maybe i misunderstand the method 'backgroundFlushCommits', if AP has error, then a exception will be thrown inside 'backgroundFlushCommits, so if AP has error, the same put will not be flush twice
but, validatePut(put); should be executed if AP has error [email protected] From: Ted Yu Date: 2015-09-07 10:06 To: [email protected] Subject: Re: improvement in htable.doput Looking at the tip of 0.98 branch, the logic in doPut() is the same as shown below. You may want to open a JIRA and attach patch if you have time. Cheers On Sun, Sep 6, 2015 at 6:15 PM, [email protected] < [email protected]> wrote: > in hbase0.98.10, doPut is as follows: > > private void doPut(Put put) throws InterruptedIOException, > RetriesExhaustedWithDetailsException { > if (ap.hasError()){ > writeAsyncBuffer.add(put); > backgroundFlushCommits(true); > } > > validatePut(put); > > currentWriteBufferSize += put.heapSize(); > writeAsyncBuffer.add(put); > > while (currentWriteBufferSize > writeBufferSize) { > backgroundFlushCommits(false); > } > } > > i have two suggesstions: > 1. if ap has error, then this put will be added to buffer twice and flush > twice, why? > 2. if really need to flush twice, then we should validate, increament > buffer size, finally add this put to buffer. > > > > [email protected] >
