https://vchord.pgsty.com/zh/pgvecto-rs/reference/ Recent content in 参考手册 on Vchord Hugo zh-CN Mon, 02 Nov 2026 19:35:19 +0800 https://vchord.pgsty.com/zh/pgvecto-rs/reference/schema/ Mon, 01 Jan 0001 00:00:00 +0000 https://vchord.pgsty.com/zh/pgvecto-rs/reference/schema/ <p>以下是 <code>pgvecto.rs</code> 提供的数据库对象定义（Schema）。</p> <h2 id="list-of-data-types">数据类型列表 </h2> <div class="td-table-scroll td-table-scroll--static"> <table> <thead> <tr> <th scope="col">名称</th> <th scope="col">说明</th> </tr> </thead> <tbody> <tr> <td>vector</td> <td>稠密向量，标量元素类型为 IEEE 754-2008 定义的 <code>binary32</code>（32 位浮点数）</td> </tr> <tr> <td>vecf16</td> <td>半精度向量，标量元素类型为 IEEE 754-2008 定义的 <code>binary16</code>（16 位浮点数）</td> </tr> <tr> <td>veci8</td> <td>整数向量，标量元素类型为 <code>8-bit Integer</code>（8 位有符号整数）</td> </tr> <tr> <td>svector</td> <td>稀疏向量，标量元素类型为 IEEE 754-2008 定义的 <code>binary32</code>（32 位浮点数）</td> </tr> <tr> <td>bvector</td> <td>二值向量，定长位串（bit string）</td> </tr> <tr> <td>vector_index_stat</td> <td>表示向量索引统计信息的复合类型</td> </tr> </tbody> </table> </div> <h2 id="list-of-operators">操作符列表 </h2> <div class="td-table-scroll td-table-scroll--static"> <table> <thead> <tr> <th scope="col">名称</th> <th scope="col">左操作数类型</th> <th scope="col">右操作数类型</th> <th scope="col">返回类型</th> <th scope="col">说明</th> </tr> </thead> <tbody> <tr> <td>+</td> <td>vector</td> <td>vector</td> <td>vector</td> <td>逐元素加法运算</td> </tr> <tr> <td>+</td> <td>vecf16</td> <td>vecf16</td> <td>vecf16</td> <td>逐元素加法运算</td> </tr> <tr> <td>+</td> <td>veci8</td> <td>veci8</td> <td>veci8</td> <td>逐元素加法运算</td> </tr> <tr> <td>+</td> <td>svector</td> <td>svector</td> <td>svector</td> <td>逐元素加法运算</td> </tr> <tr> <td>-</td> <td>vector</td> <td>vector</td> <td>vector</td> <td>逐元素减法运算</td> </tr> <tr> <td>-</td> <td>vecf16</td> <td>vecf16</td> <td>vecf16</td> <td>逐元素减法运算</td> </tr> <tr> <td>-</td> <td>veci8</td> <td>veci8</td> <td>veci8</td> <td>逐元素减法运算</td> </tr> <tr> <td>-</td> <td>svector</td> <td>svector</td> <td>svector</td> <td>逐元素减法运算</td> </tr> <tr> <td>&amp;</td> <td>bvector</td> <td>bvector</td> <td>bvector</td> <td>逐元素按位与（AND）</td> </tr> <tr> <td>|</td> <td>bvector</td> <td>bvector</td> <td>bvector</td> <td>逐元素按位或（OR）</td> </tr> <tr> <td>^</td> <td>bvector</td> <td>bvector</td> <td>bvector</td> <td>逐元素按位异或（XOR）</td> </tr> <tr> <td>=</td> <td>vector</td> <td>vector</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>=</td> <td>vecf16</td> <td>vecf16</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>=</td> <td>veci8</td> <td>veci8</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>=</td> <td>svector</td> <td>svector</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>=</td> <td>bvector</td> <td>bvector</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&lt;&gt;</td> <td>vector</td> <td>vector</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&lt;&gt;</td> <td>vecf16</td> <td>vecf16</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&lt;&gt;</td> <td>veci8</td> <td>veci8</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&lt;&gt;</td