[jira] [Commented] (HDFS-8010) Erasure coding: extend UnderReplicatedBlocks to accurately handle striped blocks
[ https://issues.apache.org/jira/browse/HDFS-8010?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14512893#comment-14512893 ] Zhe Zhang commented on HDFS-8010: - Thanks for the reviews [~jingzhao] and [~rakeshr]! And sorry for getting to this late. I just submitted a path for the rename under HDFS-8255. Will update the patch for the {{getPriority}} part soon. Erasure coding: extend UnderReplicatedBlocks to accurately handle striped blocks Key: HDFS-8010 URL: https://issues.apache.org/jira/browse/HDFS-8010 Project: Hadoop HDFS Issue Type: Sub-task Reporter: Zhe Zhang Assignee: Zhe Zhang Attachments: HDFS-8010-000.patch This JIRA tracks efforts to accurately assess the _risk level_ of a striped block groups with missing blocks, when added to {{UnderReplicatedBlocks}} -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-8010) Erasure coding: extend UnderReplicatedBlocks to accurately handle striped blocks
[ https://issues.apache.org/jira/browse/HDFS-8010?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14493261#comment-14493261 ] Jing Zhao commented on HDFS-8010: - Thanks for working on this, Zhe! The patch looks good to me. Some minors: # Please see if you want to do the rename work in trunk. But keeping the change here is also fine to me. # The patch needs rebase. Besides, {{getPriority}} now needs to use ECSchema information instead of hard-coded numbers. # After using ECSchema information for priority computation, we may need to add several more unit tests (to cover cases with different schemas). # As suggested by [~rakeshr], we can move the conditions into two separate functions. Erasure coding: extend UnderReplicatedBlocks to accurately handle striped blocks Key: HDFS-8010 URL: https://issues.apache.org/jira/browse/HDFS-8010 Project: Hadoop HDFS Issue Type: Sub-task Reporter: Zhe Zhang Assignee: Zhe Zhang Attachments: HDFS-8010-000.patch This JIRA tracks efforts to accurately assess the _risk level_ of a striped block groups with missing blocks, when added to {{UnderReplicatedBlocks}} -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-8010) Erasure coding: extend UnderReplicatedBlocks to accurately handle striped blocks
[ https://issues.apache.org/jira/browse/HDFS-8010?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14388555#comment-14388555 ] Rakesh R commented on HDFS-8010: Minor mistake in my above comments, please read the suggested way as : {code} private boolean veryUnderReplicated(int curReplicas, int expectedReplicas, boolean isStriped) { if (!isStriped) { return (curReplicas * 3) expectedReplicas; } else { return curReplicas = HdfsConstants.NUM_DATA_BLOCKS + 2; } } private boolean highestPrioirty(int curReplicas, boolean isStriped) { if(!isStriped){ return curReplicas == 1; } else { return curReplicas == HdfsConstants.NUM_DATA_BLOCKS; } } {code} Erasure coding: extend UnderReplicatedBlocks to accurately handle striped blocks Key: HDFS-8010 URL: https://issues.apache.org/jira/browse/HDFS-8010 Project: Hadoop HDFS Issue Type: Sub-task Reporter: Zhe Zhang Assignee: Zhe Zhang Attachments: HDFS-8010-000.patch This JIRA tracks efforts to accurately assess the _risk level_ of a striped block groups with missing blocks, when added to {{UnderReplicatedBlocks}} -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-8010) Erasure coding: extend UnderReplicatedBlocks to accurately handle striped blocks
[ https://issues.apache.org/jira/browse/HDFS-8010?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14388552#comment-14388552 ] Rakesh R commented on HDFS-8010: Thanks [~zhz]. Overall the patch looks good. Just few comments, # Can we rename {{requiredReplication}} and {{expectedReplication}} variables as well. This should be helpful to reflect the meaning, right? # Can we think of simplifying the below conditions to make it more readable. {code} +} else if ((!isStriped curReplicas == 1) || +(isStriped curReplicas == HdfsConstants.NUM_DATA_BLOCKS)) { +} else if ((!isStriped (curReplicas * 3) expectedReplicas) || +isStriped curReplicas = HdfsConstants.NUM_DATA_BLOCKS + 2) { {code} One suggestion that comes in my mind is moving to another private methods like below, {code} private boolean veryUnderReplicated(int curReplicas, int expectedReplicas, boolean isStriped) { if (!isStriped) { return curReplicas = HdfsConstants.NUM_DATA_BLOCKS + 2; } else { return (curReplicas * 3) expectedReplicas; } } private boolean highestPrioirty(int curReplicas, boolean isStriped) { if(isStriped){ return curReplicas == 1; } else { return curReplicas == HdfsConstants.NUM_DATA_BLOCKS; } } {code} Erasure coding: extend UnderReplicatedBlocks to accurately handle striped blocks Key: HDFS-8010 URL: https://issues.apache.org/jira/browse/HDFS-8010 Project: Hadoop HDFS Issue Type: Sub-task Reporter: Zhe Zhang Assignee: Zhe Zhang Attachments: HDFS-8010-000.patch This JIRA tracks efforts to accurately assess the _risk level_ of a striped block groups with missing blocks, when added to {{UnderReplicatedBlocks}} -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-8010) Erasure coding: extend UnderReplicatedBlocks to accurately handle striped blocks
[ https://issues.apache.org/jira/browse/HDFS-8010?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14385161#comment-14385161 ] Zhe Zhang commented on HDFS-8010: - My current thoughts are: # Move the logic of checking expected, actual, and decomm replicas into {{UnderReplicatedBlocks#add}}. This way it becomes a single place to handle both contiguous and striped blocks. # Better placement of striped blocks into priority levels (maybe more levels are needed) [~jingzhao] Not sure if you have already started working on this. If so please feel free to assign to yourself. Erasure coding: extend UnderReplicatedBlocks to accurately handle striped blocks Key: HDFS-8010 URL: https://issues.apache.org/jira/browse/HDFS-8010 Project: Hadoop HDFS Issue Type: Sub-task Reporter: Zhe Zhang This JIRA tracks efforts to accurately assess the _risk level_ of a striped block groups with missing blocks, when added to {{UnderReplicatedBlocks}} -- This message was sent by Atlassian JIRA (v6.3.4#6332)