Skip to content

Latest commit

 

History

History
572 lines (348 loc) · 23.9 KB

CHANGELOG.md

File metadata and controls

572 lines (348 loc) · 23.9 KB

Maxwell changelog

v1.8.0: "upbeat, honest, contradictory"

In version 1.8.0 Maxwell gains alpha support for GTID-based positions! All praise due to Henry Cai.

v1.7.2: "comparing self to better"

  • Fix a bug found where maxwell could cache the wrong TABLE_MAP_ID for a binlog event, leading to crashes or in some cases data mismatches.

v1.7.1: "blame it on your seratonin"

  • bootstrapping now can take a --where clause
  • performance improvements in the kafka producer

v1.7.0: "lucky me, lucky mud"

Maxwell 1.7 brings 2 major new, alpha features. The first is Mysql 5.7 support, including JSON column type support and handling of 5.7 SQL, but not including GTID support yet. This is based on porting Maxwell to Stanley Shyko's binlog-connector library. Thanks to Stanley for his amazing support doing this port.

The second major new feature is a producer for Amazon's Kinesis streams, This was contributed in full by the dogged and persistent Thomas Dziedzic. Check it out with --producer=kinesis.

There's also some bugfixes:

  • Amazon RDS heartbeat events now tick maxwell's position, thx Scott Ferguson
  • allow CHECK() statements inside column definitions

v1.6.0: "give me a quest"

This is mostly a bugfix release, but it gets a minor version bump due to a single change of behavior: dates and timestamps which mysql may accept, but are considered invalid (0000-00-00 is a notable example) previously had inconsistent behavior. Now we convert these to NULL. Other bugfixes:

  • heartbeats have moved into their own table
  • more fixes around alibaba rds
  • ignore DELETE statements that are output for MEMORY tables upon server restart
  • allow pointing maxwell to a pre-existing database

v1.5.2: "french banana"

  • add support for kafka 0.10.1 @ smferguson
  • master recovery: cleanup positions from previous master; prevent errors on flip-back.
  • fix a bug that would trigger in certain cases when dropping a column that was part of the primary-key

v1.5.1: "1.5.1 is just 1.5.1"

This is a bugfix release.

  • fixes for bootstrapping with an alternative maxwell-schema name and an include_database filter, thanks Lucian Jones
  • fixes for kafka 0.10 with lz4 compression, thanks Scott Ferguson
  • ignore the RDS table mysql.ha_health_check table
  • Get the bootstrapping process to output NULL values.
  • fix a quoting issue in the bootstrap code, thanks @mylesjao.

v1.5.0: "someone, somewhere, is still smoking cigarettes, damnit"

  • CHANGE: Kafka producer no longer ships with hard-coded defaults. Please ensure you have "compression.type", "metadata.fetch.timeout.ms", and "retries" configured to your liking.
  • bugfix: fix a regression in handling ALTER TABLE change c int after b statements
  • warn on servers with missing server_id

v1.4.2: "drawer cat is back"

  • kafka 0.10.0 support, as well as a re-working of the --kafka_version command line option.

v1.4.1: "cat snores"

  • support per-table topics, Thanks @smferguson and @sschatts.
  • fix a parser issue with DROP COLUMN CASCADE, thanks @smferguson

v1.4.0: "deep, insomniac character flaws"

1.4.0 brings us two nice new features:

  • partition-by-column: see --kafka_partition_columns. Thanks @smferguson
  • output schema changes as JSON: see --output_ddl. Thanks @xmlking
  • As well as a fix around race conditions on shutdown.

v1.3.0: "yogg-saron"

  • support for fractional DATETIME, TIME, TIMESTAMP columns, thanks @Dagnan
  • support for outputting server_id & thread_id, thanks @sagiba
  • fix a race condition in bootstrap support

v1.2.2: "bats wearing frog pajamas"

  • Maxwell will now include by default fields with NULL values (as null fields). To disable this and restore the old functionality where fields were omitted, pass --output_nulls=false
  • Fix an issue with multi-client support where two replicators would ping-pong heartbeats at each other
  • Fix an issue where a client would attempt to recover a position from a mismatched client_id
  • Fix a bug when using CHANGE COLUMN on a primary key

v1.2.1: "point-ones are a sad and inevitable fact"

This is a bugfix release.

  • fix a parser bug around ALTER TABLE CHARACTER SET
  • fix bin/maxwell to pull in the proper version of the kafka-clients library

v1.2.0: "just here, not to talk to you"

1.2.0 is a major release of Maxwell that introduces master recovery features; when a slave is promoted to master, Maxwell is now capable of recovering the position. See the --master_recovery flag for more details.

