<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>VACUUM on OLAP Studio — Independent DuckDB Technical Blog</title><link>https://duckdblab.org/en/tags/vacuum/</link><description>Recent content in VACUUM on OLAP Studio — Independent DuckDB Technical Blog</description><generator>Hugo -- gohugo.io</generator><language>en-US</language><lastBuildDate>Fri, 20 Nov 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://duckdblab.org/en/tags/vacuum/index.xml" rel="self" type="application/rss+xml"/><item><title>DuckDB Space Reclaiming, Tested: VACUUM Doesn't Shrink, CHECKPOINT Doesn't Truncate, COPY FROM DATABASE Actually Compacts</title><link>https://duckdblab.org/en/post/duckdb-vacuum-reclaim-space/</link><pubDate>Fri, 20 Nov 2026 00:00:00 +0000</pubDate><guid>https://duckdblab.org/en/post/duckdb-vacuum-reclaim-space/</guid><description>&lt;img src="https://duckdblab.org/images/posts/duckdb-vacuum-reclaim-space/cover.png" alt="Featured image of post DuckDB Space Reclaiming, Tested: VACUUM Doesn't Shrink, CHECKPOINT Doesn't Truncate, COPY FROM DATABASE Actually Compacts" /&gt;&lt;h2 id="a-common-wrong-intuition"&gt;A Common (Wrong) Intuition
&lt;/h2&gt;&lt;p&gt;If you&amp;rsquo;ve used PostgreSQL or SQLite, you carry a mental model: &lt;strong&gt;delete a lot of data, run &lt;code&gt;VACUUM&lt;/code&gt;, and the file gets smaller.&lt;/strong&gt; So it&amp;rsquo;s tempting to treat DuckDB&amp;rsquo;s &lt;code&gt;VACUUM&lt;/code&gt; as the switch that shrinks a &lt;code&gt;.duckdb&lt;/code&gt; file.&lt;/p&gt;
&lt;p&gt;It is not. The official docs say it up front:&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;The &lt;code&gt;VACUUM&lt;/code&gt; statement only has basic support in DuckDB and is mostly provided for PostgreSQL-compatibility. … The &lt;code&gt;VACUUM&lt;/code&gt; statement does &lt;strong&gt;not&lt;/strong&gt; reclaim space.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;p&gt;In other words, &lt;strong&gt;DuckDB&amp;rsquo;s &lt;code&gt;VACUUM&lt;/code&gt; neither reclaims disk space nor shrinks the file.&lt;/strong&gt; Its real jobs are a syntax-compatibility placeholder for Postgres users and (via &lt;code&gt;VACUUM ANALYZE table(col)&lt;/code&gt;) rebuilding per-column statistics.&lt;/p&gt;
&lt;p&gt;So how &lt;em&gt;do&lt;/em&gt; you actually get the file down in size after deleting lots of rows? This post measures a 1-million-row database across four stages and pins down the true division of labor among &lt;strong&gt;&lt;code&gt;VACUUM&lt;/code&gt; / &lt;code&gt;CHECKPOINT&lt;/code&gt; / &lt;code&gt;COPY FROM DATABASE&lt;/code&gt;&lt;/strong&gt;.&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;&lt;strong&gt;Scope&lt;/strong&gt;: one focused topic — space reclamation — measured on DuckDB 1.5 (Variegata). Every number is reproducible with the commands at the end.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="1-three-tools-three-roles-the-short-version"&gt;1. Three Tools, Three Roles (the short version)
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Tool&lt;/th&gt;
 &lt;th&gt;Real effect&lt;/th&gt;
 &lt;th&gt;Does the &lt;code&gt;.duckdb&lt;/code&gt; file get smaller?&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;&lt;code&gt;VACUUM;&lt;/code&gt; / &lt;code&gt;VACUUM ANALYZE;&lt;/code&gt;&lt;/td&gt;
 &lt;td&gt;Whole-database: &lt;strong&gt;no-op&lt;/strong&gt;&lt;/td&gt;
 &lt;td&gt;No&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;&lt;code&gt;VACUUM my_table(my_col);&lt;/code&gt;&lt;/td&gt;
 &lt;td&gt;Rebuilds stats for that table&amp;rsquo;s column&lt;/td&gt;
 &lt;td&gt;No (touches stats, not space)&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;&lt;code&gt;CHECKPOINT;&lt;/code&gt; / &lt;code&gt;FORCE CHECKPOINT;&lt;/code&gt;&lt;/td&gt;
 &lt;td&gt;Flushes WAL into the data file and &lt;strong&gt;merges&lt;/strong&gt; qualifying row groups&lt;/td&gt;
 &lt;td&gt;Reuses free space, but &lt;strong&gt;does not truncate&lt;/strong&gt;&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;&lt;code&gt;COPY FROM DATABASE a TO b;&lt;/code&gt;&lt;/td&gt;
 &lt;td&gt;Copies content into a &lt;strong&gt;fresh&lt;/strong&gt; database (a compaction rewrite)&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;Yes&lt;/strong&gt;, the file genuinely shrinks&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;In one line: &lt;strong&gt;the only reliable way to make the on-disk file smaller is &lt;code&gt;COPY FROM DATABASE&lt;/code&gt;; &lt;code&gt;CHECKPOINT&lt;/code&gt; stops further growth; &lt;code&gt;VACUUM&lt;/code&gt; is essentially unrelated to space.&lt;/strong&gt;&lt;/p&gt;
&lt;h2 id="2-minimal-reproducible-experiment-1m-rows-delete-half-measure-4-sizes"&gt;2. Minimal Reproducible Experiment: 1M Rows, Delete Half, Measure 4 Sizes
&lt;/h2&gt;&lt;p&gt;Everything below runs directly in the DuckDB CLI. File sizes are measured with bash &lt;code&gt;stat -c '%s'&lt;/code&gt; (bytes).&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Step 0: build 1M rows, checkpoint&lt;/strong&gt;&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" class="chroma"&gt;&lt;code class="language-sql" data-lang="sql"&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="k"&gt;CREATE&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;TABLE&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;t&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;AS&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;SELECT&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;range&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;AS&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;id&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s1"&gt;&amp;#39;payload_&amp;#39;&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="o"&gt;||&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;repeat&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s1"&gt;&amp;#39;x&amp;#39;&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mi"&gt;40&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;AS&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;payload&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;FROM&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;range&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="mi"&gt;1000000&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="k"&gt;CHECKPOINT&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;div class="highlight"&gt;&lt;pre tabindex="0" class="chroma"&gt;&lt;code class="language-bash" data-lang="bash"&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;$ stat -c &lt;span class="s1"&gt;&amp;#39;%s&amp;#39;&lt;/span&gt; exp.duckdb
&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="m"&gt;536576&lt;/span&gt; &lt;span class="c1"&gt;# S0: 1M rows, after build + checkpoint&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;&lt;strong&gt;Step 1: delete half (&lt;code&gt;id % 2 = 0&lt;/code&gt;), no checkpoint yet&lt;/strong&gt;&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" class="chroma"&gt;&lt;code class="language-sql" data-lang="sql"&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="k"&gt;DELETE&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;FROM&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;t&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;WHERE&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;id&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="o"&gt;%&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mi"&gt;2&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;div class="highlight"&gt;&lt;pre tabindex="0" class="chroma"&gt;&lt;code class="language-bash" data-lang="bash"&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;$ stat -c &lt;span class="s1"&gt;&amp;#39;%s&amp;#39;&lt;/span&gt; exp.duckdb
&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="m"&gt;798720&lt;/span&gt; &lt;span class="c1"&gt;# S1: the file got *bigger* after the delete (delete marks + WAL)&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;Note the counterintuitive bit: you deleted 50% of the data and the file &lt;em&gt;grew&lt;/em&gt; (536KB -&amp;gt; 798KB). DuckDB is append-only; a delete just writes tombstones into the WAL rather than erasing in place.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Step 2: run &lt;code&gt;VACUUM&lt;/code&gt;&lt;/strong&gt;&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" class="chroma"&gt;&lt;code class="language-sql" data-lang="sql"&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="k"&gt;VACUUM&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;div class="highlight"&gt;&lt;pre tabindex="0" class="chroma"&gt;&lt;code class="language-bash" data-lang="bash"&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;$ stat -c &lt;span class="s1"&gt;&amp;#39;%s&amp;#39;&lt;/span&gt; exp.duckdb
