libsql: read a cursor response to its end so the connection is reused by zkochan · Pull Request #2301 · tursodatabase/libsql

zkochan · 2026-10-23T18:19:11Z

OwnedCursorStep dropped its Cursor, and the /v3/cursor response body with it, as soon as it read the step_end entry. hyper had not read the final chunk yet, so it closed the connection instead of pooling it. Every query on a remote connection then opened a new TCP connection, and against Turso a new TLS handshake.

After step_end, the step now keeps reading the cursor until the stream ends, then reports no more rows. Entries after step_end are skipped. A read error at that point ends the rows without an error, because the step's rows are already complete.

A caller still has to read the rows to the end (call next() until it returns None) for the connection to be reused. A Rows dropped after its first row still drops the body early.

Testing

New unit test owned_step_reads_the_response_to_its_end. It fails without the change.
Against a local libsql-server, 50 query("SELECT 1") calls read to the end opened 50 connections before and 2 after.

Written by an agent (Claude Code, claude-opus-5-5).

OwnedCursorStep dropped its cursor, and the HTTP response body with it, on the step_end entry. hyper had not read the final chunk, so it closed the connection, and every remote query opened a new one. The step now reads the rest of the response before it reports no more rows. Fixes tursodatabase#2300 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Every token lookup against a remote libsql database opened a new TCP connection, and against Turso a new TLS handshake. A query's rows come from a chunked `/v3/cursor` response. libsql dropped the response body when it read the `step_end` entry, before hyper had read the final chunk, so hyper closed the connection instead of pooling it. pnpr's lookups also dropped their rows after the first one. - vendor/libsql carries libsql 0.9.30 with the fix from tursodatabase/libsql#2301: after `step_end` the step reads the response to its end. `src/lib.rs` allows warnings, because cargo does not cap the lints of path dependencies and the workspace builds with `-D warnings`. - The auth backend reads every single-row query through `first_row`, which reads the row's values and then the rest of the rows. A local row reads the live statement, so its values are taken first. - The pre-commit formatter skips `vendor/`, so a vendored crate keeps its upstream formatting and edition. Against a local libsql-server, 200 authenticated requests opened 193 new connections before and none after. Closes pnpm/tasks#192 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

The pnpm-0.9.30 branch of pnpm/libsql is the 0.9.30 release commit plus the cursor fix from tursodatabase/libsql#2301. `[patch.crates-io]` pins it by revision, as it does pnpm/node-semver-rs, so the repository no longer carries a copy of the crate. Cargo caps the lints of git dependencies, so the fork needs no allow-warnings change. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Every token lookup against a remote libsql database opened a new TCP connection, and against Turso a new TLS handshake. A query's rows come from a chunked `/v3/cursor` response. libsql dropped the response body when it read the `step_end` entry, before hyper had read the final chunk, so hyper closed the connection instead of pooling it. pnpr's lookups also dropped their rows after the first one. - libsql comes from the pnpm-0.9.30 branch of pnpm/libsql: the 0.9.30 release plus the fix from tursodatabase/libsql#2301, after which a step reads the cursor response to its end. - The auth backend reads every single-row query through `first_row`, which reads the row's values and then the rest of the rows. A local row reads the live statement, so its values are taken first. Against a local libsql-server, 200 authenticated requests opened 193 new connections before and none after. Part of pnpm/tasks#192

Windows Rust CI aborted pnpr-auth libsql tests with 0xc0000005 (four times recently: registration_cap_is_strict_across_backend_instances three times, sync_server::bulk_revocation_refreshes_a_replica_whose_tokens_went_elsewhere once). libsql's LibsqlConnection::drop calls disconnect() and then the inner local::Connection's own Drop calls disconnect() again. Both see the drop_ref Arc as unique, so sqlite3_close_v2 runs a second time on a handle SQLite already freed. Usually it reads freed memory as "closed" and reports misuse. When another thread reused that memory, it read garbage and crashed. The tests that crashed run several connections concurrently, which makes reuse likely. Upstream fixed it in tursodatabase/libsql#2261 (released in 0.10.0), which nulls the handle after the first close. pnpr stays on the pnpm fork because tursodatabase/libsql#2301 is still unreleased, so the fix is cherry-picked onto the fork's pnpm-0.9.30 branch. The new test installs an SQLite log callback, drops one connection and asserts no misuse was logged. It logged two before the fork bump. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Windows Rust CI aborted pnpr-auth libsql tests with 0xc0000005 (four times recently: registration_cap_is_strict_across_backend_instances three times, sync_server::bulk_revocation_refreshes_a_replica_whose_tokens_went_elsewhere once). libsql's LibsqlConnection::drop calls disconnect() and then the inner local::Connection's own Drop calls disconnect() again. Both see the drop_ref Arc as unique, so sqlite3_close_v2 runs a second time on a handle SQLite already freed. Usually it reads freed memory as "closed" and reports misuse. When another thread reused that memory, it read garbage and crashed. The tests that crashed run several connections concurrently, which makes reuse likely. Upstream fixed it in tursodatabase/libsql#2261 (released in 0.10.0), which nulls the handle after the first close. pnpr stays on the pnpm fork because tursodatabase/libsql#2301 is still unreleased, so the fix is cherry-picked onto the fork's pnpm-0.9.30 branch. The new test installs an SQLite log callback, drops one connection and asserts no misuse was logged. It logged two before the fork bump. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

This was referenced Oct 23, 2026

perf(pnpr): reuse connections to a remote libsql auth database pnpm/pnpm#17299

Merged

pnpr: libsql auth backend opens a new TCP connection for most requests pnpm/tasks#192

Open

zkochan mentioned this pull request Oct 26, 2026

fix(pnpr): close each libsql connection once pnpm/pnpm#17366

Merged

3 tasks done

Navigation Menu

Repository navigation

libsql: read a cursor response to its end so the connection is reused - #2301

libsql: read a cursor response to its end so the connection is reused#2301
zkochan wants to merge 1 commit into
tursodatabase:mainfrom
zkochan:cursor-read-to-end

zkochan commented Oct 23, 2026

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

zkochan commented Oct 23, 2026

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant