Skip to content

v0.3.7

Choose a tag to compare

@ShangmingCai ShangmingCai released this 25 Oct 02:56
· 277 commits to main since this release
9e4f96b

What's Changed

  • [Store] skip null buffer by @XucSh in #812
  • [Store] Change Default Value of eviction_high_watermark_ratio and eviction_ratio by @ykwd in #820
  • [CI/Build] gate mooncake-store test behind BUILD_UNIT_TESTS option by @peng1999 in #821
  • [Doc] Store Integrated to SGLang HiCache by @ykwd in #829
  • [TransferEngine]: remove SO_REUSEADDR in findAvailableTcpPort by @doujiang24 in #830
  • [Build] Install Python Files by @ykwd in #836
  • [TransferEngine] Make ascend TE to be released successfully and support fast recovery from failures through retry by @hjchen2 in #827
  • [Build] Install Python Files Patch by @ykwd in #839
  • [Doc] SGLang HiCache Intergration by @ykwd in #833
  • [Docs] Fix Broken Trace Link by @ykwd in #841
  • feat(store): add NUMA node binding support via bind_to_numa_node method by @xiaguan in #823
  • docs(deployment): Add Basic Mooncake Store deployment guide by @xiaguan in #825
  • refactor(store): use dedicated thread for signal handling by @xiaguan in #840
  • add ascend protocol to mooncake store by @ascend-direct-dev in #835
  • store: Add json file and improve doc by @201341 in #843
  • feat(store): add client heartbeat support for non ha mode by @xiaguan in #845
  • Fix typo in issue template by @Zane-Jiang in #858
  • Fix nvlink_transport bug: revert #683 by @ShangmingCai in #869
  • fix adxl find tcp port bug by @ascend-direct-dev in #856
  • chore: bump version to 0.3.6.post1 in pyproject.toml by @ShangmingCai in #870
  • [Transfer Engine] Post notify if all transfer tasks are completed by @alogfans in #831
  • feat(store): support transfer engine p2phandshake by @xiaguan in #852
  • [Chores] Remove Unused Variable by @ykwd in #822
  • [TransferEngine] Performance Enhancement for Heterogeneous Ascend via Intelligent Aggregation & Pipeline Design by @zuochunwei in #859
  • [Misc] feat: Support external kv_connector for vllm v1 by @dtcccc in #865
  • feat(store): disable auto discovery by default, require devices for RDMA by @xiaguan in #877
  • Allow customizing RPC port range by @peng1999 in #873
  • [Store] Check If Get Completed Within Lease by @ykwd in #778
  • [Docs] Update Obsolete Content & Fix Minor Problems by @ykwd in #880
  • fix(store): fix memory leak in client_integration_test.cpp by @JINGE-ui in #881
  • Refactor(store): Remove BufStatus and segment_name for AllocatedBuffer by @xiaguan in #883
  • [Store] Check if Connecting Master Fails by @ykwd in #886
  • [TransferEngine] clear all transport mems for fast recovery for ascend transport by @hjchen2 in #847
  • [Misc] Mooncake EP & Mooncake Backend by @UNIDY2002 in #805
  • [Docs] Update quick start and usage examples by @chestnut-Q in #893
  • fix(store): disable persistence instead of returning error by @xiaguan in #892
  • [Store]: Get start_time before calling RPC in BatchQuery by @nickyc975 in #896
  • feat(store): Add multi threading handle page fault during segment allocation by @xiaguan in #875
  • docs(store): restructure and simplify SGLang HiCache integration guide by @xiaguan in #897
  • [Store]: Add option to use jemalloc in mooncake store master by @nickyc975 in #902
  • [Misc] improvements for mooncake_connector_v1 by @dtcccc in #906
  • [TransferEngine] initiator_test script: make it works with P2PHANDSHAKE. by @doujiang24 in #907
  • [CI/Build] For Mooncake EP, fix the flag USE_CUDA that was unexpectedly turned off by @UNIDY2002 in #909
  • [Misc] For EP, pass device_name instead of nic_id when creating ep.Buffer by @UNIDY2002 in #910
  • [Doc] Add Mooncake x SGLang Hicache Design and Some Updates by @ykwd in #913
  • fix(doc): Add Hicache Design to Index.md by @ykwd in #914
  • [TransferEngine] Add Moore Threads GPUs Support by @popsiclexu in #862
  • feat(TE): add notify support for sync transfers and expose getNotifies API by @staryxchen in #894
  • mooncake-backend chunked transfer by @ympcMark in #911
  • ascend direct transport support transfer to multiple destinations in one batch by @ascend-direct-dev in #857
  • fix(store): Fix integer overflow in get_into/batch_get_into for values > 4GB by @xiaguan in #920
  • [Doc] Add Clarification for STORE_USE_ETCD Compile Option by @ykwd in #927
  • [Store] Change log level for batch operation by @stmatengss in #916
  • Reduced build-with-ep workflow by @JasonZhang517 in #926
  • [Misc] Fix chunked impl of _reduce_scatter_base by @UNIDY2002 in #931
  • feat: add batch_put_from_multi_buffers by @LCAIZJ in #929
  • [Misc] Fix the shutdown logic of Mooncake Backend by @UNIDY2002 in #933
  • [CI/Build] Always build with EP in CI by @UNIDY2002 in #922
  • [Integration] feat: introduce barex allocator by @stmatengss in #932
  • TE: adxl config without buffer pool by @ascend-direct-dev in #941
  • fix(transfer_engine): replace deprecated Json::Reader by @xiaguan in #938
  • [CI] Fix CI Error Due to RDMA Fail by @ykwd in #930
  • mlx5gda.cpp: add cleanup to destroy ah in mlx5gda_modify_rc_qp_init2rtr by @zhilishui in #945
  • [doc] Fix documentation link by @Liziqi-77 in #949
  • feat(store_service): support load config from env for mooncake store_service by @Syspretor in #951
  • [Misc] For Mooncake Backend, skip transferring to non-active ranks by @UNIDY2002 in #953
  • Enable CUDA support in CI configuration by @ShangmingCai in #937
  • Bugfix issue 946 by @uniqueni in #947
  • Bump version to 0.3.7 in pyproject.toml by @ShangmingCai in #959
  • Try to fix the release CI by @UNIDY2002 in #962

New Contributors

Full Changelog: v0.3.6...v0.3.7