Skip to main content
This page has details about each release of the following :
  • molt: and
  • replicator:
Cockroach Labs recommends using the latest available version of each tool. Refer to Installation.

​
Installation

To install MOLT, download the binary that matches your architecture and source database:
Operating SystemArchitecturePostgreSQL/MySQLOracle
WindowsAMD 64-bitDownloadN/A
ARM 64-bitDownloadN/A
LinuxAMD 64-bitDownloadDownload
ARM 64-bitDownloadN/A
MacAMD 64-bitDownloadDownload
ARM 64-bitDownloadDownload
The download package includes the following:
  • molt binary.
  • replicator binary.
  • Grafana dashboard JSON files for MOLT Fetch ( grafana_dashboard.json ) and Replicator ( replicator_grafana_dashboard.json ) metrics. Each bundled dashboard is compatible with its corresponding binary version.
    • Oracle downloads also include the Oracle-specific Replicator dashboard ( replicator_oracle_grafana_dashboard.json ).
For ease of use, keep both molt and replicator in your current working directory. To display the current version of each binary, run molt --version and replicator --version. molt is bundled with the latest replicator version available at the time of the MOLT release. This means that the MOLT download always contains the latest released version of . To verify that the molt and replicator versions match, run molt --version and replicator --version. For previous binaries, refer to the MOLT version manifest. MOLT Fetch is supported on Red Hat Enterprise Linux (RHEL) 9 and above.

​
Docker images

​
MOLT Fetch

Docker multi-platform images containing both the AMD and ARM molt and replicator binaries are available. To pull the latest image for PostgreSQL and MySQL:
To pull a specific version (for example, 1.1.3):
To pull the latest image for Oracle (note that only linux/amd64 is supported):

​
MOLT Replicator

Docker images for MOLT Replicator are also available as a standalone binary:
To pull a specific version (for example, v1.1.1):

​
Changelog

​
June 26, 2026

molt 1.4.2 is available:
  • Fixed a bug where MOLT Fetch and MOLT Verify, run against a MySQL source with set to anything other than .*, would silently filter out all tables, completing with zero rows migrated and a success (exit 0) status. The schema filter is now correctly reset for MySQL sources.
  • MOLT Verify now honors column_exclusion rules from a . Previously these rules were silently ignored, producing false missing column on target warnings for columns intentionally excluded during MOLT Fetch.
  • MOLT Fetch now handles empty (0-row) tables in export-only to import-only workflows. Previously, an empty table produced no output during export, causing a subsequent import-only run to fail with a no such file or directory error. Empty tables are now marked and skipped gracefully on import (from local filesystem, S3, GCP, and Azure).
  • MOLT Fetch now validates flag values at startup and reports all invalid flags at once, instead of failing one flag at a time across re-runs.
  • At the end of a fetch run, MOLT Fetch now prints the exact commands to drop the PostgreSQL logical replication slot and publication that it created on the source, since replication is complete and WAL no longer needs to be retained.
  • Reduced per-object and per-batch log volume at default (info) logging in MOLT Fetch and MOLT Verify for large migrations. Verbose per-table and per-batch progress is now logged at debug level.

​
May 28, 2026

molt 1.4.1 is available:
  • MOLT Fetch now logs structured telemetry events. Opt out with the --opt-out-telemetry flag.
  • MOLT Fetch no longer counts tables that cannot be migrated (no primary key or source/target collation mismatch) toward the fetch complete summary’s num_tables and tables fields. Such tables are now reported separately in a new skipped_unmigratable_tables field.

​
April 30, 2026

molt 1.4.0 is available:
  • MOLT Verify now supports PostgreSQL declarative partitioned tables.

​
March 26, 2026

molt 1.3.7 is available:
  • Fixed a bug in MOLT Fetch v1.3.3-v1.3.6 where a transient error when scanning rows from the source database could cause retried queries to resume from an incorrect position, resulting in rows missing from the target database. Migrations from PostgreSQL and CockroachDB databases may have been exposed to this issue. Fetch retry attempts are recorded in the MOLT Fetch logs, and MOLT Verify will have detected any data discrepancies introduced by this bug.

