Roadmap

As of 2025Q1, the content and order of goals are very much in flux.

Your input is very much welcome, please engage us in a support contract or on the community forums.

About this page

Tags used in the roadmap
  • #external: Likely source of the bug is not us, but in a dependency
  • #maybe: We might not do this
  • #urgency/high
  • #urgency/medium
  • #urgency/low
  • #severity/high
  • #severity/medium
  • #severity/low
  • #size/small
  • #size/medium
  • #size/big
  • #api-change: Changes already-published API
  • #bug: Likely bug in the software (not documentation)
  • #enhancement: New feature or improving behavior some other way
  • #performance: Improves performance, minimal effect on other behaviors
  • #ops: Affects operational aspects, not functionality; for example, observability
  • #dev: Only affects development and building from source, not the final product
  • #test: Topic where test coverage needs to be better
  • #doc: Improvements or additions to documentation
  • #good-first-issue: Good for newcomers
  • #need-more-info: We don't have know enough yet to work on this
  • #help-wanted: Extra attention would be appreciated
  • #ecosystem/*: We are hoping our dependencies will improve
  • #waiting: Cannot proceed at this time

The checklist items are in very rough priority order. Order will be adjusted based on feedback and discovery of critical paths.

The big picture

  • v0: Bootstrap to a real SQL database
  • v1: Production-ready database using RocksDB
  • v2: Pure Rust with new storage subsystem

Platforms

Support for non-Linux operating systems is not currently on the roadmap. We intend to actively use Linux-specific functionality for performance, security and simplicity.

For now, only x86-64 is actively tested, but everything is supposed to work cross-architecture (higher risk: 32-bit processing >4 GiB of data at once, 32-bit in general, big-endian).

Pending work

Work that is more or less "about to be done".

