Skip to content

not in lock elapsed time(ms)={}, bodyLength #3350

@Cczzzz

Description

@Cczzzz

在服务端总有莫名其妙的超时情况发生:
2021-09-13 01:44:35 WARN SendMessageThread_14 - putMessage not in lock elapsed time(ms)=980, bodyLength=22
2021-09-13 01:44:35 WARN SendMessageThread_1 - putMessage not in lock elapsed time(ms)=981, bodyLength=6864
2021-09-13 01:44:35 WARN SendMessageThread_9 - putMessage not in lock elapsed time(ms)=982, bodyLength=6786

但是并没有[NOTIFYME]putMessage in lock cost time(ms) ,说明没有写入page cache 耗时很高的情况。可能是主从同步的问题?
场景:

tps <1k
4.7.0 on dledger
16c 32g
大概3天出现一次

今天发现 出现异常时有磁盘io读取和page in 的峰值,还有出现 meory page faults 的峰值。
这因为拉取历史消息发生缺页然后需要加载整个page cache 到内存中吗,这个行为会堵塞其他的写或者读吗。
日志中可以看见
Offset not matched. Request offset: 447420757556, firstOffset: 841813590016, lastOffset: 920196743168, mappedFileSize: 1073741824, mappedFiles count: 73
说明是一直有人缺拉取历史数据的。
读取历史数据会很严重的影响写入吗

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions