- Published on
Kafka
All Posts
- flink (27)
- kafka (13)
- ai-builders-digest (11)
- bigdata (6)
- yarn (5)
- dolphinscheduler (5)
- checkpoint (5)
- streampark (4)
- ckpt (4)
- connector (4)
- dofi (4)
- ai (4)
- savepoint (3)
- format (3)
- dinky (3)
- 运维 (3)
- milvus (3)
- transaction (2)
- rag (2)
- hdfs (2)
- json (2)
- udf (2)
- 大数据 (2)
- offset (2)
- oceanbase (2)
- exactly-once (2)
- ollama (2)
- cdh (1)
- hadoop (1)
- effectively-once (1)
- at-least-once (1)
- idempotency (1)
- exactlyonce (1)
- openapi (1)
- docker (1)
- agent (1)
- skill (1)
- dolphin (1)
- scheduler (1)
- shell (1)
- flinkcdc (1)
- mysql (1)
- state (1)
- stream-processing (1)
- fault-tolerance (1)
- jackson (1)
- timestamp (1)
- groupid (1)
- scanmode (1)
- startup (1)
- parameter (1)
- tablefunction (1)
- flinksql (1)
- dlq (1)
- dead-letter-queue (1)
- raw-format (1)
- richfunction (1)
- debug (1)
- gitops (1)
- kubernetes (1)
- kustomize (1)
- argocd (1)
- sops (1)
- velero (1)
- embedding (1)
- share-groups (1)
- ai-infrastructure (1)
- distributed-systems (1)
- model (1)
- vector (1)
- cloudcanal (1)
- starrocks (1)
- 实时数仓 (1)
- kappa (1)
- continue (1)
- idea (1)
- litellm (1)
- llm-gateway (1)
- local-inference (1)
- prometheus (1)
- slots (1)
- vcore (1)
- tasknum (1)
- parallelism (1)
- primarykey (1)
- infrastructure (1)
- defensive-programming (1)
- distributed-storage (1)
- qmd (1)
- personal-knowledge-base (1)
- vector-search (1)
- reranking (1)
- llm (1)
- local-deployment (1)
- zookeeper (1)
- resourcemanager (1)
- Published on
Kafka 一条消息里，外层是真数组、内层却是“字符串包 JSON 数组”。把内层当 ARRAY<ROW> 去 UNNEST 直接反序列化报错，靠一层 UNNEST + 一层 TableFunction 才解决。- Published on
Kafka 4.2 makes Share Groups production-ready, giving Kafka a queue-like consumption model while preserving a clear boundary with Flink stateful stream processing.- Published on
以 OceanBase、CloudCanal、Kafka、Flink SQL 和 StarRocks 为例，说明如何将采集、传输、计算和查询解耦，构建一条可扩展、可回放的实时数仓链路。- Published on
Flink SQL 没有内置通用的 DLQ 语法，采用 Raw Source 读取原始消息、SQL 层容错解析、双路分流到业务 Sink 和 DLQ Topic 的方案。