Unimplemented or buggy SQL operations

  • DataFusion explicitly refuses to work with many SQL types even when it has the corresponding underlying type; see https://docs.rs/datafusion/latest/datafusion/sql/planner/struct.SqlToRel.html#method.build_schema and convert_data_type, e.g. UInt64 vs UNSIGNED BIGINT
  • large number literals behave differently than in Postgres; SELECT i64_max_plus_1 gives float, Postgres looks to be using big numbers
  • CREATE UNIQUE INDEX .. [NULLS DISTINCT] with null columns (#ref/unimplemented_sql/w9j8q485anior)
    • use a multi-index with nullable columns in value not key, prefix scan on insert to detect conflict
  • UPDATE of column in primary key (#ref/unimplemented_sql/fpb157cronqcy)
  • tables without primary keys are not usable
    • Datafusion UPDATE semantics are tricky, we need to dissect the input LogicalPlan and convert to table scan of the current data (instead of blindly executing it), update records as we see them in the scan
    • #bug DELETE (#ref/gs5muy13tu1uk) or UPDATE (#ref/i61jpodrazscs) for a table using row_id (not primary keys)
    • #bug DELETE or UPDATE for a table using row_id (not primary keys) #ref/gs5muy13tu1uk #ref/i61jpodrazscs
    • possible fix: adjust logical_plan to also get a magic-named column __kanto_row_key, use that to do deletes
    • possible fix: understand the logical_plan, don't use DataFusion to produce record batches, run a custom table/filter scan that also deletes on the fly
  • return number of rows affected
  • index existing data at CREATE INDEX time #severity/high #urgency/high
    • ideally go straight to concurrent index building logic, no table locks
  • INSERT RETURNING (#ref/hixyde6h9n77o), UPDATE RETURNING (#ref/pp5wz7nq4yhzs), DELETE RETURNING (#ref/yn84qzsk1f9ny) refused by DataFusion SQL parser, not sure how to work around #ecosystem/datafusion #severity/high #urgency/medium
  • CTEs (test coverage): SELECT, UPDATE, INSERT, DELETE inside a CTE, SELECT, UPDATE, INSERT, DELETE following a CTE
  • rest of the data types
    • should we enforce VARCHAR(30) etc length, truncate; forbidding that syntax until clarified
    • INTEGER(10) etc precision
      • type { INTEGER(precision) | INT(precision) | INT4(precision) } (#ref/unimplemented_sql/tiptbb5d8buxa)
      • type { UNSIGNED INTEGER(precision) | UNSIGNED INT(precision) | UNSIGNED INT4(precision) } (#ref/unimplemented_sql/fzt1q8mfgkr6n)
      • type { (UNSIGNED) INT2(precision) | (UNSIGNED) SMALLINT(precision) } (#ref/unimplemented_sql/g57gd8499t4hg)
      • type { BIGINT(precision) | INT8(precision) } (#ref/unimplemented_sql/88fzntq3mz14g)
      • type { UNSIGNED BIGINT(precision) | UNSIGNED INT8(precision) } (#ref/unimplemented_sql/sze33j7qzt91s)
      • type TINYINT(precision) (#ref/unimplemented_sql/hueii8mjgk8qq)
      • type UNSIGNED TINYINT(precision) (#ref/unimplemented_sql/ewt6tneh7ecps)
    • type { NUMERIC | DECIMAL | DEC } (#ref/unimplemented_sql/uoakmzwax448h)
    • FLOAT(42) etc precision
      • type FLOAT(precision) (#ref/unimplemented_sql/xfwbr486jaw5q)
      • type DOUBLE(precision) (#ref/unimplemented_sql/wo43sei9ubpon)
    • NUMERIC(precision, scale), NUMERIC(precision), NUMERIC; aka DECIMAL(precision, scale)
    • type { CHARACTER | CHAR | FIXEDSTRING } (#ref/unimplemented_sql/hdhf8ygp9zna4): fixed-length strings
    • type INT128 (ClickHouse syntax) (#ref/unimplemented_sql/4sg3mftgxu9j4)
    • type INT256 (ClickHouse syntax) (#ref/unimplemented_sql/ex9mnxxipbqqs)
    • type UINT128 (ClickHouse syntax) (#ref/unimplemented_sql/riob7j6jahpte)
    • type UINT256 (ClickHouse syntax) (#ref/unimplemented_sql/q6tzjzd7xf3ga)
    • type { MEDIUMINT (UNSIGNED) } (#ref/unimplemented_sql/m7snzsrqj68sw)
    • type { BIGNUMERIC | BIGDECIMAL } (BigQuery syntax) (#ref/unimplemented_sql/an57g4d4uzwew)
    • type BIT (VARYING) (#ref/unimplemented_sql/88uneott9owc1)
    • type UUID (#ref/unimplemented_sql/531ax7gb73ce1)
    • type TIMESTAMP (#ref/unimplemented_sql/njad51sbxzwnn)
    • type TIME (#ref/unimplemented_sql/bppzhg7ck7xhr)
    • type DATE (#ref/unimplemented_sql/1r5ez1z8j7ryo)
    • type DATE32 (#ref/unimplemented_sql/781y5fxih7pnn)
    • type DATETIME (ClickHouse syntax) (#ref/unimplemented_sql/oak3s5g3rnutq)
    • type DATETIME (MySQL syntax) (#ref/unimplemented_sql/dojz85ngwo5wr)
    • type INTERVAL (#ref/unimplemented_sql/5wxynwyfua69s)
    • type { JSON | JSONB } (#ref/unimplemented_sql/abcw878jr35qc)
    • type BINARY (#ref/unimplemented_sql/i16wqwmie17eg)
    • type ENUM (#ref/unimplemented_sql/1geqh5qxdeoko)
    • type ARRAY (#ref/unimplemented_sql/hnmqj3qbtutn4)
    • type SET (#ref/unimplemented_sql/hm53sq3gco3sw)
    • type STRUCT (#ref/unimplemented_sql/nsapujxjjk9hg)
    • type MAP (#ref/unimplemented_sql/7ts8zgnafnp7g)
    • type UNION (#ref/unimplemented_sql/oownfhnfj5b9r)
    • type TUPLE (#ref/unimplemented_sql/x7qwkwdhdznxe)
    • type NESTED (#ref/unimplemented_sql/fa6nofmo1s49g)
    • type { CHARACTER LARGE OBJECT | CHAR LARGE OBJECT | CLOB } (#ref/unimplemented_sql/ew7uhufhkzj9w)
    • type TRIGGER (#ref/unimplemented_sql/zf5jx1ykoc5s1)
    • custom type (#ref/unimplemented_sql/dt315zutjqpzc)
    • type UNSPECIFIED (#ref/unimplemented_sql/916n48obgiqh6)
    • type LOWCARDINALITY (#ref/unimplemented_sql/78exey8xjz3yk)
    • type REGCLASS (Postgres syntax) (#ref/unimplemented_sql/quzjzcsusttgc)
    • type NULLABLE (ClickHouse syntax) (#ref/unimplemented_sql/ra5kspnstauu6)
  • emulate more Postgres system tables, like pg_cancel_backend

Unimplemented SQL syntax support

See also DataFusion SQL support status: https://arrow.apache.org/datafusion/user-guide/sql/sql_status.html, roadmap https://arrow.apache.org/datafusion/contributor-guide/roadmap.html.

