[ https://issues.apache.org/jira/browse/HBASE-7201?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13537470#comment-13537470 ]
stack commented on HBASE-7201: ------------------------------ Yeah, looking at this again, hfile has the following: {code} // The below three methods take Writables. We'd like to undo Writables but undoing the below would be pretty // painful. Could take a byte [] or a Message but we want to be backward compatible around hfiles so would need // to map between Message and Writable or byte [] and current Writable serialization. This would be a bit of work // to little gain. Thats my thinking at moment. St.Ack 20121129 void appendMetaBlock(String bloomFilterMetaKey, Writable metaWriter); /** * Store general Bloom filter in the file. This does not deal with Bloom filter * internals but is necessary, since Bloom filters are stored differently * in HFile version 1 and version 2. */ void addGeneralBloomFilter(BloomFilterWriter bfw); /** * Store delete family Bloom filter in the file, which is only supported in * HFile V2. */ void addDeleteFamilyBloomFilter(BloomFilterWriter bfw) throws IOException; {code} Am thinking its not the end the of the world if hfile does Writables past 0.96. On WAL side, ditto. I'd think that HLogKey and WALEdit need to evolve anyways -- i.e. write DataBlocks rather than just kvs. As part of evolution can write different formats. Meantime WAL and HFile using Writable I don't think a blocker on 0.96. Knocking down the priority. > Convert HLog / HFile metadata content to PB > ------------------------------------------- > > Key: HBASE-7201 > URL: https://issues.apache.org/jira/browse/HBASE-7201 > Project: HBase > Issue Type: Sub-task > Components: HFile, Protobufs, wal > Reporter: Enis Soztutar > Priority: Blocker > Fix For: 0.96.0 > > Attachments: 7201.txt > > > Some of the remaining discussions for PB conversions: > - Convert the HFile/HLog metadata to PB. > - WALEdit, HLogKey should be converted? We don't want to repeat the PBMagic, > and the PB overhead can be high, but this is needed for replication? > - We said no to converting KV. > These should not block 0.96. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira