regions not release while drop database

[gaoqu...@gmail.com]

version : 1.0.8

when bechmark with sysbench oltp_update_index :10tables ,200w rows，aflter ten hours，the rigion count increase to 1400+.

then drop the database,the regions do not decreased,why?

and I also test master version,it can release the regions,but when the regions decrease or increase ,the performance will become so bad.

[yangz...@pingcap.com]

Thanks for your feedback. 

This is related with https://github.com/pingcap/tidb/issues/5915

[Arthur]

"Drop database, the regions do not decreased", this is because we don't support region merge in current multi-raft implementation.

To merge the old regions, we have to implement it in raft, it's more difficult and involves many corner cases.

The good news is, this feature is under development and maybe release after 1.1 RC.

Please keep an eye on our canonical news.

"when the regions decrease or increase ,the performance will become so bad."

You mean normal region split, or is there something wrong with TiKV machine?

在 2018年3月8日星期四 UTC+8下午12:43:28，gaoqu...@gmail.com写道：

[quan gao]

reply:

"when the regions decrease or increase ,the performance will become so bad."

You mean normal region split, or is there something wrong with TiKV machine?

tikv has many errors, such as:

region missing

can not’t find snaffle

get snapshot failed for kids=[] pear is not leader

中文（英文太差，猜测您是中国人）：

由于后来装了新版本，所以master版本的日志信息忘记留存，以上是昨天查询日志大概记得的错误信息

监控5秒刷新一次，从监控看region的增长与减少是非常急剧的，几十上百个变化

上面的错误发生时看，基本发生在storeid变更时的极短时间内

--
You received this message because you are subscribed to a topic in the Google Groups "TiDB user group" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/tidb-user/0NTA7udNYag/unsubscribe.
To unsubscribe from this group and all its topics, send an email to tidb-user+...@googlegroups.com.
To post to this group, send email to tidb...@googlegroups.com.
Visit this group at https://groups.google.com/group/tidb-user.
To view this discussion on the web visit https://groups.google.com/d/msgid/tidb-user/1d777f29-2390-465c-96db-445818aedd66%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Arthur]

When you talk about "region", do you mean "Current Regions Count" (which is a curve in the grafana) ? 

"Current Regions Count" is actually the raft peer count, one region typically have 3 copy, i.e. 3 raft peer.

PD will schedule for hot region (in case so many request send to the same region to make it a bottleneck),

that why you see its value change (decrease).

"Number of Regions" is different from "Current Regions Count", it's the total region count.

Because we haven't implement region merge yet, **it should never decrease**.

> "从监控看region的增长与减少是非常急剧的，几十上百个变化" 

There are two possible reasons for this:

1. Caused by pd schedule

2. Some TiKV instances are down.

The former is expected, while the later maybe a problem.

Could you provide more details for us.

What's the cluster information?

How did you do the test? Are you loading data?

Is there a grafana screen shot available?

Thanks!

在 2018年3月8日星期四 UTC+8下午4:13:12，quan gao写道：