.	1 month ago
Rene Cannao	8c36be3606	fix(test/tap): make create_mysql_user helper backend-version aware Fixes #5627, #5628, #5630. The shared test helper test/tap/tap/utils.cpp:create_mysql_user() was hardcoding IDENTIFIED WITH 'mysql_native_password' in its CREATE USER template. MySQL 9.0 removed that plugin entirely (the .so in the tarball is a client-lib stub only; INSTALL PLUGIN fails with "Can't find symbol '_mysql_plugin_interface_version_'"), so CREATE USER IDENTIFIED WITH 'mysql_native_password' returns ER_PLUGIN_IS_NOT_LOADED on 9.x backends. Tests that use this helper as setup — they don't care which auth plugin the backend uses, they just need a working user — were failing before executing a single assertion. Affected tests (all in mysql90-g* / mysql95-g*): test_com_reset_connection_com_change_user-t (#5627) test_default_value_transaction_isolation_attr-t (#5628) test_server_sess_status-t (#5630) Fix: select the IDENTIFIED clause at runtime based on mysql_get_server_version(): - server < 9.0 → keep the explicit 'mysql_native_password' clause (preserves behavior on 5.7 / 8.4 for any callers that implicitly depend on a native-password backend user) - server >= 9.0 → use IDENTIFIED BY (server default plugin, which is caching_sha2_password on 9.x) Auth-sensitive tests that specifically exercise plugin-iteration (e.g. test_auth_methods-t, reg_test_4935-caching_sha2-t) do not use this helper — they use create_mysql_user_exp_creds() in utils_auth.h, which takes an explicit plugin argument. Those are handled in separate commits.	1 month ago
Rene Cannao	e11411fde2	fix(infra): use 'mysql -E -e STATEMENT' instead of 'mysql -e STATEMENT\G' MySQL 9.5.0 client (no longer?) accepts '\G' as a statement terminator when the statement is passed via '-e': $ mysql -e 'SHOW REPLICA STATUS\G' ERROR at line 1: Unknown command '\G'. '\G' still works interactively, and still works in -e on 8.4 and 9.0.1. This is a silent behavior change in the 9.5 client binary. (The 9.0.1 client in the same tarball tree handles '\G' fine.) docker-mysql-post.bash uses this to query replication status after container startup. Under 9.5, the mysql command exits non-zero, the captured output is empty, the subsequent grep in a pipeline returns 1, and set -e + pipefail kill docker-mysql-post.bash. That in turn kills docker-compose-init.bash which kills ensure-infras.bash. Net effect: every mysql95 / mysql95-binlog group fails before running a single test, logging just: Verifying MySQL on ...:3308... OK Checking replication on port 3307... (stops here) Fix: switch from '-e "SHOW REPLICA STATUS\G"' to '-E -e "SHOW REPLICA STATUS"'. '-E' (a.k.a. --vertical) has been supported by mysql(1) since 3.x, works identically on MariaDB 10, MySQL 5.7 / 8.4 / 9.0 / 9.5. The grep/awk that extracts Replica_IO_Running and Replica_SQL_Running keeps working unchanged since the vertical output format is the same. Applied to all 10 docker-mysql-post.bash files under test/infra/infra-dbdeployer-*. No change needed to the ci-infra Docker images — docker-mysql-post.bash runs on the host after the container starts.	1 month ago
Rene Cannao	bee37b2cb8	fix(tester): abort when TAP_GROUP is set but has no tests registered Previously, if TAP_GROUP was set but the group had no entries in test/tap/groups/groups.json (typo, missing registration, etc.), the tester logged a warning and silently fell back to running ALL 325 TAP tests. In a multi-group parallel run this turns a misconfiguration into a multi-hour hang that only surfaces when the per-group TIMEOUT fires. We hit this in practice: mysql90-g[2-5] existed as launch targets in the runner but hadn't been populated in groups.json yet. The tester fell back to the full suite, and the groups TIMED OUT after chewing through pgsql tests they had no business running. Change behavior to log at CRITICAL and sys.exit(1) in two cases: 1. TAP_GROUP is set, groups.json is present, but the group has no tests registered. 2. TAP_GROUP is set and groups.json is missing entirely. The "no TAP_GROUP specified" path is unchanged — that's the legitimate way to run the full suite directly (outside the group runner). Error messages explicitly tell the operator what to do: either register tests for the group, or unset TAP_GROUP to run the full suite intentionally.	1 month ago

1 2 3 4 5 ...

9684 Commits (f815b778126b517ae1f07a2a4e33701c79f33efe) All Branches Search

9684 Commits (f815b778126b517ae1f07a2a4e33701c79f33efe)

All Branches