| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
Update documentation (#1970) Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 513c86bf-84d2-4521-88c6-c963e28036b0 | 2 个月前 | |
Update documentation (#1970) Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 513c86bf-84d2-4521-88c6-c963e28036b0 | 2 个月前 | |
Align READONLY and READWRITE with Redis cluster semantics (#2109) * change semantics of ro and rw commands * separate clusterSession internal write from readOnlySession context * ensure cluster test stability by issuing read-only at connect/reconnect * add focused tests excercising default behavior of readOnlySession * update docs | 25 天前 | |
Fix Vector Set RESP2 and RESP3 reply contracts (#2184) * Fix Vector Set RESP2 and RESP3 reply contracts Cover all twelve Vector Set commands under both protocols and correct VREM boolean, VSIM missing-attribute, and VINFO metadata replies. Test and document the existing VLINKS protocol types without changing its native neighbor grouping. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 8f3348cd-0095-4ac5-ad48-7ee7f210131b * Strengthen VINFO protocol field assertions Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 8f3348cd-0095-4ac5-ad48-7ee7f210131b * Expect null attributes in existing VSIM protocol tests Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 8f3348cd-0095-4ac5-ad48-7ee7f210131b * Parameterize Vector Set protocol tests at fixture level Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 8f3348cd-0095-4ac5-ad48-7ee7f210131b --------- Co-authored-by: Tiago Martins Napoli <tiagonapoli@microsoft.com> Co-authored-by: GitHub Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 8f3348cd-0095-4ac5-ad48-7ee7f210131b | 5 天前 | |
Update DiskANN FFI (#1992) * add logCallback * add valueLengthHint to ReadCallbackUnmanaged * update filterCallback to take attribute data directly * wire up continue_search * sketch out VRANDMEMBER and random_members * update vector-sets.md * sketch out VLINKS and search_neighbors * add asserts that all keys coming FROM DiskANN are aligned (i.e. multiples of 4) * removing alignment hackery, assert that data passed back to DiskANN is aligned * document aligment requirements * formatting * remove more dead alignment code * update limits doc * add missing assert * relax alignment restrictions on external ids and attributes * formatting * formatting * retry if quant task fails * add beam_width to search_vector and search_element; default to 4 * change some FFI i32s to u32s * bump diskann-garnet to 5.0.0 * get compiling again * fix key length logic in vector set batch reads * align DiskANNInsertResult with diskann-garnet names * restore length prefix - exceptions are illegal here, so simple control flow * fix test * fix VGETATTR so it looks at the new location for these * stopgap fix for attribute filter evaluation, there's a managed <-> unmanaged transiton that complicates things * condition gc suppression in reader path * fix attribute lookup * attribute storage has shrunk, kick test sizes up to force eviction * add failing test for continue_search * one other failing test for VLINKS continue_search * test and fixes for VRANDMEMBER, along with its weird Redis-isms * scale output_distances_len by sizeof(float) to match search_xxx(...) contract * guarantee we always have space for at least M distances in search_neighbors * update vector-sets.md files * pull in latest diskann-garnet, various fixes from a working version of overflow_results (nee continue_search) * bump to diskann-garnet 5.0.2; all but 1 test passing now, something about quantization * bump diskann-garnet to 5.0.3 * don't spin forever in quantization if we're shutting down * don't log so much in this test * test all quantizers; shrink duration a bit so this isn't brutal additional test time * don't truncate RMW on CopyUpdate path for VectorSessionFunctions if callback is specified, violates documented contract to do so (and corrupts quantizer data in practice) * formatting * fix VLINKS WITHSCORE responses * fix id len read in RandomMembers; technically this would never fail because we're only ever going to use 1-byte of the 4-bytes we have allocated, but still best to be correct here * use helper instead of fetching memory manually * correctly pin large values in VectorSessionFunctions.Reader path if they need to be passed to a callback * use correct size hint in batch read callback * don't leak memory on ContinueSearch * don't leak GC handles in RandomMembers * don't leak memory in RandomMembers * empty VLINKS response is valid and needed a test * forbid int.MinValue COUNT for VRANDMEMBER * add more detail to vector-set.md about new commands | 10 天前 | |
Update documentation (#1970) Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 513c86bf-84d2-4521-88c6-c963e28036b0 | 2 个月前 | |
[Cluster] Read version 2 endpoints before enabling writers (breaking) (#2128) * Accept version 2 cluster endpoints before enabling writers Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 5c19fe0a-0745-48d8-ab1d-7e1e8e3a2fd8 Address reader review comments and remove rollout internals from docs Negotiate gossip versions and retain peer routing during rolling upgrades Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 5c19fe0a-0745-48d8-ab1d-7e1e8e3a2fd8 Prepare automatic request-matched gossip for rolling upgrades Keep first-release requests in version one and persist peer metadata in version two. Preserve known peer endpoints across version-one owner replies. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 5c19fe0a-0745-48d8-ab1d-7e1e8e3a2fd8 Avoid async state machine for gossip initialization Reuse serialized config during replica failover Preserve synthetic cluster bus port output [Docs] Remove internal endpoint rollout details * [Cluster] Document version 2 configuration compatibility * Clarify cluster config version and ownership semantics * Name cluster config support range explicitly * Clarify cluster config version comment --------- Co-authored-by: Vasileios Zois <96085550+vazois@users.noreply.github.com> | 4 天前 | |
Update documentation (#1970) Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 513c86bf-84d2-4521-88c6-c963e28036b0 | 2 个月前 | |
Update docs, move purgebp to a debug subcommand (#1984) * Update documentation Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 513c86bf-84d2-4521-88c6-c963e28036b0 * Move PURGEBP to a DEBUG subcommand PURGEBP is a debug/admin utility, so move it from a top-level RESP command to `DEBUG PURGEBP <manager-type>`, alongside FORCEGC/FLUSHANDEVICT. The read-only top-level RespCommand.PURGEBP is removed; its frozen v3 AOF slot maps to NONE (never persisted). Metadata, ACL test, and docs updated accordingly. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 513c86bf-84d2-4521-88c6-c963e28036b0 --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Co-authored-by: Ted Hart <15467143+TedHartMS@users.noreply.github.com> Copilot-Session: 513c86bf-84d2-4521-88c6-c963e28036b0 | 2 个月前 |
| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
| 2 个月前 | ||
| 2 个月前 | ||
| 25 天前 | ||
| 5 天前 | ||
| 10 天前 | ||
| 2 个月前 | ||
| 4 天前 | ||
| 2 个月前 | ||
| 2 个月前 |