Adar Dembo has posted comments on this change. ( http://gerrit.cloudera.org:8080/8144 )
Change subject: KUDU-2055 [part 3]: Refactor BlockCreationTransaction and BlockDeletionTransaction ...................................................................... Patch Set 10: (2 comments) http://gerrit.cloudera.org:8080/#/c/8144/10/src/kudu/fs/block_manager.h File src/kudu/fs/block_manager.h: http://gerrit.cloudera.org:8080/#/c/8144/10/src/kudu/fs/block_manager.h@289 PS10, Line 289: // Add a block to the creation transaction. This method is not thread-safe. I don't think it's appropriate for threads to share BlockCreationTransaction or BlockDeletionTransaction. Each thread should use its own transaction, or use external synchronization to safely share a single instance. So, rather than calling out just the Add methods as being thread unsafe, could you document that the classes themselves aren't thread safe? http://gerrit.cloudera.org:8080/#/c/8144/8/src/kudu/fs/log_block_manager.cc File src/kudu/fs/log_block_manager.cc: http://gerrit.cloudera.org:8080/#/c/8144/8/src/kudu/fs/log_block_manager.cc@1152 PS8, Line 1152: std::unique_ptr<WritableBlock> block) { > Done, but I kept "std::vector<std::unique_ptr<LogWritableBlock>> created_bl Well, both pieces of state is definitely worse than just one, so I'd rather revert to the old approach or figure out what's wrong with the new one: 1. L1160 is cosmetic; it could just as easily VLOG about the number of containers with commited blocks. 2. L1162 can be changed to a double for loop to iterate on every block. Not a big deal. 3. What's the additional transformation needed on L1176? It's already iterating over created_block_map; am I missing something? -- To view, visit http://gerrit.cloudera.org:8080/8144 To unsubscribe, visit http://gerrit.cloudera.org:8080/settings Gerrit-Project: kudu Gerrit-Branch: master Gerrit-MessageType: comment Gerrit-Change-Id: I60c7d437061f98ad27b9aecda5fa89e909fb2ec6 Gerrit-Change-Number: 8144 Gerrit-PatchSet: 10 Gerrit-Owner: Hao Hao <[email protected]> Gerrit-Reviewer: Adar Dembo <[email protected]> Gerrit-Reviewer: Dan Burkert <[email protected]> Gerrit-Reviewer: Hao Hao <[email protected]> Gerrit-Reviewer: Kudu Jenkins Gerrit-Comment-Date: Tue, 10 Oct 2017 22:59:32 +0000 Gerrit-HasComments: Yes