> <td>svector</td> <td>svector</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&lt;&gt;</td> <td>bvector</td> <td>bvector</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&lt;</td> <td>vector</td> <td>vector</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&lt;</td> <td>vecf16</td> <td>vecf16</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&lt;</td> <td>veci8</td> <td>veci8</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&lt;</td> <td>svector</td> <td>svector</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&lt;</td> <td>bvector</td> <td>bvector</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&gt;</td> <td>vector</td> <td>vector</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&gt;</td> <td>vecf16</td> <td>vecf16</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&gt;</td> <td>veci8</td> <td>veci8</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&gt;</td> <td>svector</td> <td>svector</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&gt;</td> <td>bvector</td> <td>bvector</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&lt;=</td> <td>vector</td> <td>vector</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&lt;=</td> <td>vecf16</td> <td>vecf16</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&lt;=</td> <td>veci8</td> <td>veci8</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&lt;=</td> <td>svector</td> <td>svector</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&lt;=</td> <td>bvector</td> <td>bvector</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&gt;=</td> <td>vector</td> <td>vector</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&gt;=</td> <td>vecf16</td> <td>vecf16</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&gt;=</td> <td>veci8</td> <td>veci8</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&gt;=</td> <td>svector</td> <td>svector</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&gt;=</td> <td>bvector</td> <td>bvector</td> <td>boolean</td> <td>字典序比较</td> </tr> <tr> <td>&lt;-&gt;</td> <td>vector</td> <td>vector</td> <td>real</td> <td>平方欧氏距离</td> </tr> <tr> <td>&lt;-&gt;</td> <td>vecf16</td> <td>vecf16</td> <td>real</td> <td>平方欧氏距离</td> </tr> <tr> <td>&lt;-&gt;</td> <td>veci8</td> <td>veci8</td> <td>real</td> <td>平方欧氏距离</td> </tr> <tr> <td>&lt;-&gt;</td> <td>svector</td> <td>svector</td> <td>real</td> <td>平方欧氏距离</td> </tr> <tr> <td>&lt;-&gt;</td> <td>bvector</td> <td>bvector</td> <td>real</td> <td>平方欧氏距离</td> </tr> <tr> <td>&lt;#&gt;</td> <td>vector</td> <td>vector</td> <td>real</td> <td>负点积</td> </tr> <tr> <td>&lt;#&gt;</td> <td>vecf16</td> <td>vecf16</td> <td>real</td> <td>负点积</td> </tr> <tr> <td>&lt;#&gt;</td> <td>veci8</td> <td>veci8</td> <td>real</td> <td>负点积</td> </tr> <tr> <td>&lt;#&gt;</td> <td>svector</td> <td>svector</td> <td>real</td> <td>负点积</td> </tr> <tr> <td>&lt;#&gt;</td> <td>bvector</td> <td>bvector</td> <td>real</td> <td>负点积</td> </tr> <tr> <td>&lt;=&gt;</td> <td>vector</td> <td>vector</td> <td>real</td> <td>余弦距离</td> </tr> <tr> <td>&lt;=&gt;</td> <td>vecf16</td> <td>vecf16</td> <td>real</td> <td>余弦距离</td> </tr> <tr> <td>&lt;=&gt;</td> <td>veci8</td> <td>veci8</td> <td>real</td> <td>余弦距离</td> </tr> <tr> <td>&lt;=&gt;</td> <td>svector</td> <td>svector</td> <td>real</td> <td>余弦距离</td> </tr> <tr> <td>&lt;=&gt;</td> <td>bvector</td> <td>bvector</td> <td>real</td> <td>余弦距离</td> </tr> <tr> <td>&lt;~&gt;</td> <td>bvector</td> <td>bvector</td> <td>real</td> <td>杰卡德距离（Jaccard 