​
February 26, 2026

molt 1.3.6 is available:
  • Added support for selective data verification on PostgreSQL and MySQL sources, allowing users to compare a subset of data from both source and target databases based on filter predicates.
  • Fixed an issue where MOLT Fetch failed to copy data from PostgreSQL when numeric arrays contained negative values.

​
January 22, 2026

molt 1.3.5 is available:
  • Binary downloads now include version-compatible Grafana dashboards.

​
December 18, 2025

molt 1.3.4 is available:
  • Deprecated data-load-and-replication, replication-only, and failback modes in Fetch. Replication should now be performed separately with Replicator. Refer to the documentation.
  • Oracle-compatible MacOS builds are now available for arm64 and amd64. Refer to Installation.
  • Fixed a bug where the --filter-path flag for MOLT Fetch was hidden. --filter-path is now present for selective data movement in MOLT Fetch.
  • MOLT Fetch now logs a warning when --schema-filter is used with MySQL sources, which do not support schemas.
  • Fixed a character escaping bug where MOLT Verify was incorrectly reporting mismatching rows for JSON data in MySQL.
  • Added better edge case handling for cleaning up the target database after IMPORT runs so that a fresh connection is opened, which mitigates against stale or already broken long-lived connections.
  • Added better logging to handle the case where the recreation of a connection fails during a retry.
  • Removed fixup mode, live verify mode, and continuous verify mode from MOLT Verify, including the --continuous and --continuous-pause-between-runs flags.
  • Fixed a bug in retry logic to cleanly exit (without hanging) when the context is canceled and report the proper error.

​
October 24, 2025

molt 1.3.2 is available:
  • MOLT Fetch replication modes are deprecated in favor of a separate replication workflow using replicator. For details, refer to .
  • Added retry logic to the export phase for CockroachDB and PostgreSQL sources to handle transient errors while maintaining a consistent snapshot point. Not currently supported for Oracle or MySQL sources.
  • Added --export-retry-max-attempts and --export-retry-max-duration flags to control retry behavior for source query exports.

​
September 25, 2025

molt 1.3.2 is available.
  • MOLT Fetch replication modes will be deprecated in favor of a separate replication workflow in an upcoming release. This includes the data-load-and-replication, replication-only, and failback modes.
  • Fixed a bug with stats-based sharding where PostgreSQL queries were generated incorrectly when the first primary key column did not have sufficient unique values.
  • Fixed a bug in escape-password where passwords that start with a hyphen were not handled correctly. Users must now pass the --password flag when running escape-password. For example, molt escape-password --password 'a$52&'.
  • Added support for assume role authentication during with MOLT Fetch.
  • Added support to replicator for retrying unique constraint violations on the target database, which can be temporary in some cases.
  • Added exponential backoff to replicator for retryable errors when applying mutations to the target database. This reduces load on the target database and prevents exhausting retries prematurely. The new --retryInitialBackoff, --retryMaxBackoff, and --retryMultiplier control backoff behavior. The new --maxRetries flag configures the maximum number of retries. To retain the previous “immediate retry” behavior, set --retryMaxBackoff 1ns --retryInitialBackoff 1ns --retryMultiplier 1.
  • Added support to replicator for the source_lag_seconds, target_lag_seconds, apply_mutation_age_seconds, and source_commit_to_apply_lag_seconds metrics for replication from PostgreSQL and MySQL, and introduced histogram metrics source_lag_seconds_histogram and target_lag_seconds_histogram for replication from CockroachDB. source_lag_seconds measures the delay before data is ready to be processed by replicator, while target_lag_seconds measures the “end-to-end” delay until replicator has written data to the target. A steady increase in source_lag_seconds may indicate replicator cannot keep up with the source workload, while a steady increase in target_lag_seconds may indicate replicator cannot keep up with the source workload or that writes on the target database are bottlenecked.

​
August 21, 2025

molt 1.3.1 is available.
  • MOLT Fetch now supports of primary keys of any data type on PostgreSQL 11+ sources. This can be enabled with the flag.
  • Added the flag to allow data loads with planned downtime and no replication setup. The --pglogical-ignore-wal-check flag has been removed.
  • Added the --enableParallelApplies to enable parallel application of independent table groups during replication. By default, applies are synchronous. When enabled, this increases throughput at the cost of increased target pool and memory usage.
  • Improved cleanup logic for scheduled tasks to ensure progress reporting and prevent indefinite hangs.
  • Added parallelism gating to ensure the parallelism setting is smaller than the targetMaxPoolSize. This helps prevent a potential indefinite hang.
  • Added new metrics that track start and end times for progress reports ( core_progress_reports_started_count and core_progress_reports_ended_count ) and error reports ( core_error_reports_started_count and core_error_reports_ended_count ). These provide visibility into the core sequencer progress and help identify hangs in the applier and progress tracking pipeline.
  • Added new metrics target_apply_queue_utilization_percent and target_apply_queue_depth to track target apply queue utilization, indicating when the queue should be resized. These metrics apply only to PostgreSQL, Oracle, and CockroachDB sources.
  • Oracle sources now support configurable backpressure between the source and target applier. The new target_apply_queue_depth metric tracks queue depth.
  • Improved visibility into queue depth between each source frontend and the backend applier to the target. Updated logging to use lower log levels with clearer, less alarming messages.
  • Improved throughput for tables in the replication stream that have no dependencies on one another. This increases parallelism and minimizes blocking of transactions that are mutually independent.
  • The best effort window is now disabled by default to prevent unexpected mode switches that could cause consistency issues or stall replication due to failed target applies.
Bug fixes
  • Fixed a panic that could occur with ENUM types.

​
July 24, 2025

molt 1.3.0 is available.
  • Oracle is now supported on Fetch Docker images and standalone binaries.
  • Oracle is now supported on Replicator Docker images and standalone binaries.
  • Azure Blob Storage is now supported as an intermediate data store for Fetch.
  • Added a --skip-pk-check flag that lets you run the initial data load even when the source or target table is missing a primary key, or the keys do not match. When this flag is set, Fetch treats every table as keyless, disables sharding, and exports each table in a single batch, ignoring export-concurrency and row-batch-size. Querying the entire table at once may lead to high memory usage or long-running queries. Duplicate source rows are automatically removed during import when the target has primary key or unique constraints.
  • Replication now uses checkpoint polling by default. This is because polling (which periodically queries the staging table for updates) performs comparably to checkpoint streaming (which uses an internal changefeed from the staging table to broadcast real-time updates). To use checkpoint streaming, set --enableCheckpointStream. --disableCheckpointStream is deprecated and should be removed from replication commands.
  • Replaced the --copierChannel and --stageCopierChannelSize replication flags with a single --targetApplyQueueSize flag, which controls downstream apply throughput and memory usage. This feature applies only to CockroachDB and PostgreSQL ( pglogical ) sources.
  • Added support for compressed changefeed payloads, improving performance of changefeed-to-Replicator communication. This only affects failback workflows from CockroachDB v25.2 and later.
  • Improved apply-side performance by switching to a faster JSON decoder.
  • Improved batch-accumulation performance by removing unnecessary sorting.
Bug fixes
  • Fixed a bug where some error logs were not displayed when replicating to the target database.

​
June 19, 2025

molt 1.2.7 is available.
  • Updated the MOLT Grafana dashboard with the following timing metrics to better diagnose performance: Source Side Queries, Convert Source Side Queries to Datums, Writing Datums to Pipe, Preparing CSV files to be uploaded, Uploading CSV files to S3/GCP/local.
  • Upgraded the MOLT parser to support new syntax that is valid in CockroachDB v25.2.
  • Added more granular replication counter metrics to track data counts at each stage of the mutation pipeline, helping to diagnose data correctness issues.
Bug fixes
  • MOLT Fetch failback now reliably creates changefeeds with a sorted list of table names so that create changefeed operations can be properly deduplicated.
  • Fixed an issue where shard connections failed to recognize custom types (e.g., ENUM ) in primary keys during table migration. This occurred because the type map from the original pgx.Conn was not cloned. The type map is now properly cloned and attached to each shard connection.
  • Fixed a bug that could cause an integer overflow, which impacts retrieving the correct shards for exporting data.