&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="m"&gt;798720&lt;/span&gt; &lt;span class="c1"&gt;# S2: identical to S1 — VACUUM has zero effect on size&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;&lt;strong&gt;Step 3: run &lt;code&gt;CHECKPOINT&lt;/code&gt;&lt;/strong&gt;&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" class="chroma"&gt;&lt;code class="language-sql" data-lang="sql"&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="k"&gt;CHECKPOINT&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;div class="highlight"&gt;&lt;pre tabindex="0" class="chroma"&gt;&lt;code class="language-bash" data-lang="bash"&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;$ stat -c &lt;span class="s1"&gt;&amp;#39;%s&amp;#39;&lt;/span&gt; exp.duckdb
&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="m"&gt;798720&lt;/span&gt; &lt;span class="c1"&gt;# S3: still not truncated, size unchanged&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;This is the step that most surprises people: run &lt;code&gt;CHECKPOINT&lt;/code&gt;, look at the file, see no shrink, and conclude &amp;ldquo;reclamation didn&amp;rsquo;t work.&amp;rdquo; It actually &lt;strong&gt;did&lt;/strong&gt; reclaim — just not in the way you expected.&lt;/p&gt;
&lt;h2 id="3-why-checkpoint-doesnt-truncate-the-file"&gt;3. Why CHECKPOINT Doesn&amp;rsquo;t Truncate the File
&lt;/h2&gt;&lt;p&gt;DuckDB uses a &lt;strong&gt;single-file format&lt;/strong&gt;. The official &amp;ldquo;Reclaiming Space&amp;rdquo; page opens with:&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;DuckDB uses a single-file format, which has some inherent limitations w.r.t. reclaiming disk space.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;p&gt;The mechanism: at checkpoint time, DuckDB &lt;strong&gt;merges&lt;/strong&gt; adjacent row groups whose deletes cross a threshold and marks that space as reusable. It does &lt;strong&gt;not&lt;/strong&gt; shorten the file&amp;rsquo;s byte length — the file stays at its historical high-water mark, and the freed region is simply reused on the &lt;em&gt;next&lt;/em&gt; writes instead of growing the file further.&lt;/p&gt;
&lt;p&gt;So S3 staying flat is correct behavior. &lt;strong&gt;&amp;ldquo;Reclaim&amp;rdquo; ≈ &amp;ldquo;stop growing as fast,&amp;rdquo; not &amp;ldquo;immediately smaller on disk.&amp;rdquo;&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;The docs add one detail: merging kicks in when ~25% of rows in adjacent row groups are deleted. Delete too little (say 5%) and checkpoint won&amp;rsquo;t merge those row groups, so no space is freed. Here we deleted 50%, well above the threshold, so merging does happen — visible as &amp;ldquo;reusable space,&amp;rdquo; not as truncation.&lt;/p&gt;
&lt;h2 id="4-vacuums-real-job-rebuilding-stats-not-collecting-space"&gt;4. VACUUM&amp;rsquo;s Real Job: Rebuilding Stats, Not Collecting Space
&lt;/h2&gt;&lt;p&gt;Break &lt;code&gt;VACUUM&lt;/code&gt; into its forms (measured on 1.5):&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" class="chroma"&gt;&lt;code class="language-sql" data-lang="sql"&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="k"&gt;VACUUM&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="c1"&gt;-- whole DB: no-op, does nothing
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="k"&gt;VACUUM&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;ANALYZE&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="c1"&gt;-- whole DB: also a no-op
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="k"&gt;VACUUM&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;t&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;id&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="c1"&gt;-- rebuilds stats for column id of table t (silent success)
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="k"&gt;VACUUM&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;FULL&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="c1"&gt;-- error: Not implemented Error: Full vacuum option
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;&lt;code&gt;VACUUM t(id)&lt;/code&gt; is the one thing it &lt;strong&gt;can&lt;/strong&gt; actually do: when a column&amp;rsquo;s statistics (e.g. distinct count) go stale after heavy UPDATE/DELETE, rebuild them so the optimizer picks a better plan. That has &lt;strong&gt;nothing&lt;/strong&gt; to do with disk space.&lt;/p&gt;
&lt;p&gt;So &amp;ldquo;use VACUUM to shrink the file&amp;rdquo; is, semantically, the wrong tool from the start.&lt;/p&gt;
&lt;h2 id="5-actually-shrinking-the-file-copy-from-database"&gt;5. Actually Shrinking the File: &lt;code&gt;COPY FROM DATABASE&lt;/code&gt;
&lt;/h2&gt;&lt;p&gt;The official docs point to &lt;code&gt;COPY FROM DATABASE&lt;/code&gt; for compaction:&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;To compact the database, you can create a fresh copy of the database using the &lt;code&gt;COPY FROM DATABASE&lt;/code&gt; statement.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;p&gt;Measured (continuing from section 2; the file is now 900k rows at 1,060,864 bytes):&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" class="chroma"&gt;&lt;code class="language-sql" data-lang="sql"&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="n"&gt;ATTACH&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s1"&gt;&amp;#39;exp.duckdb&amp;#39;&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;AS&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;db1&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="n"&gt;ATTACH&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s1"&gt;&amp;#39;exp_compact.duckdb&amp;#39;&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;AS&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;db2&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="n"&gt;USE&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;db2&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="k"&gt;COPY&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;FROM&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;DATABASE&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;db1&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;TO&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;db2&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="k"&gt;CHECKPOINT&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;div class="highlight"&gt;&lt;pre tabindex="0" class="chroma"&gt;&lt;code class="language-bash" data-lang="bash"&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;$ stat -c &lt;span class="s1"&gt;&amp;#39;%s&amp;#39;&lt;/span&gt; exp.duckdb&lt;span class="p"&gt;;&lt;/span&gt; stat -c &lt;span class="s1"&gt;&amp;#39;%s&amp;#39;&lt;/span&gt; exp_compact.duckdb
&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="m"&gt;1060864&lt;/span&gt; &lt;span class="c1"&gt;# source, inflated&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="m"&gt;536576&lt;/span&gt; &lt;span class="c1"&gt;# new DB, compacted; row count unchanged (900k rows)&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;&lt;strong&gt;Size dropped from 1,060,864 to 536,576 — about 49% smaller&lt;/strong&gt; — with no data loss. That&amp;rsquo;s the canonical &amp;ldquo;really shrink the file&amp;rdquo; move in DuckDB: rewrite into a clean database, then switch to it.&lt;/p&gt;
&lt;p&gt;A common production pattern: &lt;code&gt;COPY FROM DATABASE&lt;/code&gt; into a new file, verify row counts / checksums, then atomically rename-replace the old file. The source stays read-only throughout, so it&amp;rsquo;s safe.&lt;/p&gt;
&lt;h2 id="6-two-extras-14--15"&gt;6. Two Extras (1.4+ / 1.5)
&lt;/h2&gt;&lt;p&gt;&lt;strong&gt;1. In-memory databases can checkpoint + compress (since 1.4)&lt;/strong&gt;&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" class="chroma"&gt;&lt;code class="language-sql" data-lang="sql"&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="n"&gt;ATTACH&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s1"&gt;&amp;#39;:memory:&amp;#39;&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;AS&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;mcmp&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;COMPRESS&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="n"&gt;USE&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;mcmp&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="k"&gt;CREATE&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;TABLE&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;x&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;AS&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;SELECT&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;range&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;AS&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;id&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;repeat&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s1"&gt;&amp;#39;z&amp;#39;&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mi"&gt;200&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;AS&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nb"&gt;blob&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;FROM&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;range&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="mi"&gt;50000&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="k"&gt;CHECKPOINT&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;Before 