Hi,

I am using HFiles on HDFS for an application and I am seeing an issue which
I dont see with HBase. Each time a seek is performed on the HFile using
HFileReaderV2, it is read 128K from disk instead of just reading a single
block - I have verified this by looking at strace and tracing the file
descriptors on which reads are being performed. It seems to always read
"dfs.client.read.shortcircuit.buffer.size" amount of data from the hfile
instead of just reading enough information in the data block. When I use
the default value of 1M for the short circuit buffer, I see reads of upto
1M.

Shouldn't the seek only fetch just the block or am i missing configuration
parameter ?

Thanks
Varun

Reply via email to