Cassandra lab

The console works like cqlsh. Type a statement (ending in ; is optional) and press Enter; Shift+Enter adds a line, and a BEGIN BATCH waits for its APPLY BATCH. Results print as cqlsh prints them — the right-aligned table, (3 rows), null, True, timestamps as 2026-01-01 10:00:00.000000+0000 — and errors carry Cassandra’s own messages, such as InvalidRequest: … message="Cannot execute this query as it might involve data filtering …".

What the built-in node runs: keyspaces (SimpleStrategy, NetworkTopologyStrategy), tables with composite partition keys, clustering columns and CLUSTERING ORDER BY, static columns, all the native types, list, set, map, tuple, frozen and user-defined types, counters, default_time_to_live and gc_grace_seconds; INSERT (and INSERT … JSON), UPDATE with collection operations, DELETE of cells, rows, clustering ranges and partitions, USING TTL and USING TIMESTAMP, batches, lightweight transactions (IF NOT EXISTS, IF EXISTS, IF col = …), and SELECT with every restriction Cassandra allows (=, IN, clustering ranges, multi-column (c1, c2) > (…), token() ranges, CONTAINS), ORDER BY, LIMIT, PER PARTITION LIMIT, DISTINCT, GROUP BY, COUNT/MIN/MAX/SUM/AVG, TTL(), WRITETIME(), token(), SELECT JSON, now(), uuid(), toTimestamp() and friends. Storage-attached indexes (CREATE INDEX … USING 'sai') let a query restrict an indexed column. Writes are last-write-wins cells with timestamps; reads merge the memtable and every SSTable holding the partition.

Pick a table (yours, an example’s, or — when you are connected — one read from your server’s system_schema):

The memtable and the SSTables of one table, newest first, with each partition fragment’s live rows and tombstones; Flush, Compact (a major compaction) and clock buttons (+1 h, +1 day, +10 days — gc_grace_seconds is 10 days by default). Compaction drops data shadowed by newer writes and tombstones, and purges tombstones — including expired TTL cells, which become tombstones — only once gc_grace_seconds has passed. A chart follows tombstones and live rows after each step. Reads over more than 1,000 tombstones print Cassandra’s warning (tombstone_warn_threshold); 100,000 fail the read (tombstone_failure_threshold). Writing a whole non-frozen collection also writes a tombstone for its old contents, and the console says so.

Nodes, data centres, replication factor, num_tokens (vnodes, 1–256) and token allocation (random as before Cassandra 4.0, or evenly spread as the allocator does). The ring colours each node’s token ranges; type a partition key to see its Murmur3 token and its replicas (NetworkTopologyStrategy picks RF distinct nodes per DC, preferring new racks; SimpleStrategy the next nodes clockwise). Choose a table to plot its partitions at their tokens (dot size: partition size). The ownership bars show how evenly the ranges are spread — more vnodes, more even. Tokens are computed exactly as Cassandra’s Murmur3Partitioner does (MurmurHash3 x64_128 with Cassandra’s sign-extension quirk; composite keys are CompositeType-encoded), and tested against tokens from a real server.

A coordinator, the replicas of one partition (RF per DC, up to three DCs) and a consistency level for a read or a write. Send another request animates the requests and replies (latencies are drawn from a seeded log-normal around the median you set, plus a cross-DC delay); the coordinator answers once enough replicas have replied. Click replicas to take them down: the result shows success, Unavailable (too few replicas alive — checked before anything is sent) or a timeout. It also says how many replicas the level needs (QUORUM = ⌊ΣRF/2⌋ + 1, LOCAL_QUORUM = ⌊RF_local/2⌋ + 1, EACH_QUORUM a quorum in every DC), how many may be lost, and a read/write matrix of which combinations satisfy R + W > RF. Full screen (⛶, or F) works on the Ring and Consistency diagrams and on the console.

Docs covers the CQL statements, the WHERE rules, ALLOW FILTERING, TTL, collections, counters, lightweight transactions, and the concepts behind them (partition keys, clustering, query-first modelling, tombstones, memtables and SSTables, consistency, replication, hints and read repair, paging, the native protocol), with examples you can Try in a fresh built-in node or run in the console; each page links to the Apache Cassandra documentation. History keeps statements per connection, only in this browser; anything with a password is never saved.

Choose Real server. The page speaks Cassandra’s native protocol itself — version 5 with its checksummed segments (CRC24 headers, CRC32 payloads), falling back to version 4 — with PasswordAuthenticator (SASL PLAIN), paging, warnings and tracing. Browsers cannot open TCP connections, so the bytes travel through:

The gitwebsite app. If it is running on your computer, keep ⚡ gitwebsite app, enter host:port (for example 127.0.0.1:9042) and Connect. The app’s TCP relay connects to Cassandra; the first time, the app asks you to allow that target (it must be on its TCP allow-list).

A WebSocket bridge when the app is not running, for example with websockify, listening on this machine only:

and connect to ws://localhost:9043. Anyone who can reach a bridge can reach your database: keep it on 127.0.0.1 and stop it when you are done.

On a real server DROP, TRUNCATE, ALTER, role changes, DELETE of whole partitions or ranges, and batches of deletes ask before they run. The username and password stay in memory unless you tick Remember (then they are kept in this browser’s storage, readable by anyone using it); they never go into links or history. Show frames prints each protocol frame (opcode, stream, length and the first bytes). The Model tab reads your server’s schema, never its data.

Related: Partitioning lab, Redis client, Postgres playground, MongoDB playground, Kafka simulator, Latency numbers, Probability.

Links can be written by hand. Add the tab after #: console, model, storage, ring, consistency, docs or history. A link never connects to a server and never carries a password.

Presets for example=: