When running with --resume option the ALTER runs on the new table
even if the new table exists and has already been altered.
This causes a deterministic failure every time the --resume option
is used. This fix tests if --resume is given and does not run
ALTER on the new table. If pt-osc did not successfully alter
the new table during the previous invocation, this could cause the
ALTER to not be applied.
The right fix would be to compare the DDL of the _new table to
the proposed DDL generated by pt-osc and only run ALTER if they
do not match.
pt-table-sync now uses up to 17 decimal digits when writing
floating point numbers in the generated SQL statements.
This is necessary to prevent unintended data changes.
The MySQL driver DBD::mysql does not decode JSON values as utf8
although MySQL uses utf8mb4 for all JSON strings.
This change decodes JSON values as utf8 (when not already done)
such that SQL statements are generated correctly.
Previously, pt-table-sync generated DML statements that included
the generated columns of a table, which is however rejected by
the database and not necessary for the required sync statements.
- Implemented functions get_replica_name and get_source_name in lib/MasterSlave.pm I did not implement similar functions in other places, because they set other variables as well, not only replica_name or source_name.
- s/slave lag/replica lag/ in tests
- reverted rename of COM_REGISTER_SLAVE to COM_REGISTER_REPLICA, because the constant still named COM_REGISTER_SLAVE in 8.4
- removed some slave leftovers from pt-replica-find
- Added test for replica lag check for pt-archiver
- Re-added deprecated slave- options
- Added tests for deprecation warnings and for legacy options for pt-archiver
- Removed practically not supported options --replica-user and --replica-password from pt-archiver, pt-kill, pt-query-digest
- Added legacy source/replica options (master/slave) variants and tests for pt-heartbeat, pt-online-schema-change, pt-replica-find, pt-replica-restart, pt-table-checksum, pt-table-sync
- Updated modules after lib/MasterSlave.pm change
- Moved data collection for THP from lib/bash/report_system_info.sh to lib/bash/collect_system_info.sh, so pt-summary reports THP status on the machine where samples were collected, not on the machine where an engineer examines samples.
- Removed runtime.txt after discussion with Anastasia Alexandrova
- Added "use VersionParser" into tests in t/lib when needed
- Removed word master from tests for pt-archiver, pt-config-diff, pt-deadlock-logger, pt-duplicate-key-checker, pt-find, pt-fk-error-logger, pt-heartbeat, pt-index-usage, pt-ioprofile, pt-kill, pt-mysql-summary
- Removed word slave from tests for pt-archiver, pt-config-diff, pt-deadlock-logger, pt-duplicate-key-checker, pt-find, pt-fk-error-logger, pt-heartbeat, pt-index-usage, pt-ioprofile, pt-kill, pt-mysql-summary
- Updated modules for pt-archiver, pt-config-diff, pt-deadlock-logger, pt-duplicate-key-checker, pt-find, pt-fk-error-logger, pt-heartbeat, pt-index-usage, pt-ioprofile, pt-kill, pt-mysql-summary
- Changed mysql_ssl patch, so it is now short option s
- Added a check for existing zombies in t/pt-kill/execute_command.t
- Added bin/pt-galera-log-explainer to .gitignore
- Reversed condition, so default is transaction_isolation, not
tx_isolation
- Removed extra variable: we have variable $version, defined before
- Fixed test cases
- Removed option print_identified_with_as_hex, because it was already
implemented in the PT-2190 fix
- Simplified patch
- Kept CREATE USER/ALTER USER sequence and extra DELETE from mysql.user
table
- Added test case