1.4, in-memory tables didn&amp;rsquo;t checkpoint or compress. Since 1.4, in-memory tables support both compression and &amp;ldquo;delete-row vacuuming&amp;rdquo; — handy for resident memory workloads that churn rows constantly (e.g. an AI agent&amp;rsquo;s in-memory working set).&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;2. ART-indexed tables + &lt;code&gt;vacuum_rebuild_indexes&lt;/code&gt; (experimental)&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;Tables with &lt;code&gt;PRIMARY KEY&lt;/code&gt;/&lt;code&gt;UNIQUE&lt;/code&gt; constraints use ART indexes under the hood. By default, &lt;code&gt;VACUUM&lt;/code&gt; &lt;strong&gt;skips&lt;/strong&gt; tables with ART indexes; the experimental &lt;code&gt;vacuum_rebuild_indexes&lt;/code&gt; setting lets vacuum compact those tables (rebuilding the indexes afterward), with &lt;code&gt;0&lt;/code&gt; (the default) disabling it.&lt;/p&gt;
&lt;p&gt;Gotcha, measured: this setting &lt;strong&gt;cannot be changed while the database is running&lt;/strong&gt; —&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" class="chroma"&gt;&lt;code class="language-sql" data-lang="sql"&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="k"&gt;SET&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;vacuum_rebuild_indexes&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mi"&gt;200000&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="c1"&gt;-- Invalid Input Error: Cannot change vacuum_rebuild_indexes setting while database is running
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;It can only be configured at startup / &lt;code&gt;ATTACH&lt;/code&gt; time.&lt;/p&gt;
&lt;h2 id="7-automatic-checkpoint-wal_autocheckpoint"&gt;7. Automatic Checkpoint: &lt;code&gt;wal_autocheckpoint&lt;/code&gt;
&lt;/h2&gt;&lt;p&gt;You don&amp;rsquo;t have to sprinkle &lt;code&gt;CHECKPOINT&lt;/code&gt; by hand. DuckDB auto-checkpoints based on WAL size:&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" class="chroma"&gt;&lt;code class="language-sql" data-lang="sql"&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="k"&gt;SELECT&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;name&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;value&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;FROM&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;duckdb_settings&lt;/span&gt;&lt;span class="p"&gt;()&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="k"&gt;WHERE&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="n"&gt;name&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s1"&gt;&amp;#39;wal_autocheckpoint&amp;#39;&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="line"&gt;&lt;span class="cl"&gt;&lt;span class="c1"&gt;-- wal_autocheckpoint | 16.0 MiB
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;The default threshold is 16 MiB: once the WAL reaches 16MB, an automatic checkpoint fires. On write-heavy workloads you can lower it (flush WAL more eagerly, control the file high-water mark); on read-mostly workloads raising it is harmless. Understanding this default explains &amp;ldquo;why does my &lt;code&gt;.duckdb&lt;/code&gt; keep growing&amp;rdquo; — usually it&amp;rsquo;s WAL growth triggering checkpoints, while the file never truncates.&lt;/p&gt;
&lt;h2 id="8-when-to-use-what-decision-table"&gt;8. When to Use What (Decision Table)
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Your goal&lt;/th&gt;
 &lt;th&gt;Correct action&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Deleted lots of rows; want to &lt;strong&gt;stop growing / reuse free space&lt;/strong&gt;&lt;/td&gt;
 &lt;td&gt;&lt;code&gt;CHECKPOINT&lt;/code&gt; (or rely on auto &lt;code&gt;wal_autocheckpoint&lt;/code&gt;)&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Want the &lt;strong&gt;on-disk &lt;code&gt;.duckdb&lt;/code&gt; to genuinely shrink&lt;/strong&gt;&lt;/td&gt;
