----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/26478/#review56010 -----------------------------------------------------------
src/main/java/org/apache/aurora/scheduler/storage/log/SnapshotDeduplicator.java <https://reviews.apache.org/r/26478/#comment96362> This javadoc would highly benefit from some details about the source of duplication and a proposed solution. It's not obvious for a newcomer why TaskConfigs are duplicated. src/main/java/org/apache/aurora/scheduler/storage/log/SnapshotDeduplicator.java <https://reviews.apache.org/r/26478/#comment96365> Why result field here? src/main/java/org/apache/aurora/scheduler/storage/log/SnapshotDeduplicator.java <https://reviews.apache.org/r/26478/#comment96360> There is a remote possibility of numOutputTasks to be zero in a snapshot for an empty cluster. src/main/java/org/apache/aurora/scheduler/storage/log/SnapshotDeduplicator.java <https://reviews.apache.org/r/26478/#comment96363> Inverse log message of a hydration ratio would be useful here along with a "Starting redupulication". src/main/thrift/org/apache/aurora/gen/storage.thrift <https://reviews.apache.org/r/26478/#comment96364> Please, document fields. What is taskConfigId here? src/test/java/org/apache/aurora/scheduler/storage/log/SnapshotDeduplicatorImplTest.java <https://reviews.apache.org/r/26478/#comment96368> How about a roundtrip test with no tasks in a snapshot? - Maxim Khutornenko On Oct. 9, 2014, 2:39 a.m., Kevin Sweeney wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/26478/ > ----------------------------------------------------------- > > (Updated Oct. 9, 2014, 2:39 a.m.) > > > Review request for Aurora, David McLaughlin, Bill Farner, and Zameer Manji. > > > Bugs: AURORA-722 > https://issues.apache.org/jira/browse/AURORA-722 > > > Repository: aurora > > > Description > ------- > > Add a new format for deduplicated storage snapshots. Microbenchmarks show a > 10x deduplication ratio on Twitter's production snapshots. > > This format is backwards-incompatible, so this patch introduces a flag to > control its use (defaulting off). > > This only changes the format used to write to the replicated log (where time > is of the essence since all writes are done holding the global storage lock) > - the format of backups written to disk is unchanged, as backups don't hold > the lock. > > > Diffs > ----- > > build.gradle 2e1bf78d7797f17afd51a94a22eff80e00aba464 > src/main/java/org/apache/aurora/codec/ThriftBinaryCodec.java > 65e986eaa2c4193431ca048425a1ed3ab60f5882 > src/main/java/org/apache/aurora/scheduler/storage/log/EntrySerializer.java > 7239a6a5eb5479e395e16423c83fdf80a77e5a83 > src/main/java/org/apache/aurora/scheduler/storage/log/LogManager.java > 4b50e2069407dc263b4fc93f1827d3a8836253bf > src/main/java/org/apache/aurora/scheduler/storage/log/LogStorage.java > f806297d1d0700155c976743f936b2b8a3a390fb > src/main/java/org/apache/aurora/scheduler/storage/log/LogStorageModule.java > 769348e6b8a5c701734afff391b1c77de35222c6 > > src/main/java/org/apache/aurora/scheduler/storage/log/SnapshotDeduplicator.java > PRE-CREATION > src/main/java/org/apache/aurora/scheduler/storage/log/StreamManager.java > 22db80eaf34fe736fa5a3a9289836c9ac9e59906 > > src/main/java/org/apache/aurora/scheduler/storage/log/StreamManagerImpl.java > e5cfbf5cf43bf5bbc38c42fe685a7e9f0d03af2a > src/main/thrift/org/apache/aurora/gen/storage.thrift > 5350ec945fbe028ee4641683815a068ce00b5efc > src/test/java/org/apache/aurora/scheduler/storage/log/LogManagerTest.java > 39729b374fe4e383f9b5ada7d016923766df9af7 > src/test/java/org/apache/aurora/scheduler/storage/log/LogStorageTest.java > 7a8c3b882633376a1bf6a78616d55aaa7401d13f > > src/test/java/org/apache/aurora/scheduler/storage/log/SnapshotDeduplicatorImplTest.java > PRE-CREATION > > Diff: https://reviews.apache.org/r/26478/diff/ > > > Testing > ------- > > ./gradlew -Pq build > > > Thanks, > > Kevin Sweeney > >