​
May 22, 2025

molt 1.2.6 is available.
  • Fixed a bug in that occurred when was set to false (default). Previously, the COPY query could use incorrect column names in some cases during data transfer, causing errors. The query now uses the correct column names.
  • Fixed a bug in how origin messages were handled during replication from PostgreSQL sources. This allows replication to successfully continue.
  • ENUM types can now be replicated from MySQL 8.0 sources.

​
April 25, 2025

molt 1.2.5 is available.
  • During data export, MOLT Fetch now treats empty STRING values on source Oracle databases as NULL values on the target database. This is because Oracle does not differentiate between empty STRING and NULL values.

​
April 7, 2025

molt 1.2.4 is available.
  • MOLT Fetch now supports PostgreSQL 11.
  • MOLT Fetch failback to CockroachDB is now disallowed.
  • MOLT Verify can now compare tables that are named differently on the source and target schemas.
  • The molt logging date format is now period-delimited for Windows compatibility.
  • During replication, an index is now created on all tables by default, improving replication performance. Because index creation can cause the replication process to initialize more slowly, this behavior can be disabled using the --stageDisableCreateTableReaderIndex .
  • Added a failback metric that tracks the time to write a source commit to the staging schema for a given mutation.
  • Added a failback metric that tracks the time to write a source commit to the target database for a given mutation.

​
February 26, 2025

molt 1.2.3 is available.
  • MOLT Fetch users can now set and to values greater than 1 for MySQL sources.
  • MOLT Fetch now supports case-insensitive comparison of table and column names by default. Previously, case-sensitive comparisons could result in no matching table on target errors. To disable case-sensitive comparisons explicitly, set . If = is not included (e.g., --case-sensitive false ), this is interpreted as --case-sensitive (i.e., --case-sensitive=true ).

​
February 5, 2025

molt 1.2.2 is available.
  • Added an flag that is used to set the AWS_REGION query parameter explicitly in the .
  • Fixed the schema mode for cases where there are uppercase table or schema names.
  • Fixed an issue with unsigned BIGINT values overflowing in replication.
  • Added a --schemaRefresh that is used to configure the schema watcher refresh delay in the replication phase. Previously, the refresh delay was set to a constant value of 1 minute. Set the flag as follows: --replicator-flags "--schemaRefresh {value}".

​
December 13, 2024

molt 1.2.1 is available.
  • MOLT Fetch users now can use to specify a service account for assume role authentication to cloud storage. --assume-role must be used with --use-implicit-auth, or the flag will be ignored.
  • MySQL 5.7 and later are now supported with MOLT Fetch replication modes.
  • Fetch replication mode now defaults to a less verbose INFO logging level. To specify DEBUG logging, pass in the --replicator-flags '-v' setting, or --replicator-flags '-vv' for trace logging.
  • MySQL columns of type BIGINT UNSIGNED or SERIAL are now auto-mapped to type in CockroachDB. MySQL regular BIGINT types are mapped to type in CockroachDB.
  • The pglogical replication workflow was modified in order to enforce safer and simpler defaults for the , data-load-and-replication, and replication-only workflows for PostgreSQL sources. Fetch now ensures that the publication is created before the slot, and that replication-only defaults to using publications and slots created either in previous Fetch runs or manually.
  • Fixed scan iterator query ordering for BINARY and TEXT (of same collation) PKs so that they lead to the correct queries and ordering.
  • For a MySQL source in replication-only mode, the can now be used to resume streaming replication after being interrupted. Otherwise, the is used to start initial replication after a previous Fetch run in mode, or as an override to the current replication stream.

​
October 29, 2024

molt 1.2.0 is available.
  • Added failback mode to MOLT Fetch, which allows the user to replicate changes on CockroachDB back to the initial source database. Failback is supported for MySQL and PostgreSQL databases.
  • The , which specifies the address of the pprof endpoint, is now configurable. The default value is '127.0.0.1:3031'.
  • involving replication now state that MySQL 8.0 and later are supported for replication between MySQL and CockroachDB.
  • can now be moved to CockroachDB using .
  • Improved logging for the schema check phases under the trace logging level, which is set with .
  • Added a sample Grafana dashboard for monitoring MOLT tasks.
  • Fetch now logs the name of the staging database in the target CockroachDB cluster used to store metadata for .
  • String that use C on PostgreSQL can now be compared to the default en_US.utf8 on CockroachDB.
  • MOLT is now distributed under the Cockroach Labs Product License Agreement, which is bundled with the binary.

​
August 26, 2024

molt 1.1.7 is available.
  • When a is used to rename a table or map partitioned tables, a script in the format partitionTableScript.{timestamp}.ts is now automatically generated to ensure that works properly if enabled.

​
August 19, 2024

molt 1.1.6 is available.
  • Fixed a Fetch bug where was ignored when --table-filter and --schema-filter were set to the default ( '.*' ).

​
August 15, 2024

molt 1.1.5 is available.
  • Deprecated the --ongoing-replication flag in favor of --mode data-load-and-replication, using the new --mode flag. Users should replace all instances of --ongoing-replication with --mode data-load-and-replication.
  • Fetch can now be run in an export-only mode by specifying . This will export all the data in csv or csv.gz format to the specified cloud or local store.
  • Fetch can now be run in an import-only mode by specifying . This will load all data in the specified cloud or local store into the target CockroachDB database, effectively skipping the export data phase.
  • Strings for the --mode flag are now word-separated by hyphens instead of underscores. For example, replication-only instead of replication_only.

​
August 8, 2024

molt 1.1.4 is available.
  • Added a replication-only mode for Fetch that allows the user to run ongoing replication without schema creation or initial data load. This requires users to set --mode replication_only and --replicator-flags to specify the defaultGTIDSet ( MySQL ) or slotName ( PostgreSQL ).
  • Partitioned tables can now be mapped to renamed tables on the target database, using the Fetch .
  • Added a new --metrics-scrape-interval flag to allow users to specify their Prometheus scrape interval and apply a sleep at the end to allow for the final metrics to be scraped.
  • Previously, there was a mismatch between the errors logged in log lines and those recorded in the exceptions table when an IMPORT INTO or COPY FROM operation failed due to a non-PostgreSQL error. Now, all errors will lead to an exceptions table entry that allows the user to continue progress from a certain table’s file.
  • Fixed a bug that will allow Fetch to properly determine a GTID if there are multiple source_uuids.

​
July 31, 2024

molt 1.1.3 is available.
  • 'infinity'::timestamp values can now be moved with Fetch.
  • Fixed an issue where connections were not being closed immediately after sharding was completed. This could lead to errors if the was set to a low value.
  • Fetch users can now exclude specific tables from migration using the .

​
July 18, 2024

molt 1.1.2 is available.
  • Fetch users can now specify columns to exclude from table migrations in order to migrate a subset of their data. This is supported in the schema creation, export, import, and direct copy phases.
  • Fetch now automatically maps a partitioned table from a PostgreSQL source to the target CockroachDB schema.
  • Fetch now supports column exclusions and computed column mappings via a new .
  • The new Fetch flag specifies a JSON file for schema/table/column transformations, which has validation utilities built in.

​
July 10, 2024

molt 1.1.1 is available.
  • Fixed a bug that led to incorrect list continuation file behavior if a trailing slash was provided in .
  • Fixed a bug with extracting the filename from a failed import URL. Previously, an older filename was being used, which could result in duplicated data. Now, the filename that is used in import matches what is stored in the exceptions log table.
  • Added a flag that determines whether is used for cloud storage import URIs.

