hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Manoj Govindassamy (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (HDFS-12191) Provide option to not capture the accessTime change of a file to snapshot if no other modification has been done
Date Thu, 24 Aug 2017 22:07:00 GMT

    [ https://issues.apache.org/jira/browse/HDFS-12191?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16140691#comment-16140691
] 

Manoj Govindassamy edited comment on HDFS-12191 at 8/24/17 10:06 PM:
---------------------------------------------------------------------

Thanks for working on the patch revision [~yzhangal]. Few comments below:

1. {{INode}}
{noformat}
public abstract class INode {
.. 
   private static boolean dontCaptureAccessTimeOnlyChangeInSnapshot = false;
   
   public static void setDontCaptureAccessTimeOnlyChangeInSnapshot(boolean s) {
       LOG.info("Setting dontCaptureAccessTimeOnlyChangeInSnapshot to " + s);
       dontCaptureAccessTimeOnlyChangeInSnapshot = s;
   }

   public static boolean getDontCaptureAccessTimeOnlyChangeInSnapshot() {
       return dontCaptureAccessTimeOnlyChangeInSnapshot;
   }
{noformat}
* Abstract class INode doesn't look a right place for placing the snapshot logic. The callers
of {{INode#setAccessTime()}} can pass in the needed details to skip recording the modifications
* FSNamesystem#setAccessTimes() has all the needed details to make the decision whether to
record the accesstime changes in the snapshots or not. So, shall we pass in the details from
here?

2. Config 
* typos in {{DFS_NAMENODE_SNAPTHOT_SKIP_ACCESSTIME_ONLY_CHANG_DEFAULT}} 

3. 
{noformat}
<description>
4228	  <name>dfs.namenode.snapshot.skip.accesstime-only-change</name>
4229	  <value>false</value>
4231	    If accessTime of a file changed but there is no other modification
4232	    made to the file, the changed accesstime will not be captured in next
4233	    snapshot. However, if there is other modification made to the file,
4234	    the latest access time will be captured together with the modification
4235	    in next snapshot.
4236	  </description>
{noformat}
* How about something on the below lines?
"When enabled, for access time only modification operations, HDFS snapshots will skip capturing
pre modification copy of files/directories metadata. However, for all other metadata modification
operations, files/directories latest access time will be captured along with the other metadata."


was (Author: manojg):
Thanks for working on the patch revision [~yzhangal]. 

1. {{INode}}
{noformat}
public abstract class INode {
.. 
   private static boolean dontCaptureAccessTimeOnlyChangeInSnapshot = false;
   
   public static void setDontCaptureAccessTimeOnlyChangeInSnapshot(boolean s) {
       LOG.info("Setting dontCaptureAccessTimeOnlyChangeInSnapshot to " + s);
       dontCaptureAccessTimeOnlyChangeInSnapshot = s;
   }

   public static boolean getDontCaptureAccessTimeOnlyChangeInSnapshot() {
       return dontCaptureAccessTimeOnlyChangeInSnapshot;
   }
{noformat}
* Abstract class INode doesn't look a right place for placing the snapshot logic. The callers
of {{INode#setAccessTime()}} can pass in the needed details to skip recording the modifications
* FSNamesystem#setAccessTimes() has all the needed details to make the decision whether to
record the accesstime changes in the snapshots or not. So, shall we pass in the details from
here?


> Provide option to not capture the accessTime change of a file to snapshot if no other
modification has been done
> ----------------------------------------------------------------------------------------------------------------
>
>                 Key: HDFS-12191
>                 URL: https://issues.apache.org/jira/browse/HDFS-12191
>             Project: Hadoop HDFS
>          Issue Type: Bug
>          Components: hdfs, namenode
>    Affects Versions: 3.0.0-beta1
>            Reporter: Yongjun Zhang
>            Assignee: Yongjun Zhang
>         Attachments: HDFS-12191.001.patch, HDFS-12191.002.patch
>
>
> Currently, if the accessTime of a file changed before a snapshot is taken, this accessTime
will be captured in the snapshot, even if there is no other modifications made to this file.
> Because of this, when we calculate snapshotDiff, more work need to be done for this file,
e,g,, metadataEquals method will be called, even if there is no modification is made (thus
not recorded to snapshotDiff). This can cause snapshotDiff to slow down quite a lot when there
are a lot of files to be examined.
> This jira is to provide an option to skip capturing accessTime only change to snapshot.
Thus snapshotDiff can be done faster.
> When accessTime of a file changed, if there is other modification to the file, the access
time will still be captured in snapshot.
> Sometimes we want accessTime be captured to snapshot, such that when restoring from the
snapshot, we know the accessTime of this snapshot. So this new feature is optional, and is
controlled by a config property.
> Worth to mention is, how accurately the acessTime is captured is dependent on the following
config that has default value of 1 hour, which means new access within an hour of previous
access will not be captured.
> {code}
> public static final String  DFS_NAMENODE_ACCESSTIME_PRECISION_KEY =
>       HdfsClientConfigKeys.DeprecatedKeys.DFS_NAMENODE_ACCESSTIME_PRECISION_KEY;
> public static final long    DFS_NAMENODE_ACCESSTIME_PRECISION_DEFAULT = 3600000;
> {code}
> .



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-help@hadoop.apache.org


Mime
View raw message