[
https://issues.apache.org/jira/browse/HDFS-7941?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15316948#comment-15316948
]
Johannes Herr commented on HDFS-7941:
-------------------------------------
The file may not appear to grow because the new length is not sent to the
Namenode on hsync or because Block-compression is used and the data is still in
the compress buffer.
See http://stackoverflow.com/a/37664114/136247 for more details.
> hsync() not working
> -------------------
>
> Key: HDFS-7941
> URL: https://issues.apache.org/jira/browse/HDFS-7941
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: hdfs-client
> Affects Versions: 2.6.0
> Environment: HDP 2.2 running on Redhat
> Reporter: Sverre Bakke
>
> When using SequenceFile.Writer and appending+syncing to file repeatedly, the
> sync does not appear to work other than:
> - once after writing headers
> - when closing.
> Imagine the following test case:
> http://pastebin.com/Y9xysCRX
> This code would append a new record every second and then immediately sync
> it. One would also imagine that the file would grow for every append,
> however, this does not happen.
> After watching the behavior I have noticed that it only syncs the headers at
> the very beginning (providing a file of 164 bytes) and then never again until
> its closed. This despite it is asked to hsync() after every append.
> Looking into the debug logs, this also claims the same behavior (executed the
> provided code example and grepped for "sync"):
> SLF4J: Failed to load class "org.slf4j.impl.StaticLoggerBinder".
> SLF4J: Defaulting to no-operation (NOP) logger implementation
> SLF4J: See http://www.slf4j.org/codes.html#StaticLoggerBinder for further
> details.
> 2015-03-17 15:55:14 DEBUG ProtobufRpcEngine:253 - Call: fsync took 11ms
> This was the only time the code ran fsync throughout the entire execution.
> This has been tested (with similar result) for the following deployments:
> - sequencefile with no compression
> - sequencefile with record compression
> - sequencefile with block compression
> - textfile with no compression
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
