1.6 KiB
【M22项目】OLAP Flink Job may be stuck due to full G1 old GC.
| ID | Creation Date | Assignee | Status |
|---|---|---|---|
| OMPUB-1410 | 2024-08-13T16:59:05.000+0800 | 王成成 | 已关闭 |
现象描述
YGN NDC 出现OLAP Flink Job may be stuck due to full G1 old GC告警 |OLAP Flink Job may be stuck due to full G1 old GC.| |The NDC_YGN - session_record_processed_kafka_to_clickhouse05de2e85ce58b29ee49b5270ac3e5b42 triggers Old GC| |The NDC_YGN - session_record_processed_kafka_to_clickhouse1b026fc850105dd6120b2d405bab4671 triggers Old GC| |The NDC_YGN - session_record_processed_kafka_to_clickhouse65c76eadcb883c33af62426bd7166678 triggers Old GC| |The NDC_YGN - session_record_processed_kafka_to_clickhouse7159ebbdd4054f4c7503107eb4d0322f triggers Old GC| |The NDC_YGN - session_record_processed_kafka_to_clickhousebda1c07c47e30da4b48de8cd3c3e4fb0 triggers Old GC|wangchengcheng commented on 2024-08-16T16:29:54.704+0800:
问题原因：
写ck的列buffer会缓存共享，列buffer大小只会增加不会减少，列buffer大小的最大值为所有写入批次此列的最大值
多线程情况下，列buffer可能存在交换，某些字符串列比较大，随着时间的增加，会导致比较大的列buffer越来越多
wangchengcheng commented on 2024-08-16T16:30:17.693+0800:
已将groot-stream1.3.1版本更新至M22现场，任务运行72小时后未发现相关问题。
wangjunhao commented on 2024-08-29T11:35:45.872+0800:
M22现场更新后运行两周未发现相关问题