hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Wei-Chiu Chuang <weic...@cloudera.com.INVALID>
Subject Re: Does Hedged-read usually happen in real workloads?
Date Mon, 19 Aug 2019 21:50:11 GMT
On Mon, Aug 19, 2019 at 5:30 AM Daegyu Han <hdg9400@gmail.com> wrote:

> Hi all,
>
> To my understand, Hedged-Read is a technique for copying IO to a replica
> when pread () is delayed to mitigate IO latency such as speculative
> execution in MapReduce.
>
> - Does Hedged-read really happen where it uses hbase databases?
>
There was a pretty good talk at HBaseCon 2018
https://www.youtube.com/watch?v=l6S-Vbs9WsU&list=PLdOhKvENNn6WH71iHXRNbEd1pTN2xY6P3&index=9
It talked about HBase Read Replica, but at the same time, also touched upon
hedged read.

Additionally, I am aware this is enabled for HBase and Impala among my
customers.
It's not clear to me what impact does it bring to Impala, but Impala is
quite sensitive to slow HDFS as well.

> - Does hedged-read often occur because of a bottleneck in HDFS or a
> bottleneck in the storage itself?
> - What is the best number to set the timeout and thread for Hedged-Read?
>
There are some configurations mentioned in the talk above. Check it out.

>
> Thank you
> ᐧ
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message