[
https://issues.apache.org/jira/browse/HDFS-10195?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15209434#comment-15209434
]
Jing Zhao commented on HDFS-10195:
----------------------------------
Thanks a lot for the great work, [~anu]. Besides Chris's comments, I have some
further comments. Please let me know if they make sense to you.
# My current feeling is that the {{init}}/{{shutdown}}/{{getLock}} do not need
to be APIs in ContainerManager. They are more related to the ContainerManager
implementation instead of container operations.
# In ContainerManagerImpl, maybe we do not need to use another two class
attributes for readLock and writeLock. How about letting ContainerManagerImpl
implement the RwLock interface?
# IIUC the patch creates a leveldb instance for each container's metadata? Can
we use one single leveldb instance for all the container's metadata, to avoid
loading many db instances at the same time and/or the performance penalty when
loading a new db store. We can assign individual prefix for each container's
metadata keys to separate containers in the db.
# It will be helpful if we have javadoc explaining the content and format of
each file (data file, metadata file, and metadata db file etc.).
# We also need to handle the case that the metadata file is missing. In that
case {{readContainerInfo}} may throw a NPE which cannot be caught by the
current catch section.
> Ozone: Add container persistence
> --------------------------------
>
> Key: HDFS-10195
> URL: https://issues.apache.org/jira/browse/HDFS-10195
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: ozone
> Affects Versions: HDFS-7240
> Reporter: Anu Engineer
> Assignee: Anu Engineer
> Fix For: HDFS-7240
>
> Attachments: HDFS-10195-HDFS-7240.001.patch
>
>
> Adds file based persistence for containers.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
