Speed and security
Tabletist is made to be quick over a slow network and careful with what you trust it with. This page gathers what that means, with links to the details.
Fast
A native program
Tabletist is written in Rust and draws its own window, so there is no browser inside. Connecting, queries, exports and downloads run away from the window, so it keeps answering while the database works, and Cmd/Ctrl+. (Ctrl+C in the Omarchy look) cancels a running query.
Sorting, filtering and paging are done by the server, and only the page you look at comes over the network, so a table of millions of rows opens without loading them.
Few round trips
On a remote server most of the wait is the network, not the database. Each time a client sends something and waits for the answer is a round trip, and a server in another region or behind a VPN can take 100 to 200 ms for each. Tabletist sends what a step needs together, so on PostgreSQL, MySQL and MariaDB the steps you take most wait on the server once:
| What you do | Round trips |
|---|---|
| List the schemas and the default schema’s tables, once connected | 1 |
| Open a table from the sidebar: its structure and its first page | 1 |
| Load a page: the next one, or after a sort or a filter | 1 |
| Count a table’s rows | 1 |
Run a read-only SELECT in the SQL editor |
1 |
Through a network that adds 200 ms to every round trip, a table opens in about 0.2 seconds. From pressing Connect to seeing the tables takes about 1.2 seconds on PostgreSQL and 0.9 on MySQL and MariaDB, a little more with TLS, whose handshake has round trips of its own.
A few steps take more: a run that writes, a table whose key changed
since the sidebar listed it, a table whose structure the sidebar’s
listing did not read (opening it waits once for the structure and again
for the first page), and on MySQL and MariaDB a table with a FLOAT
column. A SQLite database is a file on your computer, with no
network in between.
Secure
Connections
- TLS. Choose
verify-fulland Tabletist checks the server’s certificate and its host name. When the mode you chose would let someone on the network read the password, the connection dialog says so. Read about TLS. - No password in plain text. When a MySQL server asks for the password in plain text, as AWS RDS IAM authentication does, Tabletist sends it only over TLS. Without TLS the connection fails instead.
- SSH tunnels. Sign in with a password, a key file or your SSH agent,
1Password’s included, and go through the
ProxyJumpandProxyCommandhosts your~/.ssh/confignames. For aProxyCommand, a host name or user with a character the shell would read is refused, not put into the command. Read about SSH tunnels. - Host keys. The first time you connect to an SSH server, Tabletist shows its key’s fingerprint and asks whether to trust it. If the key changes later, Tabletist refuses to connect, since that can mean someone is in the middle. Read about host keys.
Passwords
Passwords are never written to Tabletist’s files or its log. They are kept in the system keyring (the Keychain on macOS, Credential Manager on Windows, the Secret Service on Linux), or asked for each time you connect. A password you type when asked is saved only once the server accepted it, and agents never get it. Read about passwords.
Read-only, kept by the database
Read-only is the database’s to keep, not a guess from a statement’s words. With Open read-only the session itself is read-only on PostgreSQL and MySQL, and a SQLite file is opened read-only. A run in the SQL editor’s Read-only mode happens in a read-only transaction that is rolled back. BigQuery has no transactions, so a BigQuery connection is always read-only and runs only the statements Tabletist can tell are reads.
Production is careful by default:
- A production connection opens read-only unless you turn that off.
- With it off, a save asks first, with its statements on screen, and the SQL editor still runs read-only.
On every connection, a save is one transaction, and when someone else
changed the same row it writes nothing and asks you what to do. In the
SQL editor, a run that would change every row of a table (an UPDATE or
DELETE with no WHERE, or a TRUNCATE) asks before it runs.
Agents
Agents read only the connections you allow, within the rows and the time you set: on PostgreSQL, MySQL and SQLite in a read-only transaction that is rolled back, and on BigQuery only the statements Tabletist can tell are reads. They never get a connection’s password. Tables you hide and the values of columns you mask are kept from them, and so are views that would read what they may not. Every call goes into a log you can read, which keeps what was asked and never a row. On Linux and macOS, an administrator can set limits in a policy file that an agent running as you cannot change. Connect an agent and read about the administrator’s policy.
Releases and updates
- Each release lists its files in a
checksums.txtsigned with the Tabletist release key. The macOS app is signed by its developer and notarized by Apple. See the download page. - Before an update is installed, Tabletist checks the signature of the release’s checksums, the file’s checksum, and that the new version starts; on macOS also that it is Tabletist, signed by its developer and approved by Gatekeeper. A file that fails is deleted, and nothing is installed. Read about updates (the Updates pane).
The code
- Tabletist is open source under the MIT license, so anyone can read what it does with your connections. The code is on GitHub.
- Rust keeps memory safe. The code forbids
unsafe, apart from three small crates that call macOS’s AppKit, the menu bar and SQLite, each behind a safe interface. - The dependencies are checked against the RustSec advisory database every day, and on every change to them.