LiveCompare v2

LiveCompare is designed to compare any number of databases to verify they are identical. The tool compares any number databases and generates a comparison report, a list of differences and handy DML scripts so the user can optionally apply the DML and fix the inconsistencies in any of the databases.

By default, the comparison set will include all tables in the database. LiveCompare allows checking of multiple tables concurrently (multiple worker processes) and is highly configurable to allow checking just a few tables or just a section of rows within a table.

Each database comparison is called a "comparison session". When the program starts for the first time, it will start a new session and start comparing table by table. In standalone mode, once all tables are compared, the program stops and generates all reports. LiveCompare can be stopped and started without losing context information, so it can be run at convenient times.

Each table comparison operation is called a "comparison round". If the table is too big, LiveCompare will split the table into multiple comparison rounds that will also be executed in parallel, alongside with other tables that are being carried on by other workers at the same time.

In standalone mode, the initial comparison round for a table starts from the beginning of the table (oldest existing PK) to the end of the table (newest existing PK). New rows inserted after the round started are ignored. LiveCompare will sort the PK columns in order to get min and max PK from each table. For each PK column which is unsortable, LiveCompare will cast it's content to string. In PostgreSQL that's achieved by using ::text and in Oracle by using to_char.

When executing the comparison algorithm, each worker requires N+1 database connections, being N the number of databases being compared. The extra required connection is to an output/reporting database, where the program cache is kept too, so the user is able to stop/resume a comparison session.

Any differences found by the comparison algorithm can be manually re-checked by the user at a later convenient time. This is recommended to be done to allow a replication consistency check. Upon the difference re-check, maybe replication caught up on that specific row and the difference does not exist anymore, so the difference is removed, otherwise it is marked as permanent.

At the end of the execution the program generates a DML script so the user can review it, and fix differences one by one, or simply apply the entire DML script so all permanent differences are fixed.

LiveCompare can be potentially used to ensure logical data integrity at row-level; for example, for these scenarios:

  • Database technology migration (Oracle x Postgres);
  • Server migration or upgrade (old server x new server);
  • Physical replication (primary x standby);
  • After failover incidents, for example to compare the new primary data against the old, isolated primary data;
  • In case of an unexpected split-brain situation after a failover. If the old primary was not properly fenced and the application wrote data into it, it is possible to use LiveCompare to know exactly which data is present in the old primary and is not present in the new primary. If desired, the DBA can use the DML script that LiveCompare generates to apply those data into the new primary;
  • Logical replication. Three kind of logical replication technologies are supported: Postgres native logical replication, pglogical and BDR.

Comparison Performance

LiveCompare has been optimized for use on production systems and has various parameters for tuning, described later. Comparison rounds are read-only workloads. An example use case compared 43,109,165 rows in 6 tables in 9m 17s with 4 connections and 4 workers, giving comparison performance of approximately 77k rows per second, or 1 billion rows in <4 hours.

The use case above can be considered a general use case. For low-load, testing, migration and other specific scenarios, it might be possible to improve speed by changing the data_fetch_mode setting to use server-side cursors. Each kind of server side cursors, in our experiments, provides an increase in performance on use cases involving either small or large tables.

Security Considerations for the User

For PostgreSQL 13 and older, LiveCompare requires an user that is able to read all data being compared. PostgreSQL 14 introduced a new role pg_read_all_data, which can be used for LiveCompare.

When logical_replication_mode = bdr, LiveCompare requires a user that has been granted the bdr_superuser role. When logical_replication_mode = pglogical, LiveCompare requires a user that has been granted the pglogical_superuser role.

To apply the DML scripts in BDR, then all divergent connections (potentially all data connections) require a user that has been granted the bdr_superuser in order to disable bdr.xact_replication.

If BDR is being used, LiveCompare will associate all fixed rows with a replication origin called bdr_local_only_origin. LiveCompare will also apply the DML with the transaction datetime far in the past, so if there are any BDR conflicts with real DML being executed on the database, LiveCompare DML always loses the conflict.

With the default setting of difference_fix_start_query, the transaction in apply scripts will change role to the owner of the table in order to prevent database users from gaining access to the role applying fixes by writing malicious triggers. As a result the user for the divergent connection needs to have ability to switch role to the table owner.