For reference, see Postgres SQL, SQLite SQL, and MariaDB SQL.

  • SET LOCAL (#ref/unimplemented_sql/7b1ab6tms3uz4)
  • SHOW variable (#ref/unimplemented_sql/7a69iu5nr1oxk)
  • CREATE TABLE IF NOT EXISTS (#ref/unimplemented_sql/pf1mfhd9sz3jn)
  • DROP TABLE (#ref/unimplemented_sql/47hnz51gohsx6)
  • ALTER TABLE (#ref/unimplemented_sql/dny7j9hx3ihha)
  • CREATE INDEX IF NOT EXISTS (#ref/unimplemented_sql/tzwc96kk9173n)
  • DROP INDEX (#ref/unimplemented_sql/4kgkg4jhqhfrw)
  • ALTER INDEX (#ref/unimplemented_sql/ha1fybdqqwa8k)
  • SET TRANSACTION (#ref/unimplemented_sql/yybqytw7q8bmn)
    • may cause a bigger refactoring of our transaction start; we currently do eager creation
  • CREATE TABLE without columns (#ref/unimplemented_sql/g7hdmcm5bk69a)
  • constraints on tables #severity/high #urgency/high
    • CREATE TABLE with table constraint CHECK (#ref/unimplemented_sql/8zaqnh9mb9tor)
  • constraints on columns
    • CREATE TABLE with column option { PRIMARY KEY | UNIQUE } (#ref/unimplemented_sql/hwcjakao83bue) #severity/high #urgency/medium
    • CREATE TABLE with column option CHECK (#ref/unimplemented_sql/dwcpda8aj66d6) #severity/high #urgency/medium
  • default and computed values for columns
  • CREATE INDEX with dotted column name (#ref/unimplemented_sql/kbezqzh7p6yky)
  • CREATE INDEX with non-column expression (#ref/unimplemented_sql/8cjqfzsfpnduw)
  • CREATE INDEX WHERE (#ref/unimplemented_sql/9134bk3fe98x6): partial indexes
  • CREATE INDEX .. WITH (#ref/unimplemented_sql/w7t6c8xrehsnh)
  • CREATE INDEX .. INCLUDE (#ref/unimplemented_sql/ip415b5s8sa6h): non-key included columns
  • foreign keys #severity/high #urgency/high
    • CREATE TABLE with table constraint FOREIGN KEY (#ref/unimplemented_sql/mamfde4rdzgeo)
    • CREATE TABLE with column option REFERENCES (#ref/unimplemented_sql/ufnoiinkkp3wy)
    • enforce
  • roles and access control #severity/high #urgency/medium
    • CREATE ROLE (#ref/unimplemented_sql/hqyj3srdk1g4h)
    • DROP ROLE (#ref/unimplemented_sql/q8wh8nqixoytq)
    • ALTER ROLE (#ref/unimplemented_sql/u6set5k1gzyhh)
    • GRANT (#ref/unimplemented_sql/mekqqm6hxy64s)
    • REVOKE (#ref/unimplemented_sql/wha7884jy4bty)
    • SET ROLE (#ref/unimplemented_sql/ryqxdr55r49sr)
  • prepared statements
    • PREPARE (#ref/unimplemented_sql/fgi986yc4d7cw)
    • EXECUTE (#ref/unimplemented_sql/3w91mfrbzguyh)
    • DEALLOCATE (#ref/unimplemented_sql/fn1q45ys1t8ks)
  • views, first dynamic, then materialized, then incremental updates etc extras #severity/high
    • CREATE VIEW (#ref/unimplemented_sql/b9m5uhu9pnnsw)
    • ALTER VIEW (#ref/unimplemented_sql/uhwpi8nzojcxy)
    • DROP VIEW (#ref/unimplemented_sql/hwk7gp6ffh8zy)
  • TRUNCATE (#ref/unimplemented_sql/uf3rr4diw3o1o)
  • temporary tables
    • CREATE TEMPORARY TABLE (#ref/unimplemented_sql/3u561ykck5m76)
  • sequences (exposed to SQL)
    • CREATE SEQUENCE (#ref/unimplemented_sql/rdkjx9ryredcw)
    • DROP SEQUENCE (#ref/unimplemented_sql/ca149fq1ixm5h)
  • START TRANSACTION READ ONLY (#ref/unimplemented_sql/bhetj3emnc8n1)
  • multiple schemas
    • make schema creation explicit not implicit
    • CREATE SCHEMA (#ref/unimplemented_sql/x9cna73uf8cew)
    • DROP SCHEMA (#ref/unimplemented_sql/5rnpn33hia5a6)
  • multiple databases/catalogs/backends; all three might be the same thing for us
    • re-evaluate the whole "catalog name is backend" idea, especially when a backend can store multiple catalogs, duplication can lead to mismatch, renames are weird; maybe it's a lookup order?
      • is a backend a TABLESPACE?
        • sqlparser-rs doesn't have TABLESPACE support as of 2025-03
    • re-evaluate the idea of invisibly spreading every transaction across all backends; maybe force each transaction to stick to one, maybe make it lazy
    • Postgres docs specifically say "Databases are equivalent to catalogs"
    • CREATE DATABASE (#ref/unimplemented_sql/6c76p5eqsuea6)
