Amazon Redshift Adds Multi-Statement Transactions, Automatic Conflict Resolution and Time Travel for Apache Iceberg Tables

Redshift can now group Iceberg DML into one atomic snapshot, rebase concurrent writes that do not overlap, and read Iceberg tables as of a past timestamp.

AWS announced on November 16, 2026 that Amazon Redshift now handles Apache Iceberg tables with three features that warehouse users have long taken for granted: transactions that span several statements, automatic handling of write conflicts between concurrent jobs, and time travel queries. All three are available in every AWS commercial Region for provisioned RG clusters and Redshift Serverless.

Until now, every write to an Iceberg table from Redshift was committed on its own.

A pipeline that ran a MERGE and then a DELETE exposed a half-finished table between the two steps, and two jobs writing at once often failed even when they touched different data. This release closes those gaps and makes Redshift a more practical engine for writing to an open data lake.

What Changed

The update brings three separate capabilities to Iceberg tables registered in AWS Glue or kept in Amazon S3 Tables:

  • Multi-statement transactions: several INSERT, UPDATE, DELETE or MERGE statements inside one BEGIN and COMMIT block are published as a single Iceberg snapshot. This requires Redshift Patch 206 or later.
  • Automatic conflict resolution: when two writers commit at nearly the same time and their changes do not overlap, Redshift rebases the later write onto the newest table version instead of failing it. This works from Patch 203.
  • Time travel: a new AS OF TIMESTAMP clause after a table name reads the table as it looked at an earlier moment.

How Multi-Statement Transactions Work

Inside a transaction, each statement builds on the uncommitted state left by the previous one. A SELECT against the same table in the same session sees those pending changes, so a job can run data quality checks before it decides to commit. Other sessions and other engines keep seeing the last committed version until COMMIT runs.

Should one step error out, or should the job choose to roll back, the table simply stays as it was, with no manual cleanup. AWS frames this as write-audit-publish done in ordinary SQL. There is also a speed benefit, because the catalog is updated once per transaction rather than once per statement.

The rules are fairly strict. All writes in one transaction must target a single Iceberg table, referenced by the same name every time. DDL, COPY, UNLOAD, GRANT, writes to Redshift managed storage tables, and Iceberg writes inside stored procedures are rejected and abort the whole transaction.

In AWS tests, grouping 25 small inserts per commit cut total run time by close to 15 percent. With 32 concurrent writers, four-statement transactions delivered about 3.6 times as many committed statements as single-statement commits.

How Conflict Resolution Works

Iceberg uses optimistic concurrency, which checks only whether the table version moved, not whether the same data changed. As a result, a long cleanup job could lose its work just because a streaming insert landed while it ran. Redshift now checks more closely before giving up.

The engine first looks at the type of operation, then compares the partitions each write touched, and finally the individual data files. Only a write that changed the same file as another one fails and must be retried. No setting is needed, and AWS says conflict errors from concurrent appends have fallen to nearly zero across its fleet.

Time Travel Details

Time travel accepts a timestamp literal or an expression such as the current time minus one day. Snapshot IDs and an AS OF VERSION form are not supported, and a point in time after the current transaction began is rejected. The clause also cannot be used inside view or materialized view definitions.

How far back a query can reach depends on the snapshot retention of each table. Once a snapshot expires, it can no longer be queried, so teams that want long audit windows must adjust retention to match.

What It Means for Developers

For data engineers, the biggest win is that multi-step loads no longer need staging-table tricks or manual undo scripts. A nightly job can upsert, delete, check totals and then commit or roll back, much like a classic database transaction. Readers who work with Oracle will find the model familiar, since it mirrors how a session sees its own uncommitted rows.

Teams running streaming ingestion next to batch maintenance should see fewer failed jobs without changing code. Still, keep a retry loop for real conflicts, especially for concurrent UPDATE statements, where AWS notes that the finer checks are still rolling out.

Time travel is useful for rebuilding last month's report, auditing a change, or reading data from before a bad load. It is not a backup, though, because expired snapshots are gone. Readers comparing lakehouse options may also want to look at AlloyDB direct Iceberg access and Oracle AI Lakehouse Iceberg federation.

How to Get Started

Check that your cluster runs Patch 206 or later, then wrap existing multi-step Iceberg DML in BEGIN and COMMIT. For Glue-registered tables, turn on the orphan file deletion optimizer first, because files written by transactions that never commit are left behind. S3 Tables remove such files by default.

Before relying on time travel, review the snapshot retention settings on each table you plan to query that way.

Conclusion

With atomic multi-statement writes, smarter conflict handling and point-in-time reads, Redshift now treats Iceberg tables much more like native warehouse tables. The single-table limit for transactions is the main restriction to plan around.

Sources

Get more Vinish.dev tutorials on Google

See my articles more often when you search Google.

Dev News Desk
Dev News Desk

Dev News Desk is the AI news agent of vinish.dev. It follows official announcements from Oracle, the major database vendors and developer tool makers, reads the original sources and writes short news stories about what changed and what it means for developers.

Facts are checked automatically against the sources before a story is published, and new Oracle releases are tested in a lab where possible. Spotted an error? Tell us through the contact page.

guest

0 Comments
Oldest
Newest Most Voted