[ https://issues.apache.org/jira/browse/HBASE-14790?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15135606#comment-15135606 ]
Duo Zhang commented on HBASE-14790: ----------------------------------- Fine. Let's do it in HBase. A little problem is that if we get an error then the only way to close the file is calling recoverLease. The reason is that I do not want to resend data to datanode but in the current DTP there is no way to truncate block data from client side so the only way to make a consensus on the block length is calling recoverLease... In the new design, so this is not a big problem. And also, we need to resolve HBASE-14949 first before applying this patch. In the current implementation, we do not do pipeline recovery, so it is easier for us to meet a sync failure. If we simply fail the sync request, we will also easier to meet the inconsistency... So here we need the logic described in HBASE-14004, write the unacked entries to new WAL file if sync failed. This will lead to the problem described in HBASE-14949 that we may have two WAL files with different data mapping to the same name when splitting. On the implementation, the problem is I need to share lots of code of the original FSHLog and other related classes. But this should not be a blocker, let me try implementing a new WALProvider next. And for the reviewing, you just do it on Github? Or I upload it here and on reviewboard? I'm always happy with a review. Thanks. > Implement a new DFSOutputStream for logging WAL only > ---------------------------------------------------- > > Key: HBASE-14790 > URL: https://issues.apache.org/jira/browse/HBASE-14790 > Project: HBase > Issue Type: Improvement > Reporter: Duo Zhang > > The original {{DFSOutputStream}} is very powerful and aims to serve all > purposes. But in fact, we do not need most of the features if we only want to > log WAL. For example, we do not need pipeline recovery since we could just > close the old logger and open a new one. And also, we do not need to write > multiple blocks since we could also open a new logger if the old file is too > large. > And the most important thing is that, it is hard to handle all the corner > cases to avoid data loss or data inconsistency(such as HBASE-14004) when > using original DFSOutputStream due to its complicated logic. And the > complicated logic also force us to use some magical tricks to increase > performance. For example, we need to use multiple threads to call {{hflush}} > when logging, and now we use 5 threads. But why 5 not 10 or 100? > So here, I propose we should implement our own {{DFSOutputStream}} when > logging WAL. For correctness, and also for performance. -- This message was sent by Atlassian JIRA (v6.3.4#6332)