sqlite2dir
exposes the contents of an SQLite 3 database as a
collection of plain-text files. It's intended use case is not for
database backups -- the view provided is intended to allow humans to
more easily inspect and track changes to an SQLite database. The
output format is chosen so that tools designed to operate on
plain-text files, like diff
and git
should work well.
To allow for change tracking, sqlite2dir
supports committing the
tree of files resulting from the database export directly to a bare
git repository, which allows inspecting the history of changes using
regular git tools.
Note that sqlite2dir
is currently in its initial development phase,
and hasn't even been deployed by its author. The usual caveats apply.
The documentation for sqlite2dir
comes in the form of man
page. The markdown file can be turned in to troff
format for viewing with the man
command using pandoc. Note that to
the markdown source is tailored toward producing good output when fed
through pandoc, and will not be rendered nicely on github or alike,
and is not ideal to read in plain, either.
Generate and view the man page using the Unix man
command:
pandoc -s -t man sqlite2dir.1.md -o sqlite2dir.1
man -l sqlite2dir.1
You can also find a pandoc HTML rendering of the manpage online.
As sqlite2dir
is written in Rust, you need a Rust toolchain. Rust
1.37 or newer is required. To obtain the latest release from
crates.io, use:
cargo install sqlite2dir
Alternatively, you can run it directly from the source checkout:
cargo run -- --help
cargo run -- db.sqlite3 db-contents
To install from locally checked-out source, use cargo install --path .
, which will end up installing the executable in
~/.cargo/bin/sqlite2dir
, which should already be in your PATH
environment variable, if you followed the Rust toolchain installations
instructions.
For deployment to a Linux target, an attractive option is to create a statically linked binary using Rust's MUSL target. This will result in a completely standalone binary, which depends only on the Linux kernel's system call ABI.
In this case, you need to enable the vendored-sqlite
feature flag to
link against an embedded, newly-compiled, copy of libsqlite3
:
# If you haven't installed the MUSL target already, let's do that now:
rustup target add x86_64-unknown-linux-musl
# Build using a compiled-in copy of libsqlite3
cargo build --target x86_64-unknown-linux-musl --features vendored-sqlite --release
# Let's check it's really a static binary
file target/x86_64-unknown-linux-musl/release/sqlite2dir \
| grep -q 'statically linked' || echo "nope"
Create a dump of an sqlite3 database to a directory:
sqlite2dir db.sqlite3 db-contents
Inside the newly created db-contents
directory, you will find a
collection of SQL files containing the database Schema, and a JSON
file per table with the table contents.
The format of the SQL table data files is a stream of JSON arrays, each row being a single line containing a stand-alone JSON array containing the column data for a single database row. This format has been chosen to fulfill the following criteria:
- Reasonable diff output, while preserving the type of the values. In
particular, NULL values are represented as JSON
null
, and so can be disambiguated from a "NULL" string or an empty string. - Allow streaming creation and consumption with JSON parsers and serializers that operate on whole values.
Note that the SQLite "blob" data type is not yet supported, and the database dump will be aborted if a blob is encountered. See "Planned features" below for details.
These features are planned, roughly in the order of the author's perceived importance. During development, items will be moved from below into the changelog upon completion.
- An option to generate a short report, suitable as an email message body.
- A test harness including some basic smoke tests.
- Support for the SQLite "blob" data type. A basic implementation
would be to hash the blob content, and spit it out disk as its own
file. The DB column would then contain a reference like
{"blob-sha3-256": "SHA-3-here"}
. An improvement would be to base64-encode small blobs, and store them inline.
- Add support for a
--run
argument, to specify a config file allowing for multiple DB extractions in a single run. - With
--run
, add possibility for multi-threaded operation. - Additional database backends. I don't anticipate having the need for this feature, so I probably won't add it myself. Pull requests welcome!
sqlite2dir
is not, is not intended to be, and will, in all
likelihood, never become a database backup tool. SQLite provides the
.dump
and .backup
meta-commands its command-line tool, these
should be used instead. That way, it is even possible to restore the
data!
This is the scenario which prompted the development of sqlite2dir
.
The PowerDNS (aka pdns
) authoritative nameserver
provides several database backend, in addition to the "bind" backend,
which operates on plain-text zone files. The use of a database backend
is more flexible, but prevents easily tracking changes to the zone
content. When using plain text zone files, change tracking is easily
achieved by just putting the zone files into a git repository. Using
sqlite2dir
, you can recover that functionality when using the SQLite
pdns backend.
The following command will extract the database and commit to a bare git repository:
sqlite2dir --git-name="Clara Root" --git-email="root@localhost" \
/var/lib/pdns/pdns.sqlite3 /var/lib/pdns/pdns.git
By adding a periodic job executing the above command, e.g., via cron
or systemd
timers, one can accumulate history in a bare git
repository, which can be cloned and inspected for troubleshooting or
other analysis.
The code and documentation in the sqlite2dir
crate is free
software, licensed under
the GNU GPL, version 3.0 or later, at your option.