距离）</td> </tr> </tbody> </table> </div> <h2 id="list-of-functions">函数列表 </h2> <div class="td-table-scroll td-table-scroll--static"> <table> <thead> <tr> <th scope="col">名称</th> <th scope="col">参数类型</th> <th scope="col">返回类型</th> <th scope="col">说明</th> </tr> </thead> <tbody> <tr> <td>pgvectors_upgrade</td> <td></td> <td>void</td> <td>用于升级 <code>pgvecto.rs</code> 的管理函数。</td> </tr> <tr> <td>to_svector</td> <td>dims integer, index integer[], value real[]</td> <td>svector</td> <td>根据下标数组和值数组构造一个稀疏向量。</td> </tr> <tr> <td>binarize</td> <td>vector vector</td> <td>bvector</td> <td>将向量二值化：所有正数元素设为 1，其余元素设为 0。</td> </tr> <tr> <td>text2vec_openai</td> <td>input text, model text</td> <td>vector</td> <td>调用 OpenAI Embeddings API 的嵌入生成函数。</td> </tr> <tr> <td>text2vec_openai_v3</td> <td>input text</td> <td>vector</td> <td>调用 OpenAI Embeddings API 的嵌入生成函数，默认模型设为 <code>text-embedding-3-small</code>。</td> </tr> </tbody> </table> </div> <h2 id="list-of-casts">类型转换列表 </h2> <div class="td-table-scroll td-table-scroll--static"> <table> <thead> <tr> <th scope="col">源类型</th> <th scope="col">目标类型</th> <th scope="col">是否隐式转换？</th> </tr> </thead> <tbody> <tr> <td>real[]</td> <td>vector</td> <td>yes</td> </tr> <tr> <td>vector</td> <td>real[]</td> <td>yes</td> </tr> <tr> <td>vector</td> <td>vecf16</td> <td>no</td> </tr> <tr> <td>vecf16</td> <td>vector</td> <td>no</td> </tr> <tr> <td>vector</td> <td>veci8</td> <td>no</td> </tr> <tr> <td>veci8</td> <td>vector</td> <td>no</td> </tr> <tr> <td>vector</td> <td>svector</td> <td>no</td> </tr> <tr> <td>svector</td> <td>vector</td> <td>no</td> </tr> <tr> <td>vector</td> <td>bvector</td> <td>no</td> </tr> <tr> <td>bvector</td> <td>vector</td> <td>no</td> </tr> </tbody> </table> </div> <h2 id="list-of-access-methods">访问方法（Access Methods）列表 </h2> <div class="td-table-scroll td-table-scroll--static"> <table> <thead> <tr> <th scope="col">名称</th> <th scope="col">类型</th> <th scope="col">说明</th> </tr> </thead> <tbody> <tr> <td>vectors</td> <td>Index</td> <td>pgvecto.rs 向量索引方法</td> </tr> </tbody> </table> </div> <h2 id="list-of-operator-families">操作符族（Operator Families）列表 </h2> <div class="td-table-scroll td-table-scroll--static"> <table> <thead> <tr> <th scope="col">访问方法</th> <th scope="col">操作符族</th> <th