It also upgrades the kafka producer library to 0.9. If you're using maxwell with a kafka 0.8 server, you must now pass the --kafka0.8 flag to maxwell.

v1.1.6: "pithy"

  • minor bugfix in which maxwell with --replay mode was trying to write heartbeats

v1.1.5: "my brain is a polluted mess"

  • @dadah89 adds --output_binlog_position to optionally output the position with the row
  • @dadah89 adds --output_commit_info to turn off xid/commit fields
  • maxwell now supports tables with partitions
  • maxwell now supports N maxwells per-server. see the client_id / replica_server_id options.
  • two parser fixes, for engine=innodb and CHARSET ASCII
  • lay the ground work for doing master recovery; we add a heartbeat into the positions table that we can co-ordinate around.

v1.1.4: "george flunk"

  • add support for a bunch more charsets (gbk, big5, notably)
  • fix Maxwell's handling of kafka errors - previously we were trying to crash Maxwell by throwing a RuntimeException out of the Kafka Producer, but this was a failure. Now we log and skip all errors.

v1.1.3: "the button I push to not have to go out"

This is a bugfix release, which fixes:

  • zendesk#376, a problem parsing RENAME INDEX
  • zendesk#371, a problem with the SERIAL datatype
  • zendesk#362, we now preserve the original casing of columns
  • zendesk#373, we were incorrectly expecting heartbeats to work under 5.1

v1.1.2: "scribbled notes on red pages"

  • pick up latest mysql-connector-j, fixes #369
  • fix an issue where maxwell could skip ahead positions if a leader failed.
  • rework buffering code to be much kinder to the GC and JVM heap in case of very large transactions / rows inside transactions
  • kinder, gentler help text when you specify an option incorrectly

v1.1.1: scribbled notes on blue pages

  • fixes a race condition setting the binlog position that would get maxwell stuck

v1.1.0: "sleep away the afternoon"

  • much more efficient processing of schema updates storage, especially when dealing with large schemas.
  • @lileeyao added --exclude-columns and the --jdbc_options features
  • @lileeyao added --jdbc_options
  • can now blacklist entire databases
  • new kafka key format available, using a JSON array instead of an object
  • bugfix: unsigned integer columns were captured incorrectly. 1.1 will recapture the schema and attempt to correct the error.

v1.1.0-pre4: "buck buck buck buck buck buck-AH!"

  • Eddie McLean gives some helpful patches around bootstrapping
  • Bugfixes for the patch-up-the-schema code around unsigned ints
  • forgot to include some updates that back-patch unsigned column problems

v1.1.0-pre2: "yawn yawn"

  • fix performance issues when capturing schema in AWS Aurora
  • fix a bug in capturing unsigned integer columns

v1.0.1: "bag of oversized daisies"

  • fixes a parsing bug with CURRENT_TIMESTAMP()

v1.0.0: "Maxwell learns to speak"

Since v0.17.0, Maxwell has gotten:

  • bootstrapping support
  • blacklisting for tables
  • flexible kafka partitioning
  • replication heartbeats
  • GEOMETRY columns
  • a whole lotta lotta bugfixes

and I, Osheroff, think the damn thing is stable enough for a 1.0. So there.

v1.0.0-RC3: "C'mon and take it"

pull in support for replication heartbeats. helps in the flakier network environs.

v1.0.0-RC2: "same thing, just without the v"

  • fixes the way ALTER DATABASE charset= was handled
  • adds proper handling of ALTER TABLE CONVERT TO CHARSET

v1.0.0-RC1: "Richard Buckner's release"

  • modifications to the way the bootstrap utility works
  • fix a race condition crash bug in bootstrapping
  • fix a parser bug

v1.0.0-PRE2: "an embarassment of riches"

1.0.0-PRE2 brings in a lot of changes that got merged while we were testing out PRE1. so, hey.

  • Configurable names for the maxwell schema database (Kristian Kaufman)
  • Configurable key (primary key, id, database) into the kafka partition hash function (Kristian Kaufman)
  • Configurable Kafka partition hash function (java hashCode, murmur3) (Kristian Kaufman)
  • support GEOMETRY columns, output as well-known-text
  • add --blacklist_tables option to fully ignore excessive schema changes (Nicolas Maquet)
  • bootstrap rows now have 'bootstrap-insert' type

v1.0.0-PRE1: "drunk conversations with sober people"

  • Here we have the preview release of @nmaquet's excellent work around bootstrapping initial versions of mysql tables.

v0.17.0: "wrists of William"

v0.17 is a large bugfix release with one new feature.

  • FEATURE: allow specifying an alternative mysql schema-storage server and replication server
  • BUGFIX: properly handle case-sensitivity by aping the behavior of the master server. Fixes #230.
  • BUGFIX: parse some forms of CHECK( ... ) statements. Fixes #203.
  • BUGFIX: many more SQL-parser fixes. We are mostly through some thousands of lines of SQL produced by mysql-test.

v0.16.2: "The best laid plans"

This is a large-ish bugfix release.

