[
https://issues.apache.org/jira/browse/HDDS-14244?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ivan Andika updated HDDS-14244:
-------------------------------
Description:
One of our clusters consists of multiple OM services which point to a single
HDDS layer (e.g. SCM and DN). However, currently when we want to migrate bucket
between these OM services (same bucket name), we are using a simple sync tool
that simply download the key from the source OM service to the client and the
upload it to the new OM service although technically only OM DB keyTable
metadata and bucketTable metadata needs to be migrated since the block
locations remain the same.
We can support migrating only the OM metadata without any actual block data
transfer.
There are a few approach
# We can ingest the only the SST files from the old OM DB to the new OM DB.
This should be the most performant solution. However, I don't think SST files
between keys in different buckets are isolated, so we might need to so
something about it.
# Introduce a new API for metadata only migration. (e.g. createMetaKeys,
deleteMetaKeys, or even a new atomic migrate API). This might cause a lot of
network overheads.
This might also be used in the general OM sharding solution (splitting and
merging a shard) in the future.
We can also take a look at other distributed key value store solutions for
ideas (e.g. TiKV, CockroachDB, Yugabyte, FoundationDB, etc).
was:
One of our clusters consists of multiple OM services which point to a single
HDDS layer (e.g. SCM and DN). However, currently when we want to migrate bucket
between these OM services (same bucket name), we are using a simple sync tool
that simply download the key from the source OM service to the client and the
upload it to the new OM service although technically only OM DB keyTable
metadata and bucketTable metadata needs to be migrated since the block
locations remain the same.
We can support migrating only the OM metadata without any actual block data
transfer.
There are a few approach
# We can ingest the only the SST files from the old OM DB to the new OM DB.
This should be the most performant solution. However, I don't think SST files
between keys in different buckets are isolated, so we might need to so
something about it.
# Introduce a new API for metadata only migration. (e.g. createMetaKeys,
deleteMetaKeys). This might cause a lot of network overheads.
This might also be used in the general OM sharding solution (splitting and
merging a shard) in the future.
We can also take a look at other distributed key value store solutions for
ideas (e.g. TiKV, CockroachDB, Yugabyte, FoundationDB, etc).
> Support OM metadata-only bucket migration
> -----------------------------------------
>
> Key: HDDS-14244
> URL: https://issues.apache.org/jira/browse/HDDS-14244
> Project: Apache Ozone
> Issue Type: New Feature
> Reporter: Ivan Andika
> Assignee: Ivan Andika
> Priority: Major
>
> One of our clusters consists of multiple OM services which point to a single
> HDDS layer (e.g. SCM and DN). However, currently when we want to migrate
> bucket between these OM services (same bucket name), we are using a simple
> sync tool that simply download the key from the source OM service to the
> client and the upload it to the new OM service although technically only OM
> DB keyTable metadata and bucketTable metadata needs to be migrated since the
> block locations remain the same.
> We can support migrating only the OM metadata without any actual block data
> transfer.
> There are a few approach
> # We can ingest the only the SST files from the old OM DB to the new OM DB.
> This should be the most performant solution. However, I don't think SST files
> between keys in different buckets are isolated, so we might need to so
> something about it.
> # Introduce a new API for metadata only migration. (e.g. createMetaKeys,
> deleteMetaKeys, or even a new atomic migrate API). This might cause a lot of
> network overheads.
> This might also be used in the general OM sharding solution (splitting and
> merging a shard) in the future.
> We can also take a look at other distributed key value store solutions for
> ideas (e.g. TiKV, CockroachDB, Yugabyte, FoundationDB, etc).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