    • ALTER DATABASE: not in sqlparser v0.54.0 #ecosystem/sqlparser-rs
    • DROP DATABASE (#ref/unimplemented_sql/qn6s3c397johw)
  • EXPLAIN (#ref/unimplemented_sql/fugjaua7zkfd4)
  • CREATE INDEX DESC (#ref/unimplemented_sql/n7x3zhjk7bf8k): descending sort order
  • CREATE INDEX CONCURRENTLY (#ref/unimplemented_sql/wya1a3w874xba)
  • COMMIT AND CHAIN (#ref/unimplemented_sql/ic5jpcqsjghm6)
  • ROLLBACK AND CHAIN (#ref/unimplemented_sql/q1pay866nux9n)
  • COPY (#ref/unimplemented_sql/1fxo667s3wwes): the kind that transfers data over wire; missing support in https://github.com/sunng87/pgwire #waiting
    • we do not intend to support anything where server-side file read/write gets path from user over network
  • RELEASE SAVEPOINT (#ref/unimplemented_sql/hyzhkewjpnug6): #severity/low #urgency/low #size/small
    • test error from trying to use it again
  • { LISTEN | UNLISTEN | NOTIFY } (#ref/unimplemented_sql/8t9hqpx4umzaq)
  • user-defined functions #severity/medium #urgency/low
    • CREATE FUNCTION (#ref/unimplemented_sql/fj9ny1gu79cbg)
    • DROP FUNCTION (#ref/unimplemented_sql/ptp9d3xe1456h)
    • SHOW FUNCTIONS (#ref/unimplemented_sql/yqiswgsqjc5uh): #severity/low
  • user-defined types #severity/medium #urgency/low
    • CREATE TYPE (#ref/unimplemented_sql/ur33y834br1wa): user defined types (see [Postgres](https://www.postgresql.org/docs/16/sql-createtype.html, DataFusion issue)
    • DROP TYPE (#ref/unimplemented_sql/t1fti3yz8jsdn)
  • collation, as in locale-specific ordering:
  • procedures #severity/medium #urgency/low
    • CREATE PROCEDURE (#ref/unimplemented_sql/69jzm59u8c6jc)
    • CALL (#ref/unimplemented_sql/dx5ic5bkaaq7c): see Postgres
    • DROP PROCEDURE (#ref/unimplemented_sql/w173ctuodtcaa)
  • triggers #severity/medium #urgency/low
    • CREATE TRIGGER (#ref/unimplemented_sql/cxts4du59pwqn)
    • DROP TRIGGER (#ref/unimplemented_sql/i1omqjczqksqk)
  • KILL [CONNECTION] (#ref/unimplemented_sql/jqji1j6my4f5w): #severity/medium #urgency/low
  • KILL QUERY (#ref/unimplemented_sql/77ecapaed3bso): #severity/medium #urgency/low
  • MERGE (#ref/unimplemented_sql/y5gzyy5kxg5ws): #severity/low #urgency/low
  • CREATE TABLE .. COMMENT (#ref/unimplemented_sql/dgyhzknpfx756)
  • CREATE TABLE with column option COMMENT (#ref/unimplemented_sql/4itjrujemcztq) #severity/medium #urgency/low
  • miscellaneous column options
  • COMMENT ON (#ref/unimplemented_sql/e8uhkzz1krcgw): #severity/low #urgency/low
  • populate table with data at creation time #severity/low #urgency/low
  • READ COMMITTED: currently upgrading to READ REPEATABLE, could support directly (#ref/bm4q45m5ajz9s)
  • START TRANSACTION ISOLATION LEVEL SERIALIZABLE (#ref/unimplemented_sql/dy1ch15p7n3ze): maybe just a better error message: #severity/low #urgency/low
    • i think supporting this requires two things: TransactionOptions::set_snapshot(true) and every get is upgraded to get_for_update.
    • then marking transactions as READ ONLY can relax that for performance
  • cursors (see Postgres; not in standard interactive SQL) #severity/low #urgency/low
    • DECLARE (#ref/unimplemented_sql/159jmakdyuewy)
    • FETCH (#ref/unimplemented_sql/9nf8hju6ie9e1)
    • CLOSE (#ref/unimplemented_sql/gc1557zek4hne)
  • row-level security (see Postgres)
    • CREATE POLICY (#ref/unimplemented_sql/gce4sb6hptjfy)
    • DROP POLICY (#ref/unimplemented_sql/n9zgxwn6736wr)
    • ALTER POLICY (#ref/unimplemented_sql/wj1tnq4hkth7s)
  • DISCARD (#ref/unimplemented_sql/9a5zkcwgbhewh): (see Postgres): #severity/low #urgency/low
  • REPLACE INTO (MySQL syntax) or INSERT OR REPLACE (SQLite syntax) (#ref/unimplemented_sql/54xqh7ornzpso): #severity/low #urgency/low
  • CREATE OR REPLACE TABLE (#ref/unimplemented_sql/65so99qr1hyck): #severity/low #urgency/low
  • ASSERT (PowerSQL syntax) (#ref/unimplemented_sql/qq1s1fq45ztdw): #severity/low #urgency/low
