> On May 13, 2014, 6:37 p.m., Timothy Chen wrote: > > clients/src/main/java/org/apache/kafka/clients/producer/internals/Sender.java, > > line 323 > > <https://reviews.apache.org/r/21398/diff/2/?file=580820#file580820line323> > > > > Should we only call forceUpdate once if we get multiple unknown nodes?
The benefit of forcing metadata update whenever there is one ready partition with unknown leader is to minimize latency for these cases. Since partitions with unknown leaders (e.g. newly created topic, extended partitions, etc) would be a rare case, I think this would not add too much load on metadata request and a good tradeoff for latency. - Guozhang ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/21398/#review42869 ----------------------------------------------------------- On May 13, 2014, 6:25 p.m., Guozhang Wang wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/21398/ > ----------------------------------------------------------- > > (Updated May 13, 2014, 6:25 p.m.) > > > Review request for kafka. > > > Bugs: KAFKA-1445 > https://issues.apache.org/jira/browse/KAFKA-1445 > > > Repository: kafka > > > Description > ------- > > 0. Add the partitionsForNode index in Cluster;\n 1. Ready would return a list > of ready nodes instead of partitions;\n 2. Ready would also check if there is > any ready partitions with unknown leader, if yes indicate the > processReadyNode to force metadata refresh;\n 3. Drain would take a list of > nodes and drain the batches per node until the max request size is reached;\n > 4. Collocate would not be just tranform batches per node into a producer > request;\n 5. Corresponding unit test changes; \n 6. One minor compilation > warning fix > > > Diffs > ----- > > > clients/src/main/java/org/apache/kafka/clients/producer/internals/Partitioner.java > fbb732a57522109ac0e0eaf5c87b50cbd3a7f767 > > clients/src/main/java/org/apache/kafka/clients/producer/internals/RecordAccumulator.java > 2d7e52d430fa267ee3689a06f8a621ce5dfd1e33 > > clients/src/main/java/org/apache/kafka/clients/producer/internals/Sender.java > f0152fabbdd44e9f1a24291e84c17edf8379f4fc > clients/src/main/java/org/apache/kafka/common/Cluster.java > 426bd1eec708979149cbd6fa3959e6f9e73c7e0e > clients/src/main/java/org/apache/kafka/common/Node.java > 0e47ff3ff0e055823ec5a5aa4839d25b0fac8374 > > clients/src/test/java/org/apache/kafka/clients/producer/RecordAccumulatorTest.java > f37ab770b1794830154f9908a0156e7e99b4a458 > > clients/src/test/java/org/apache/kafka/common/utils/AbstractIteratorTest.java > 1df226606fad29da47d81d0b8ff36209c3536c06 > > Diff: https://reviews.apache.org/r/21398/diff/ > > > Testing > ------- > > unit tests > > > Thanks, > > Guozhang Wang > >