​
July 8, 2024

molt 1.1.0 is available.
  • Fixed a bug where integer values for JSONB / JSON types were not correctly handled when migrating from a PostgreSQL or CockroachDB source.
  • Optimized the logic for clearing continuation tokens. Now, the clearing option is only presented in the relevant modes and when there are tokens to clear.
  • CockroachDB sources now exclusively accept connections from admin or root users.
  • The statement timeout can now be adjusted during the export phase. This guards against hanging queries.

​
May 31, 2024

molt 1.0.0 is available.
  • Renamed the --table-splits flag to , which is more descriptive.
  • Increased the default value of to 1000. This leads to better performance on the target post-migration. Each individual import job will take longer, since more data is now imported in each batch, but the sum total of all jobs should take the same (or less) time.

​
May 29, 2024

molt 0.3.0 is available.
  • Added an flag, which configures the number of files to be imported in each IMPORT job.
  • In some cases on the previous version, binaries would not work due to how molt was being built. Updated the build method to use static linking, which creates binaries that should be more portable.
  • <> conversion is now allowed for Fetch and Verify. The bit array is first converted to UINT64. A resulting 1 or 0 is converted to true or false accordingly. If the UINT64 is another value, an error is emitted.

​
May 20, 2024

molt 0.2.1 is available.
  • MOLT tools now enforce secure connections to databases as a default. The --allow-tls-mode-disable flag allows users to override that behavior if secure access is not possible.
  • When using MySQL as a source, and are strictly set to 1.
  • Fixed a bug involving history retention for values.

​
May 3, 2024

molt 0.2.0 is available.
  • Fetch now supports CockroachDB .
  • Fetch now supports continuous replication for PostgreSQL and MySQL source databases via the flag. When Fetch finishes the initial data load phase, it will start the replicator process as a subprocess, which runs indefinitely until the user ends the process with a SIGTERM ( ctrl-c ).
  • Replicator flags for ( PostgreSQL and MySQL ) are now supported, allowing users to further configure the mode for their use case.
  • Added the flag, which enables custom type mapping for schema creation.
  • Fixed a bug where primary key positions could be missed when creating a schema with multiple primary keys.
  • Added a default mode for MySQL sources that ensures consistency and does not leverage parallelism. New text is displayed that alerts the user and links to documentation in cases where fetching from MySQL might not be consistent.
  • Logging for continuation tokens is now omitted when data export does not successfully complete.

​
April 10, 2024

molt 0.1.4 is available.
  • Fixed an inconsistency where inserting 'null' as a string into JSON / JSONB columns did not align with PostgreSQL’s case-sensitive null handling.

​
March 22, 2024

molt 0.1.2 is available.
  • Fixed a bug where running Verify with a PostgreSQL or CockroachDB source would cause an error when there are dropped columns on the table.

​
March 13, 2024

molt 0.1.1 is available.
  • An error is now displayed if the original column is a PostgreSQL nested array.
  • Added sharded data export functionality for faster performance.
  • Added support for the --table-handling 'drop-on-target-and-recreate' flag for MySQL sources, enabling automatic schema recreation.
  • Added support for ENUM data types in schema creation.
  • Deprecated the --s3-bucket and --gcp-bucket flags in favor of a single --bucket-path flag.