  • CREATE TABLE .. LIKE (#ref/unimplemented_sql/xutp6m7rjnc54)
  • CREATE TABLE .. CLONE (#ref/unimplemented_sql/7b5kah43yqquc): optionally also copies data; likely RocksDB backend will not support that option
  • CREATE TABLE .. WITH (#ref/unimplemented_sql/3kozdfoeczmrk): see Postgres, likely parse superficially and error if given #severity/low #urgency/low
  • RENAME TABLE (MySQL syntax) (#ref/unimplemented_sql/85ax9gtrkcneg)
  • CREATE TABLE with column option CHARACTER SET (#ref/unimplemented_sql/ifi7dygtua3oc): parse superficially and refuse non-utf8 #severity/low #urgency/low
  • CREATE TABLE .. DEFAULT CHARSET (MySQL syntax) (#ref/unimplemented_sql/dcan9mz8g4eja): parse superficially and refuse non-utf8 #urgency/low #severity/low #size/small
  • relaxing constraints to end of transaction:
    • CREATE TABLE with table constraint PRIMARY KEY .. DEFERRABLE (#ref/unimplemented_sql/uzf4oabd178rr)
    • CREATE TABLE with table constraint PRIMARY KEY .. INITIALLY DEFERRED (#ref/unimplemented_sql/ysnkhzw6ukbms)
  • BEGIN {DEFERRED | IMMEDIATE | EXCLUSIVE} (SQLite syntax) (#ref/unimplemented_sql/o969ikdsjz76r): #severity/low #urgency/low
  • CREATE TABLE with column option ON CONFLICT (SQLite syntax) (#ref/unimplemented_sql/ehefbqgnnfqwe) #severity/medium #urgency/low
  • PRAGMA (SQLite syntax) (#ref/unimplemented_sql/a1k38hk48fpzr): #severity/low #urgency/low
  • SHOW CREATE (MySQL syntax) (#ref/unimplemented_sql/c7cewp15zm5ys)
  • SHOW STATUS (MySQL syntax) (#ref/unimplemented_sql/ejj6ir6dsxhrc)
  • SHOW DATABASES (MySQL syntax) (#ref/unimplemented_sql/kskgotui7rupc)
  • SHOW COLUMNS (MySQL syntax) (#ref/unimplemented_sql/zx98bjxguht5o)
  • SHOW VARIABLES (MySQL syntax) (#ref/unimplemented_sql/n5cs7ue5gzpde)
  • SHOW SCHEMAS (Snowflake syntax) (#ref/unimplemented_sql/poi5oxunjyq5a)
  • SHOW VIEWS (Snowflake syntax) (#ref/unimplemented_sql/zzi1wb3jq5jeh)
  • { EXPLAIN | DESCRIBE } TABLE (MySQL syntax) (#ref/unimplemented_sql/4tpmma7ewdf3a): #severity/low #urgency/low
  • CREATE TABLE .. AUTO_INCREMENT (MySQL syntax) (#ref/unimplemented_sql/18kxgbi8rzb5r): #severity/low #urgency/low
  • USE (#ref/unimplemented_sql/ng34e6kt6rrt4): #severity/low #urgency/low
  • CREATE TABLE with table constraint PRIMARY KEY .. COMMENT (MySQL syntax) (#ref/unimplemented_sql/ss9iqz1c8ffrh)
  • MySQL-style names on constraints or indexes:
    • CREATE TABLE with table constraint PRIMARY KEY given a name (MySQL syntax) (#ref/unimplemented_sql/5rr7ouyasn1dk)
    • CREATE TABLE with table constraint PRIMARY KEY given an index name (MySQL syntax) (#ref/unimplemented_sql/mexuaio6ngad4)
  • CREATE TABLE with table constraint UNIQUE {INDEX | KEY} (MySQL syntax) (#ref/unimplemented_sql/m3op6enhb5pgw)
  • CREATE TABLE with table constraint PRIMARY KEY .. NOT ENFORCED (MySQL syntax) (#ref/unimplemented_sql/u339u8jzzkqdy)
  • CREATE TABLE with column option ON UPDATE timestamp function (MySQL syntax) (#ref/unimplemented_sql/zaub41ukgqfmq) #severity/low #urgency/low
  • CREATE TABLE with column option MATERIALIZED (ClickHouse syntax) (#ref/unimplemented_sql/ouq6dfrgbxn9w) #severity/low #urgency/low
  • CREATE TABLE with column option EPHEMERAL (ClickHouse syntax) (#ref/unimplemented_sql/78qipewmqn31h) #severity/low #urgency/low
  • CREATE TABLE with column option ALIAS (ClickHouse syntax) (#ref/unimplemented_sql/1adkef6yd1qgc) #severity/low #urgency/low
  • CREATE TABLE with column option OPTIONS (BigQuery syntax) (#ref/unimplemented_sql/7drpfhozbre1k) #severity/low #urgency/low
  • CREATE TABLE with column option TAG (Snowflake syntax) (#ref/unimplemented_sql/sjpe4f636kn8n) #severity/low #urgency/low
  • CREATE TABLE with column option { MASKING POLICY | PROJECTION POLICY } (Snowflake syntax) (#ref/unimplemented_sql/ynucdy5u7a35w) #severity/low #urgency/low
  • CREATE TABLE with weird column option (#ref/unimplemented_sql/4jqhus4cj4fho) #severity/low #urgency/low

