------------------------- Riak 0.13.0 Release Notes ------------------------- Map/Reduce performance has been substantially improved: * Improved Javascript call scheduling Previous versions of Riak MapReduce didn't handle over load situations well especially when the number of available Javascript VMs was low. Riak does a better job of tracking which Javascript VMs are busy and attempts to spread the load more evenly over all VMs. * Improved Javascript call caching The caching layer for Javascript MapReduce calls has been completely re-implemented. This should result in better performance when MapReduce jobs are run repeatedly. This work includes a new in-memory vnode LRU cache just for map operations. The size of the cache, as in the number of entries, can be controlled via the 'vnode_cache_entries' entry in the riak_kv section of app.config. The cache size defaults to 100. Setting the entry to 0 will disable the cache entirely. * Fixed MapReduce job input handling This operation was non-blocking in prior versions of Riak. The lack of backpressure caused excessive memory bloat and poor performance when the input producer overran the cluster's ability to process new MapReduce inputs. This has been fixed in the current release and should result in lower resource consumption and more stable run times. The riak-admin script now has two new commands, "ringready" and "wait-for-service". These enable the administrator to script detection of cluster convergence on a given ring and wait for KV to come up, respectively. Bitcask now uses 40% less memory per key entry and starts up significantly faster when the dataset is large. It also reclaims memory of expired key entries. Key listing when using Bitcask as a backend is substantially faster. The structure of the Riak source repos has been changed to have a single Erlang app per repo. This permits 3rd parties to use riak_core (the abstracted Dynamo logic) in other applications. The consistent hash ring balancing logic has been improved to enable large clusters (>32 nodes) to converge quickly on a common ring. Further work has been done to improve Riak's performance and robustness when dealing with large datasets and failure scenarios. Enhancements ------------ 140 - bin/riak now warns if ulimit -n is < 1024 168 - Performance of list_keys and mapred_bucket has been improved 213 - Default number of async threads has been increased to 64 446 - Each Erlang app has a dedicated repo 514 - dialyzer make target 578 - Re-examine garbage collection management of vnodes 651 - Merge should exit cleanly instead of with throw if files do not exist 696 - Garbage collection in Erlang J/S occurs more frequently 748 - Hint file reading is now buffered 081 - Support large uploads (chunk or stream automagically) Bugs Fixed ---------- 26 - The 'devrel' target now builds on CentOS. 201 - RpbPutResp has a typo: "contents" instead of "content" 217 - bin/riak stop doesn't seem to be working on Solaris 270 - Content-type set using riak_object:new/4 doesn't saved in db. 507 - Bitcask leaks file-handles when reading lock data for merge 510 - Riak 0.12 cannot be started after restarting the server 512 - Riak EE RPM removes bitcask directory when uninstalled 523 - M/R job timeout not propagating in all cases 544 - PBC interface starts listening before socket supervisor is started 552 - Bitcask expiry does not reclaim keydir memory usage at merge 559 - riak_object:new auto-update only when initial metadata 570 - Errors in console for javascript client tests, post-M/R speedup 589 - Typo in bin/riak stop 619 - Listing keys triggers enormous memory consumption 621 - Starting Riak with largish dataset causes riak_core_ring_handler to crash 656 - Protobuffs erlang client returns "function_clause" error when listing buckets 658 - riak_client:list_buckets returns no results 706 - erlang_js segfaults under load with large JSON objects 708 - Running multiple MapReduce jobs causes errors 719 - Vnode exits during/after? handoff 723 - ets backend throws exception in stop/1 728 - "riak-admin test" exits with status 0 on timeout 731 - ebloom:clear() fails on Solaris platforms 735 - riak_kv_bitcask_backend:is_empty can leak file descriptors 737 - Debian init script reload doesn't work 742 - Packaging Changes for 0.13.0rc2 746 - riak build fails on R14 757 - Bitcask should properly count dead bytes when writing in same file 773 - "Not Found" tokens not correctly handled in streaming M/R results 774 - riak-admin wait-for-service riak responds "riak is up" 776 - Update admin.org and riak*-admin man files to include new commands 784 - Solaris packages shouldn't delete modified config files on removal Riak has been updated with the necessary changes to compile on Erlang R14, but has not been thoroughly tested on R14. Please continue to run Riak on R13B04 in production. All bug and issue numbers reference https://issues.basho.com.
