>>> Ozone 0.5.0
Ozone 0.5.0 voting ended and passed. Dinesh will prepare the rest of the
release. Will be published soon.
>>> 1 billion key test (Bharat)
* 9 node (AWS) / 3 pipelines
* SSD: Raftlog, OM, SCM. HDDS: chunk data
* 5 client * 50 threads --> ~ 2+ days
* consistent 5 keys / sec write
* 10kb keys
Jitendra: good to see that non of OM or SCM are stressed
can be interesting to check details of RPC latency
data distribution between volumes
More details will be published as a blog post
>>> some ideas for the future (advanced viewfs) (Jitendra)
* A potential use-case is for adopting Ozone is to use HDFS but move
some of the data (especially the dirs with small files)
* Limitation of today's viewfs
* requires schema change, not transparent
* mount table should be managed manually
Would be great to solve them.
Router based federation is server-side but it's more heavy-weight,
use-case as we wouldn't like to proxy the data-path.
>>> Discussion about the new proposal to simplify s3 bucket name ->
ozone vol/bucket name (Marton)
1. Use only one specific volume for s3 buckets
2. expose any bucket with bind-mount it to the specific volume
Xiaoyu: using one volume can cause problem with the volume level locking
Summary: we need more discussion. Let's continue it on the mailing list.
Marton
---------------------------------------------------------------------
To unsubscribe, e-mail: ozone-dev-unsubscr...@hadoop.apache.org
For additional commands, e-mail: ozone-dev-h...@hadoop.apache.org