 &lt;td&gt;&lt;code&gt;COPY FROM DATABASE&lt;/code&gt; into a new file, then swap&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Column stats stale, query plans degraded&lt;/td&gt;
 &lt;td&gt;&lt;code&gt;VACUUM my_table(my_col)&lt;/code&gt; to rebuild stats&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Reach for &lt;code&gt;VACUUM FULL&lt;/code&gt; by habit&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;Not supported&lt;/strong&gt; on 1.5 — it errors with &lt;code&gt;Not implemented&lt;/code&gt;&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Want to compact a PK/UNIQUE-constrained table&lt;/td&gt;
 &lt;td&gt;&lt;code&gt;vacuum_rebuild_indexes&lt;/code&gt; (experimental, set at startup)&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;In-memory resident DB churning rows&lt;/td&gt;
 &lt;td&gt;&lt;code&gt;ATTACH ':memory:' (...) (COMPRESS)&lt;/code&gt; + &lt;code&gt;CHECKPOINT&lt;/code&gt; (1.4+)&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="go-deeper"&gt;Go Deeper
&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;&lt;a class="link" href="https://duckdblab.org/en/post/duckdb-complete-guide/" &gt;DuckDB Complete Guide&lt;/a&gt; — the overall roadmap from install to advanced&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://duckdblab.org/en/post/duckdb-performance-tuning-ultimate-guide/" &gt;DuckDB Performance Tuning Ultimate Guide&lt;/a&gt; — space reclamation is only one piece; read it alongside the tuning guide&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://duckdblab.org/en/post/duckdb-predicate-pushdown-filter-indexes/" &gt;DuckDB Predicate Pushdown &amp;amp; Indexes&lt;/a&gt; — how the stats rebuilt by &lt;code&gt;VACUUM ANALYZE&lt;/code&gt; drive the optimizer&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://duckdblab.org/en/post/duckdb-production-deployment-guide/" &gt;DuckDB Production Deployment Guide&lt;/a&gt; — how to safely run &amp;ldquo;compaction rewrite + atomic swap&amp;rdquo; in production&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="reproducibility"&gt;Reproducibility
&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Version&lt;/strong&gt;: DuckDB 1.5.2 (Variegata), Linux x86_64, run directly in the CLI; all sizes measured with &lt;code&gt;stat -c '%s'&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Reproduce&lt;/strong&gt;: run &lt;code&gt;duckdb exp.duckdb &amp;lt; setup.sql&lt;/code&gt; in the order of sections 2 and 5; use section 5&amp;rsquo;s &lt;code&gt;COPY FROM DATABASE&lt;/code&gt; for the shrink comparison.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Official references&lt;/strong&gt;: &lt;a class="link" href="https://duckdb.org/docs/1.5/sql/statements/vacuum.html" target="_blank" rel="noopener"
 &gt;VACUUM Statement (1.5)&lt;/a&gt;, &lt;a class="link" href="https://duckdb.org/docs/1.5/sql/statements/checkpoint.html" target="_blank" rel="noopener"
 &gt;CHECKPOINT Statement (1.5)&lt;/a&gt;, &lt;a class="link" href="https://duckdb.org/docs/1.5/operations_manual/footprint_of_duckdb/reclaiming_space.html" target="_blank" rel="noopener"
 &gt;Reclaiming Space (1.5)&lt;/a&gt;, &lt;a class="link" href="https://duckdb.org/docs/1.5/configuration/overview.html" target="_blank" rel="noopener"
 &gt;Configuration (1.5)&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Last verified&lt;/strong&gt;: 2026-11-17. If you spot an issue, please open a &lt;a class="link" href="https://github.com/pengzz9527/duckdb-blog/issues" target="_blank" rel="noopener"
 &gt;GitHub Issue&lt;/a&gt;.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="further-reading"&gt;Further Reading
&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;&lt;a class="link" href="https://duckdblab.org/en/post/essbase-duckdb-offline-analytics/" &gt;Essbase x DuckDB: offline analytics for a classic MOLAP cube&lt;/a&gt; — a real data pipeline where file footprint and rewrite cost are first-order concerns&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://duckdblab.org/en/post/duckdb-zero-diff-reconciliation/" &gt;Zero-Diff Reconciliation in DuckDB&lt;/a&gt; — the same &amp;ldquo;measure, don&amp;rsquo;t assume&amp;rdquo; discipline applied to data integrity instead of disk space&lt;/li&gt;
&lt;/ul&gt;</description></item></channel></rss>