There are plenty of corners of the SQL syntax we have chosen to not support. These are marked in the source code as NotSupportedSql. We may have judged wrong; let us know if you need some non-supported syntax.

Later

Features and improvements we would like to work on, but that aren't started yet.

Maybe

Possible features we have not at this time committed to implementing, or that are so far out that we just don't have a clear picture of how to implement them.

Small cleanups

  • exact_zip helper: we use Iterator::zip in many places where having non-equal lengths would be a bug, systematize

Ecosystem wishlist

DataFusion

Specific bugs and missing features:

Architectural:

sqlparser-rs

sqllogictest-rs

clippy (?)

tracing

Rust

misc

Details

RocksDB bind to C++ directly #maybe

RocksDB is a C++ library we're currently using through a less-featureful C shim. This crate allows calling C++ directly: https://github.com/dtolnay/cxx/. What are the downsides?

See also https://github.com/cozodb/cozo/tree/main/cozorocks, https://crates.io/crates/autocxx.

Compare to https://github.com/hkalbasi/zngur, https://crates.io/crates/cpp.

Prior work at https://github.com/google/autocxx/issues/1073. autocxx has been stagnant since 2022, so I'm pessimistic on this approach now.

MSRV policy

Streaming remote backup

Streaming remote backup for RocksDB backend