scope="col">适用类型</th> </tr> </thead> <tbody> <tr> <td>vectors</td> <td>vector_cos_ops</td> <td>vector</td> </tr> <tr> <td>vectors</td> <td>vector_dot_ops</td> <td>vector</td> </tr> <tr> <td>vectors</td> <td>vector_l2_ops</td> <td>vector</td> </tr> <tr> <td>vectors</td> <td>vecf16_cos_ops</td> <td>vecf16</td> </tr> <tr> <td>vectors</td> <td>vecf16_dot_ops</td> <td>vecf16</td> </tr> <tr> <td>vectors</td> <td>vecf16_l2_ops</td> <td>vecf16</td> </tr> <tr> <td>vectors</td> <td>veci8_cos_ops</td> <td>veci8</td> </tr> <tr> <td>vectors</td> <td>veci8_dot_ops</td> <td>veci8</td> </tr> <tr> <td>vectors</td> <td>veci8_l2_ops</td> <td>veci8</td> </tr> <tr> <td>vectors</td> <td>svector_cos_ops</td> <td>svector</td> </tr> <tr> <td>vectors</td> <td>svector_dot_ops</td> <td>svector</td> </tr> <tr> <td>vectors</td> <td>svector_l2_ops</td> <td>svector</td> </tr> <tr> <td>vectors</td> <td>bvector_cos_ops</td> <td>bvector</td> </tr> <tr> <td>vectors</td> <td>bvector_dot_ops</td> <td>bvector</td> </tr> <tr> <td>vectors</td> <td>bvector_l2_ops</td> <td>bvector</td> </tr> <tr> <td>vectors</td> <td>bvector_jaccard_ops</td> <td>bvector</td> </tr> </tbody> </table> </div> <h2 id="list-of-operator-classes">操作符类（Operator Classes）列表 </h2> <div class="td-table-scroll td-table-scroll--static"> <table> <thead> <tr> <th scope="col">访问方法</th> <th scope="col">输入类型</th> <th scope="col">操作符类</th> <th scope="col">是否默认？</th> <th scope="col">说明</th> </tr> </thead> <tbody> <tr> <td>vectors</td> <td>vector</td> <td>vector_l2_ops</td> <td>no</td> <td>平方欧氏距离</td> </tr> <tr> <td>vectors</td> <td>vector</td> <td>vector_dot_ops</td> <td>no</td> <td>负点积</td> </tr> <tr> <td>vectors</td> <td>vector</td> <td>vector_cos_ops</td> <td>no</td> <td>余弦距离</td> </tr> <tr> <td>vectors</td> <td>vecf16</td> <td>vecf16_l2_ops</td> <td>no</td> <td>平方欧氏距离</td> </tr> <tr> <td>vectors</td> <td>vecf16</td> <td>vecf16_dot_ops</td> <td>no</td> <td>负点积</td> </tr> <tr> <td>vectors</td> <td>vecf16</td> <td>vecf16_cos_ops</td> <td>no</td> <td>余弦距离</td> </tr> <tr> <td>vectors</td> <td>veci8</td> <td>veci8_l2_ops</td> <td>no</td> <td>平方欧氏距离</td> </tr> <tr> <td>vectors</td> <td>veci8</td> <td>veci8_dot_ops</td> <td>no</td> <td>负点积</td> </tr> <tr> <td>vectors</td> <td>veci8</td> <td>veci8_cos_ops</td> <td>no</td> <td>余弦距离</td> </tr> <tr> <td>vectors</td> <td>svector</td> <td>svector_l2_ops</td> <td>no</td> <td>平方欧氏距离</td> </tr> <tr> <td>vectors</td> <td>svector</td> <td>svector_dot_ops</td> <td>no</td> <td>负点积</td> </tr> <tr> <td>vectors</td> <td>svector</td> <td>svector_cos_ops</td> <td>no</td> <td>余弦距离</td> </tr> <tr> <td>vectors</td> <td>bvector</td> <td>bvector_l2_ops</td> <td>no</td> <td>平方欧氏距离</td> </tr> <tr> <td>vectors</td> <td>bvector</td> <td>bvector_dot_ops</td> <td>no</td> <td>负点积</td> </tr> <tr> <td>vectors</td> <td>bvector</td> <td>bvector_cos_ops</td> <td>no</td> <td>余弦距离</td> </tr> <tr> <td>vectors</td> <td>bvector</td> <td>bvector_jaccard_ops</td> <td>no</td> <td>杰卡德距离（Jaccard 距离）</td> </tr> </tbody> </table> </div> <h2 id="list-of-views">视图列表 </h2> <div class="td-table-scroll td-table-scroll--static"> <table> <thead> <tr> <th scope="col">名称</th> <th scope="col">说明</th> </tr> </thead> <tbody> <tr> <td>pg_vector_index_stat</td> <td>提供向量索引统计信息的监控视图</td> </tr> </tbody> </table> </div> https://vchord.pgsty.com/zh/pgvecto-rs/reference/indexing_options/ Mon, 01 Jan 0001 00:00:00 +0000 https://vchord.pgsty.com/zh/pgvecto-rs/reference/indexing_options/ <p>我们使用 <a href="https://toml.io/en/v1.0.0">TOML 语法</a>来表达索引的配置参数。