DTX Conference 2025 in Birmingham focused on modern data engineering — from stream processing to lakehouse architectures. UUID v7 emerged as a cross-cutting theme: its timestamp prefix enables time-ordered storage in ClickHouse, Flink watermarking, and Kafka correlation without secondary timestamp columns.
Event details
| Conference | DTX Conference 2025 |
| Date | October 2–3, 2025 |
| Location | Birmingham, UK |
| Focus | UUID v7 in ClickHouse, Apache Flink, Kafka, and lakehouse architectures |
Key sessions
UUID v7 in ClickHouse analytics
-- ClickHouse: UUID v7 primary key with built-in function (ClickHouse 24.5+)
CREATE TABLE events (
id UUID DEFAULT generateUUIDv7(),
user_id UUID NOT NULL,
event_type String,
payload String,
PRIMARY KEY (id) -- UUID v7 sort order = chronological
) ENGINE = MergeTree()
ORDER BY id;
-- Time-range query using UUID v7 timestamp prefix
SELECT
id,
UUIDNumToString(id) AS id_str,
toDateTime(reinterpretAsUInt128(id) >> 80 / 1000) AS created_at,
event_type
FROM events
WHERE toDateTime(reinterpretAsUInt128(id) >> 80 / 1000) > now() - INTERVAL 1 HOUR
ORDER BY id
LIMIT 100;
Apache Flink with UUID v7 event IDs
// Flink: extract event time from UUID v7 message IDs
public class UuidV7TimestampAssigner implements AssignerWithPunctuatedWatermarks<KafkaEvent> {
@Override
public long extractTimestamp(KafkaEvent event, long previousTimestamp) {
// UUID v7 top 48 bits = Unix milliseconds
String uuidStr = event.getMessageId().replace("-", "");
return Long.parseUnsignedLong(uuidStr.substring(0, 12), 16);
}
@Override
public Watermark checkAndGetNextWatermark(KafkaEvent event, long extractedTimestamp) {
return new Watermark(extractedTimestamp - 1000); // 1 second lag
}
}
// Register with Flink stream
DataStream<KafkaEvent> stream = env
.addSource(kafkaConsumer)
.assignTimestampsAndWatermarks(new UuidV7TimestampAssigner());
Kafka correlation with UUID v7
// Producer: embed UUID v7 as Kafka message key for time-ordered partitioning
ProducerRecord<String, String> record = new ProducerRecord<>(
"orders",
UuidCreator.getTimeOrderedEpoch().toString(), // UUID v7 as message key
orderJson
);
// Consumer: extract event time from UUID v7 key
ConsumerRecord<String, String> consumed = ...; // from poll()
String msgKey = consumed.key(); // UUID v7
long eventMs = Long.parseUnsignedLong(msgKey.replace("-", "").substring(0, 12), 16);
Instant eventTime = Instant.ofEpochMilli(eventMs);
See UUID v7 in Kafka Event Streaming.
Delta Lake with UUID v7 primary keys
from delta.tables import DeltaTable
from pyspark.sql.functions import expr
# Write Delta Lake table ordered by UUID v7 primary key
df.write \
.format("delta") \
.option("delta.dataSkippingNumIndexedCols", "1") \
.sortBy("id") \ # UUID v7 — chronological sort order
.save("/delta/events")
# Z-ordering on UUID v7 improves time-range scan performance
DeltaTable.forPath(spark, "/delta/events") \
.optimize() \
.executeZOrderBy("id")
Conference takeaways
- ClickHouse —
generateUUIDv7()built-in from version 24.5; time-range queries use the timestamp prefix - Apache Flink — UUID v7 event IDs replace explicit timestamp fields for watermark extraction
- Kafka — UUID v7 message keys give deterministic partition ordering within a key
- Delta Lake — UUID v7 sort order improves Z-ordering and file compaction efficiency
- Unified ID — one UUID v7 value serves as the primary key in the OLTP database, the Kafka message key, and the lakehouse partition sort key
Further reading
- UUID v7 in ClickHouse — Analytics Primary Keys
- UUID v7 in Apache Flink — Stream Processing IDs
- UUID v7 in Kafka Event Streaming
- UUID v7 in Apache Spark — Distributed Jobs
- UUID in Databases — Storage and Indexing
External references
Frequently asked questions
Why is UUID v7 preferred over UUID v4 for analytics databases?
UUID v7's 48-bit timestamp prefix enables time-range queries to prune storage ranges without a separate date partition column. In ClickHouse, ordering by UUID v7 primary key sorts rows chronologically — ORDER BY id replaces ORDER BY created_at. UUID v4 requires a secondary sort column for chronological queries. See UUID v7 in ClickHouse.
How does UUID v7 work in Apache Flink stream processing?
Flink processing-time watermarks can use UUID v7 event IDs to extract creation timestamps when event-time clocks are unavailable. UUID v7 event IDs embedded in Kafka messages carry their creation time in the first 12 hex digits. A Flink UDF can extract this to reconstruct event time: new Date(Long.parseLong(uuidStr.replaceAll("-", "").substring(0, 12), 16)). See UUID v7 in Apache Flink.
How does UUID v7 simplify data warehouse primary keys?
Data warehouses traditionally use surrogate keys (auto-increment integers) or natural keys (business IDs). UUID v7 combines both: it is globally unique (suitable as a surrogate key) and time-ordered (sortable like an auto-increment key). Lakehouse formats like Delta Lake and Iceberg benefit from UUID v7's sort order for Z-ordering and file compaction decisions.