chejinge
chejinge
https://github.com/OpenAtomFoundation/pika/issues/2295
### Which PikiwiDB functionalities are relevant/related to the feature request? 如rsync线程等 flush 及 compaction线程等 ### Description 如title ### Proposed solution 如title ### Alternatives considered 如title
### Is this a regression? Yes ### Description  ### Please provide a link to a minimal reproduction of the bug _No response_ ### Screenshots or videos   ###...
### Is this a regression? Yes ### Description  ### Please provide a link to a minimal reproduction of the bug _No response_ ### Screenshots or videos  ### Please...
### Is this a regression? Yes ### Description  ### Please provide a link to a minimal reproduction of the bug _No response_ ### Screenshots or videos 空db flushdb用了2s的时间,这样肯定会影响主从复制的一致性测试 ###...
1.slave节点写引擎可能会触发rocksdb的限速,导致写入耗时长,可能超过了20s。因此就触发了master的synctimeout,master会将slave删除掉。master将slave删除掉之后,也会删除其对binlog文件的引用,所以会删除binlog。如果期间一直有写入,那么等到slave重新连master之后,会发现binlog文件被删除,就需要重新同步历史。(已修复) 2.slave节点在处理binlogsync请求时,是将请求放到了bgworker后台线程进行处理的。但是在写rocksdb时,会丢到另一个bgworker后台线程去执行。会阻塞多个线程。(已修复) 3.master节点的checksynctimeout逻辑会在距离LastSendTime 2秒时向slave发送心跳包,但条件是只有距离lastSendTime超过2秒并且slavenode的sent_offset == acked_offset,当出现1这种情况时,会导致心跳包也发不出去。(无须改动) 4.当前的主从复制,使用syncwindow限制了发送的速度,但是没有感知slave节点的压力。(无须改动) 5.slave重连master之后,重新初始化了binlogFileReader,但是没有初始化write_queue和sync——window,多线程并发执行时,可能导致sent_offset和acked_offset永远不一致,这样可能导致不一致。(已修复) 6. 主从复制老主降备手动slaveof的过程中会出现判断 sent_offset == acked_offset的逻辑,当出现不相等的时候,会导致主从slaveof ERROR的情况。(@yuecai)
本次主要支持rate_limit_bankwidow block_cache max_background_jobs
### Which PikiwiDB functionalities are relevant/related to the feature request? other ### Description cache中不存入大key ### Proposed solution cache中不存入大key key维度string超过多少不存cache:16K feild维度:cache-field-num-per-key 代码中已经实现,但是代码写的名字有问题 ### Alternatives considered zset-cache-field-num-per-key改为cache-field-num-per-key
### Which PikiwiDB functionalities are relevant/related to the feature request? _No response_ ### Description 参考咨询Alice.g ### Proposed solution 参考咨询Alice.g ### Alternatives considered 参考咨询Alice.g
### Which PikiwiDB functionalities are relevant/related to the feature request? _No response_ 目前的逻辑是: 如果是1个key 先走cache在走db,如果是多个key,直接走db ### Description cache中没有的标记下去查磁盘 ### Proposed solution pika mget多个key的时候支持查缓存和磁盘 ### Alternatives considered pika mget多个key的时候支持查缓存和磁盘