  • Support, with reservations, binlog_row_image=MINIMAL
  • parser bug: handle table names that look like floating points
  • parser bug: fix for entity names that have '.', '', etc in them
  • handle UPPERCASE encoding names
  • support UCS2 (start trying to operate ok on the mysql-test suite)
  • use ObjectOutputStream.reset to fix memory leaks when buffering to disk

v0.16.1: "me and room service"

This is a bug-fix-roundup release:

  • support ALTER DATABASE
  • fix a bunch of parse errors: we've started running mysql-test at maxwell and are fixing up failures.
  • some modifications to the overflow-to-disk logic; we buffer the input and output, and we fix a memory leak

v0.16.0: "Kristian Kaufmann's version"

Version 0.16.0 introduces a feature where UPDATE statements will now show both the new row image and the old values of the fields that changed. Thanks @kristiankaufmann

v0.15.0: "the littlest little city"

  • fix a parse problem with indices ordered by ASC/DESC

v0.15.0-RC1: "it's later than you think"

  • large transactions now buffer to disk instead of crushing maxwell.
  • support ALGORITHM=[algo], LOCK=[lock] for 5.6 alters

v0.14.6: "It's about being American. Sort of."

  • fix TIME column support
  • fix parsing on millisecond precision column defintions
  • fix CREATE SCHEMA parsing

v0.14.5: "false is the new true"

  • handle BOOLEAN columns with true/false defaults

v0.14.4: "You'd think we'd be at 1.0 by now, wouldn't you?"

  • fixes parsing of "mysql comments" (/*! .. */)
  • More performance improvements, another 10% in a tight loop.

v0.14.3: "Peanuts. My girlfriend thinks about peanuts."

  • fixes a regression in 0.14.2 that creates duplicate copies of the "mysql" database in the schema.

v0.14.2: "Maxwell Sandvik 88"

  • capture the mysql database along with the rest of the schema. Eliding it was a bad premature optimization that led to crashes when tables in the mysql database changed.

v0.14.1: "be liberal in what you accept. Even if nonsensical."

  • fixes a parser bug around named PRIMARY KEYs.

v0.14.0: "the slow but inevitable slide"

This release introduces row filters, allowing you to include or exclude tables from maxwell's output based on names or regular expressions.

v0.13.1: "well that was somewhat expected"

v0.13.1 is a bug fix of v0.13.0 -- fixes a bug where long rows were truncated.

v0.13.0 contains:

  • Big performance boost for maxwell: 75% faster in some benchmarks
  • @davidsheldon contributed some nice bug fixes around CREATE TABLE ... IF NOT EXISTS, which were previously generating new, bogus copies of the schema.
  • we now include a "scavenger thread" that will lazily clean out old, deleted schemas.

v0.13.0: "Malkovich Malkovich Malkovich Sheldon?"

Lucky release number 13 brings some reasonably big changes:

  • Big performance boost for maxwell: 75% faster in some benchmarks
  • @davidsheldon contributed some nice bug fixes around CREATE TABLE ... IF NOT EXISTS, which were previously generating new, bogus copies of the schema.
  • we now include a "scavenger thread" that will lazily clean out old, deleted schemas.

This release has a pretty bad bug. do not use.

v0.12.0: "what do I call them? Slippers? Why, are you jealous?"

  • add support for BIT columns.

v0.11.4: "13 steps"

this is another bugfix release that fixes a problem where the replication thread can die in the middle of processing a transaction event. I really need to fix this at a lower level, ie the open-replicator level.

v0.11.3: ".. and the other half is to take the bugs out"

this is a bugfix release:

  • fix problems with table creation options inside alter statements ( ALTER TABLE foo auto_increment=10 )
  • fix a host of shutdown-procedure bugs

the test suite should also be way more reliable, not like you care.

v0.11.2: "savage acts of unprovoked violence are bad"

This is a bugfix release. It includes:

  • soft deletions of maxwell.schemas to fix A->B->A master swapping without creating intense replication delay
  • detect and fail early if we see binlog_row_image=minimal
  • kill off maxwell if the position thread dies
  • fix a bug where maxwell could pick up a copy of schema from a different server_id (curse you operator precedence!)

v0.11.1: "dog snoring loudly"

  • maxwell gets a very minimal pass at detecting when a master has changed, in which it will kill off schemas and positions from a server_id that no longer is valid. this should prevent the worst of cases.

v0.11.0: "cat waving gently"

This release of Maxwell preserves transaction information in the kafka stream by adding a xid key in the JSON object, as well as a commit key for the final row inside the transaction.

It also contains a bugfix around server_id handling.

v0.10.1: "all 64 of your bases belong to... shut up, internet parrot."

