Home / Platform

A change data platform that checks its own work

Capture, initial load, apply, verification, high availability and operations — one agent, one repository, and the same guarantees for every engine.

01 · Capture

Read the log the database already writes

No triggers, no polling queries. Each engine has a native reader that follows its change log from a durable position.

Oracle

LogMiner or XStream. Single instance, RAC and ASM; non-CDB and multitenant with mining in CDB$ROOT and the catalog read in the PDB.

Tibero

Native redo reader, including per-thread capture on TAC clusters and archive catch-up that never skips to the current log.

PostgreSQL

Logical replication slots, PG 10 and later, with DDL captured through an event-trigger queue.

MySQL & MariaDB

Row-based binlog with GTID resume, for MySQL 5.7 and 8 and MariaDB 10 and 11.

SQL Server & Db2

SQL Server native CDC tables; Db2 LUW recovery logs and Db2 for i journal receivers.

Altibase, Kafka & files

Altibase Log Analyzer, Kafka consumer groups with JSON, Debezium or Avro payloads, and watched folders of CSV or JSONL.

02 · Initial load

Bring existing data across while changes keep flowing

Choose the read strategy per pipeline. Every mode checkpoints by primary key, so an interrupted load continues instead of starting again.

ModeHow it readsBest for
ConsistentOne read as of a single SCN, so every table reflects the same point in time.Most pipelines; exact cut-over point
Flashback-free chunksShort primary-key ranges of current data; the change stream restores consistency.Very large tables where long reads would exhaust undo
Parallel partitionsContiguous PK ranges copied concurrently, each with its own resume pointer.Big single-table copies with numeric keys
Capture-earlyCapture starts first; apply for each table waits until its snapshot completes.Zero-gap migrations on busy systems
On-demand reloadReload one table from the console, with or without truncating the target.Repairs and newly added tables
03 · Apply & delivery

Deliver to databases, streams, indexes and data lakes — at the same time

One agent can fan out to several targets of different types. Each target tracks its own progress, so a slow sink never holds back the others.

Relational targets

Oracle, Tibero, PostgreSQL, MySQL, MariaDB and Altibase, with identifier case and quoting handled per engine.

Kafka & Confluent

Keyed by primary key, JSON or Avro with Schema Registry, and transactional exactly-once delivery.

MongoDB & Elasticsearch

Upserts and deletes by key through bulk APIs, carrying changes only so documents are never overwritten with partial images.

Big data & data lake

Changes are staged per table, fsynced, and rolled into JSONL, Avro Object Container Files or Apache Parquet. Each rolled file can be uploaded to HDFS or S3-compatible storage, merged into Snowflake or another warehouse over JDBC, and registered as a Hive or Impala partition — the Cloudera data-lake pattern. Failed uploads stay marked and are retried, so a NameNode restart or an expired Kerberos ticket never strands a file.

target.type: FILE · format: JSONL | AVRO_OCF | PARQUET
04 · Verify & heal

Treat “no errors” as a claim to verify

Two independent services compare what the source holds with what the target received.

  • Reconciler — finds rows missing on the target and re-applies them
  • Drift validator — scheduled count and value checks with a loss tolerance you set
  • Loss counters — decode and capture gaps written to logs and metrics
  • Targeted heal — re-reads a table as soon as a record could not be decoded
Row blocks compared · source vs target
MatchedDrift foundHealed
Failover simulation
NODE Aneuroncdc-agentACTIVE
NODE Bneuroncdc-agentSTANDBY
05 · High availability

Fail over without applying anything twice

Agents coordinate through a lease in the repository. Only the lease holder may write, and every write carries the lease epoch.

  • Standby resumes from the last durable checkpoint
  • A returning node with a stale epoch is fenced out
  • Automatic or operator-confirmed takeover
  • Keeps its position through repository outages
06 · Operate

Tools for the people who run it at 3 a.m.

Neuron Manager

Web console for agents, lag, initial-load status, reloads, transactions, sinks and the audit trail.

Admin CLI

viriyaneuron_adm for status, recovery, reloads and transaction repair from any shell.

Monitoring & logs

Metrics endpoint plus concise operator log lines designed to be alerted on — progress, gaps and verdicts.

Connectors

Sources and targets

Mix any supported source with one or more targets. Filter by role.

Databases

OR
OracleSource + target · LogMiner, XStream
TB
TiberoSource + target · single and TAC
AL
AltibaseSource + target · Log Analyzer 7.1+
PostgreSQLSource + target · logical replication
MySQLSource + target · binlog, GTID
MariaDBSource + target · binlog
SQL
SQL ServerSource · 2016+ native CDC
DB2
Db2 LUWSource · 11.x recovery log
DB2
Db2 for iSource · IBM i 7.x journals

Streaming & NoSQL

Apache KafkaSource + target · JSON, Debezium, Avro
CF
Confluent Schema RegistryTarget · Avro, exactly-once
MongoDBTarget · bulkWrite by key
ElasticsearchTarget · _bulk index / delete

Big data & data lake

Apache ParquetTarget · columnar files per roll
Apache AvroTarget · Object Container Files
Apache HiveTarget · partition registration
IM
Apache ImpalaTarget · ADD PARTITION + REFRESH
Hadoop HDFSTarget · upload per rolled file
ClouderaTarget · Avro to HDFS + Impala
S3
Object storageTarget · S3-compatible upload command
SnowflakeTarget · COPY INTO + MERGE over JDBC
{ }
JSONL filesTarget · staged, fsynced, rolled

Files

CSV
Watched folderSource · CSV, JSONL, archive-on-commit

Product names and logos are trademarks of their respective owners and indicate compatibility only. Logo artwork from Simple Icons (CC0).

Proof of concept

Run it against your own source this week

We install the agent on your servers, load and replicate a schema you choose, and hand over the verification results.