以下是配置中各个顶层键（表）的含义：</p> <div class="td-table-scroll td-table-scroll--static"> <table> <thead> <tr> <th scope="col">键名</th> <th scope="col">类型</th> <th scope="col">说明</th> </tr> </thead> <tbody> <tr> <td>segment</td> <td>table</td> <td>分段（Segment）相关选项。</td> </tr> <tr> <td>optimizing</td> <td>table</td> <td>后台优化相关选项。</td> </tr> <tr> <td>indexing</td> <td>table</td> <td>用于构建索引的算法配置。</td> </tr> </tbody> </table> </div> <h2 id="options-for-table-segment"><code>segment</code> 表选项 </h2> <div class="td-table-scroll td-table-scroll--static"> <table> <thead> <tr> <th scope="col">键名</th> <th scope="col">类型</th> <th scope="col">取值范围</th> <th scope="col">默认值</th> <th scope="col">说明</th> </tr> </thead> <tbody> <tr> <td>max_growing_segment_size</td> <td>integer</td> <td><code>[1, 4_000_000_000]</code></td> <td><code>20_000</code></td> <td>未建立索引的增长中分段的最大向量数量。</td> </tr> <tr> <td>max_sealed_segment_size</td> <td>integer</td> <td><code>[1, 4_000_000_000]</code></td> <td><code>1_000_000</code></td> <td>已封存（建立索引）分段的最大向量数量。</td> </tr> </tbody> </table> </div> <h2 id="options-for-table-optimizing"><code>optimizing</code> 表选项 </h2> <div class="td-table-scroll td-table-scroll--static"> <table> <thead> <tr> <th scope="col">键名</th> <th scope="col">类型</th> <th scope="col">取值范围</th> <th scope="col">默认值</th> <th scope="col">说明</th> </tr> </thead> <tbody> <tr> <td>optimizing_threads</td> <td>integer</td> <td><code>[1, 65535]</code></td> <td>CPU 核心数的平方根</td> <td>后台构建索引的最大线程数。</td> </tr> <tr> <td>sealing_secs</td> <td>integer</td> <td><code>[1, 60]</code></td> <td><code>60</code></td> <td>见下方说明。</td> </tr> <tr> <td>sealing_size</td> <td>integer</td> <td><code>[1, 4_000_000_000]</code></td> <td><code>1</code></td> <td>见下方说明。</td> </tr> </tbody> </table> </div> <p>如果一个写入分段（write segment）中的向量数大于 <code>sealing_size</code>，且在连续 <code>sealing_secs</code> 秒内没有接收到新数据，该写入分段将被转为已封存分段（sealed segment）以构建索引。</p> https://vchord.pgsty.com/zh/pgvecto-rs/reference/search_options/ Mon, 01 Jan 0001 00:00:00 +0000 https://vchord.pgsty.com/zh/pgvecto-rs/reference/search_options/ <p>搜索选项通过 <a href="https://www.postgresql.org/docs/current/config-setting.html">PostgreSQL GUC 参数</a>进行指定。</p> <h2 id="options-for-ivf"><code>ivf</code> 选项 </h2> <p>以下选项仅对 IVF 算法生效。详情请参阅<a href="https://vchord.pgsty.com/zh/pgvecto-rs/usage/indexing/#inverted-file-index-ivf">索引指南</a>。</p> <div class="td-table-scroll td-table-scroll--static"> <table> <thead> <tr> <th scope="col">选项</th> <th scope="col">类型</th> <th scope="col">取值范围</th> <th scope="col">默认值</th> <th scope="col">说明</th> </tr> </thead> <tbody> <tr> <td>vectors.ivf_nprobe</td> <td>integer</td> <td><code>[1, 1_000_000]</code></td> <td><code>10</code></td> <td>扫描的列表数量。</td> </tr> </tbody> </table> </div> <h2 id="options-for-hnsw"><code>hnsw</code> 选项 </h2> <p>以下选项仅对 HNSW 算法生效。详情请参阅<a href="https://vchord.pgsty.com/zh/pgvecto-rs/usage/indexing/#hierarchical-navigable-small-world-graph-hnsw">索引指南</a>。</p>