Tengfei Huang created SPARK-42582: ------------------------------------- Summary: Persisted RDD blocks can be inconsistent if the RDD computation is indeterminate Key: SPARK-42582 URL: https://issues.apache.org/jira/browse/SPARK-42582 Project: Spark Issue Type: Improvement Components: Spark Core Affects Versions: 3.3.2 Reporter: Tengfei Huang
When a rdd includes indeterminate operations, the rdd results can be different each time we recompute it. And when we cache such a rdd, we may have multiple rdd block replicas having different data. Here is an example: 1. Task A generated the rdd block rdd_1_1 on executor E1; 2. Task B on executor E2 tried to fetch remote rdd_1_1 from E1 but failed, then it will compute and cache another block on E2; If the results on E1 and E2 are differnet, we'll have 2 blocks for the same rdd partition with different data. The behavior will be unexpcted for such cases. -- This message was sent by Atlassian Jira (v8.20.10#820010) --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org