  • proper support for BLOB, BINARY, VARBINARY columns (base 64 encoded)
  • fix a problem with the SQL parser where specifying encoding or collation in a string column in the wrong order would crash
  • make table option parsing more lenient

v0.11.0-RC1: "goin' faster than a rollercoaster"

  • merge master fixes

v0.10.0: "The first word is French"

  • Mysql 5.6 checksum support!
  • some more bugfixes with the SQL parser

v0.11.0-PRE4: "except for that other thing"

  • bugfix on v0.11.0-PRE3

v0.11.0-PRE3: "nothing like a good night's sleep"

  • handle SAVEPOINT within transactions
  • downgrade unhandled SQL to a warning

v0.11.0-PRE2: "you really need to name a PRE release something cutesy?"

  • fixes for myISAM "transactions"

v0.11.0-PRE1: "A slow traffic jam towards the void"

  • fix a server_id bug (was always 1 in maxwell.schemas)
  • JSON output now includes transaction IDs

v0.10.0-RC4: "Inspiring confidence"

  • deal with BINARY flag in string column creation.

v0.9.5: "Long story short, that's why I'm late"

  • handle the BINARY flag in column creation

v0.10.0-RC3: "Except for that one thing"

  • handle "TRUNCATE [TABLE_NAME]" statements

v0.10.0-RC2: "RC2 is always a good sign."

  • fixes a bug with checksum processing.

v0.10.0-RC1: "verify all the things"

  • upgrade to open-replicator 1.3.0-RC1, which brings binlog checksum (and thus easy 5.6.1) support to maxwell.

v0.9.4: "we've been here before"

  • allow a configurable number (including unlimited) of schemas to be stored

v0.9.3: "some days it's just better to stay in bed"

  • bump open-replicator to 1.2.3, which allows processing of single rows greater than 2^24 bytes

v0.9.2: "Cat's tongue"

  • bump open-replicator buffer to 50mb by default
  • log to STDERR, not STDOUT
  • --output_file option for file producer

v0.9.1: "bugs, bugs, bugs, lies, statistics"

  • Maxwell is now aware that column names are case-insenstive
  • fix a nasty bug in which maxwell would store the wrong position after it lost its connection to the master.

v0.9.0: Vanchi says "eat"

Also, vanchi is so paranoid he's worried immediately about this.

  • mysql 5.6 support (without checksum support, yet)
  • fix a bunch of miscellaneous bugs @akshayi1 found (REAL, BOOL, BOOLEAN types, TRUNCATE TABLE)

v0.8.1: "Pascal says Bonjour"

  • minor bugfix release around mysql connections going away.

v0.8.0: the cat never shuts up

  • add "ts" field to row output
  • add --config option for passing a different config file
  • support int1, int2, int4, int8 columns

v0.7.2: "all the sql ladies"

  • handle inline sql comments
  • ignore more user management SQL

v0.7.1: "not hoarders"

  • only keep 5 most recent schemas

v0.7.0: 0.7.0, "alameda"

  • handle CURRENT_TIMESTAMP parsing properly
  • better binlog position sync behavior

v0.6.3: 0.6.3

  • better blacklist for CREATE TRIGGER

v0.6.2: v0.6.2

  • maxwell now ignores SAVEPOINT statements.

v0.6.1: v0.6.1

  • fixes a bug with parsing length-limited indexes.

v0.6.0: kafkakafkakafa

Version 0.6.0 has Maxwell outputting a JSON kafka key, so that one can use Kafka's neat "store the last copy of a key" retention policy. It also fixes a couple of bugs in the query parsing path.

v0.5.0: 0.5.0 -- "People who put commas in column names deserve undefined behavior"

  • maxwell now captures primary keys on tables. We'll use this to form kafka key names later.
  • maxwell now outputs to a single topic, hashing the data by database name to keep a database's updates in order.

v0.4.0: 0.4.0, "unboxed cat"

v0.4.0 fixes some bugs with long-lived mysql connections by adding connection pooling support.

v0.3.0: 0.3.0

This version fixes a fairly nasty bug in which the binlog-position flush thread was sharing a connection with the rest of the system, leading to crashes.

It also enables kafka gzip compression by default.

v0.2.2: 0.2.2

Version 0.2.2 sets up the LANG environment variable, which fixes a bug in utf-8 handling.

v0.2.1: v0.2.1

version 0.2.1 makes Maxwell ignore CREATE INDEX ddl statements and others.

v0.2.0: 0.2.0

This release gets Maxwell storing the last-written binlog position inside the mysql master itself.

v0.1.4: 0.1.4

support --position_file param

v0.1.3: 0.1.3

Adds kafka command line options.

v0.1.1: 0.1.1

v0.1.1, a small bugfix release.

v0.1: 0.1

This is the first possible release of Maxwell that might work. It includes some exceedingly basic kafka support, and JSON output of binlog deltas.