​
February 22, 2024

molt 0.1.0 is available.
  • Added implicit authentication for GCP sources.
  • Resolved a CSV write deadlock issue for large file exports.
  • Added new dashboard metrics: molt_verify_num_tables (categorizes tables as verified, missing, or extraneous ) and molt_verify_duration_ms (how long a Verify run took to complete, in milliseconds).
  • Previously, if a comparison between heterogeneous column types occurred, Verify would panic and stop execution. Now, Verify no longer panics, but records the error and reports it to logging, and also marks the row as having mismatching columns. The error message is logged at most once per shard to avoid log spewing. Additionally, introduced the concept of conditional success, which means that all comparable row columns were compared and succeeded with the same value. Upon conditional success, the user is instructed to check all warnings and verify the data.
  • Marked the --live and --continuous flags as dependencies for other sub-configuration flags. If a flag is set without its dependency, MOLT will indicate which dependency flag is not correctly set.
  • Added the RFC for MOLT Fetch, which explains the architectural decisions and considerations of how Fetch was built. The RFC also outlines limitiations of the current system and further improvements around usability, performance, and integration with replication tools.
  • Added logging annotations for data-level errors and information that surfaced during the Verify process, as well as summary status logs about the overall functioning of the job (e.g., 10000 truth rows seen, 5000 mismatches, etc.). These can be filtered so that exact tables and rows can be triaged in one place.
  • Changed the output for the running summary statistics so that they are now more easily indexable by log processing tools. Previously, it was more human-readable, but fields were not easily indexable out of the box. Now, there is a balance between readability and downstream processing.
  • Added tagging for summary and data logs so that users can filter the logs in a variety of log processing tools, or grep in the command line. This solves the pain point of all the logs going to the same location without any differentiation. Now, it is possible to view types of logs one at a time. Summary logs report aggregated results on either a task or step in a task (e.g., task took 10 minutes; 500 tables processed overall). Data logs report on individual or row-level operations and quantitative markers (e.g., 1000 rows completed so far; failed to verify item with PK 1000)
  • Marked the --s3-bucket, --gcp-bucket, and --local-path flags as mutually exclusive because they all represent different intermediary stores.
  • Adjusted settings for the number of rows and the size of data to be flushed so that only one rule is applied at any given time.
  • Fixed a bug where outputs containing special characters were not formatted correctly.
  • When running the import query for Fetch, the import statement is now printed if the logger is in debug mode.
  • Updated the MOLT README with an OS + Docker-based edge case with Fetch.
  • Logging for Fetch and Verify now defaults to structured JSON logging, which contains at least the message, level, and formatted time string. The --use-console-writer flag enables prettier console writing that has colored and human-readable log output. Structured logging is the default because it is easier to parse with log ingestion tools and is more efficient for the application to output.
  • Task/data logs can now be written to specified log files using --log-file. If the flag is left empty, the MOLT tool will only write to stdout, which is more efficient.

​
June 26, 2026

replicator 1.4.0 is available:
  • Replicator now serializes target transactions that touch foreign-key-related rows so that a parent and its child rows commit in order, preventing transient foreign key violation errors (SQLSTATE 23503) that could previously occur when parent and child rows were applied in separate sequencer rounds. This foreign-key-aware scheduling is enabled by default and can be disabled with the new flag.
  • Replicator now supports replicating schemas whose foreign key relationships form cycles. Such schemas were previously rejected.
  • The Oracle source connector now refuses to start against an Oracle major version that is not on the supported list (19c and 21c), failing fast at preflight with a clear error message and hint.
  • Oracle source database redo log parsing now uses a faster parser by default, falling back to ANTLR parsing when needed.
  • Fixed a bug where replicating a CockroachDB BOOLEAN column to a MySQL BIT(1) column would fail with a type coercion error.
  • Added an flag. By default, Replicator refuses to start against an Oracle major version outside of the supported list (19c, 21c). Unsupported versions are not supported by Cockroach Labs and may cause silent data corruption or permanent data loss in the worst case. This flag bypasses the default safety check and starts replication anyway. It is to be used at your own risk.
  • Added an flag (PostgreSQL source) that drops DELETEs for the named tables before buffering, to avoid memory pressure from large batch deletions already handled by target-side TTLs. Filtered deletions are surfaced via the pglogical_filtered_deletions_total metric.
  • Added a flag (default: 1000) to tune the buffer between the userscript source acceptor and the core sequencer’s copier. Tuning this value can help avoid replication bottlenecks when using userscripts.
  • Replicator now logs a warning when is used with a single-feed source (pglogical, mylogical, oraclelogminer), where the flag has no effect.
  • Fixed a bug that occurs when non-standard INT sizes are used for the CockroachDB cluster during failback replication.

​
May 28, 2026

replicator 1.3.4 is available:
  • Updated the Replicator log format to include the year in all log timestamps.

​
April 30, 2026

replicator 1.3.3 is available:
  • Fixed a bug where MySQL BLOB, BINARY, VARBINARY, and MEDIUMBLOB columns containing non-UTF-8 binary data were corrupted during replication to CockroachDB. Binary values are now correctly hex-encoded before transmission, preserving data integrity.
  • Replication from MySQL sources now correctly replicates NULL values in ENUM columns.

​
March 26, 2026

replicator 1.3.2 is available:
  • Oracle source and target lag values now use the LogMiner redo log generation timestamp for better accuracy. Note that clock skew may still affect these values.
  • replicator pglogical and replicator mylogical now respect the --flushSize and --flushPeriod flags passed in.

​
February 26, 2026

replicator 1.3.1 is available:
  • Added support for Oracle’s Large Object ( CLOB, NCLOB, BLOB, LONG ) and binary ( RAW, LONG RAW ) datatypes when failing back from CockroachDB to Oracle, within currently supported size limits.
  • Added support for failback replication with mismatched time types between CockroachDB and Oracle (for example, TIMESTAMPTZ to TIMESTAMP / DATE ), and support for Oracle timestamp replication regardless of NLS format settings.
  • Replication from MySQL sources now supports core sequencer and the --target-apply-queue-size flag with core sequencer metrics, leading to improved throughput for migrations from MySQL.
  • Added support for metrics snapshotting to aid in support investigations. This feature is off by default but can be configured to capture metrics dumps. Included export-metrics-snapshots.sh and import-metrics-snapshots.sh helper scripts to package metrics snapshots.
  • Added the --dataDir flag to configure the base data directory for metrics snapshotting.

​
January 22, 2026

replicator 1.3.0 is available:
  • Added Prometheus metrics for userscript execution (execution time, invocation counts, rows processed/filtered, errors) with schema, table, and function labels.
  • Binary downloads now include version-compatible Grafana dashboards.
  • Added userscript API v2 features.

​
December 18, 2025

replicator 1.2.0 is available:
  • Updated metric names and descriptions and updated the dashboard to reflect these changes. Major metric name updates are as follows:
    • mutations_{received,success,error}_count -> acceptor_mutations_total{type="received|success|error"}
    • source_lag_seconds_histogram -> core_source_lag_seconds
    • target_lag_seconds_histogram -> target_apply_transaction_lag_seconds
    • target_apply_queue_depth -> target_apply_queue_size
    • commit_to_stage_lag_seconds -> stage_commit_lag_seconds
    • apply_mutation_age_seconds -> target_apply_mutation_age_seconds
    • target_mutations_sent_to_apply_queue_count -> target_mutations_sent_to_apply_queue
    • checkpoint_proposed_going_backwards_errors -> checkpoint_proposed_going_backwards_errors_total
    • Oracle LogMiner metrics are now prefixed with oraclelogminer_ (including oraclecheckpoint metrics) For the rest of the metric name and description changes, run curl http://localhost:{port}/_/varz while running replicator with the --metricsAddr flag set.
  • RAW16 stored UUIDs on Oracle can now be replicated into CockroachDB UUID columns when doing Oracle Log Miner source replication.
  • Fixed a bug where Oracle timestamp parsing could use the wrong DST offset, resulting in incorrect UTC times for certain dates.
  • Fixed observability for redo log parsing buckets so that the correct buckets and minimum durations are shown. Previously, the default for Prometheus buckets were too large, leading to incorrect redo log parsing time.
  • Updated the Replicator sentinel table name to REPLICATOR_SENTINEL for Oracle LogMiner operation (Oracle source replication) to address a compatibility issue with JDBC drivers calling getIndexInfo on tables with leading underscores.
  • Added metrics for the number of mutations sent to the apply queue from the pglogical frontend.

​
October 24, 2025

replicator 1.1.3 is available:
  • Added commit_to_stage_lag_seconds Prometheus histogram metric to track the distribution of source CockroachDB to staged data times.
  • Added core_parallelism_utilization_percent gauge to track parallelism utilization and identify when the system is nearing parallelism limits, and should be sized up.
  • Added core_flush_count metric to track the number of flushed batches in the applier flow and the reason for each flush.