Apache Cassandra Cheat Sheet

01cqlsh & Sessionthe shell

02Keyspacesreplication container

CREATE KEYSPACE IF NOT EXISTS app WITH replication = {'class':'NetworkTopologyStrategy','dc1':3};★
Production default — per-datacenter replica count.
CREATE KEYSPACE dev WITH replication = {'class':'SimpleStrategy','replication_factor':1};
Single-DC / dev only — never for multi-DC.
ALTER KEYSPACE app WITH replication = {…};
Change RF, then run nodetool repair.
DROP KEYSPACE app;destroys
Deletes the keyspace and every table in it.

03Tablescolumns + keys

CREATE TABLE users ( id uuid PRIMARY KEY, email text, name text);★
Simple key — id is both partition & only key.
CREATE TABLE events ( user_id uuid, ts timestamp, kind text, PRIMARY KEY ((user_id), ts)) WITH CLUSTERING ORDER BY (ts DESC);★
Partition by user, rows newest-first. The core pattern.
PRIMARY KEY ((a, b), c, d)
Composite partition key (a,b) + clustering c,d.
ALTER TABLE users ADD phone text;
Add a column (cheap — no data rewrite).
DROP TABLE users;destroys
Removes the table and all its data.

04Primary Key Anatomythe whole game

PRIMARY KEY ( partition, clustering… )
First component(s) = partition key; the rest sort within it.
partition key★
Hashed to a token → picks the node. Every query must supply it.
clustering key
Sorts & uniquely locates rows inside a partition; enables range scans.
no partition key in WHEREanti
Forces a full-cluster scan → ALLOW FILTERING. Avoid.
keep partitions bounded
Aim < 100 MB / < 100k rows each — huge partitions kill nodes.

05Insert & Updatewrites are upserts

INSERT INTO users (id, email) VALUES (uuid(), 'a@b.io');★
Insert = upsert; same key silently overwrites.
INSERT INTO … USING TTL 86400;
Row expires after N seconds (auto-delete).
UPDATE users SET name='Zoe' WHERE id=?;★
Updates the row, or creates it if absent (upsert).
UPDATE t SET n = n + 1 WHERE …;
Counter columns — increment/decrement only.
INSERT … IF NOT EXISTS;LWT
Lightweight transaction — Paxos, much slower.

06Select & Queryquery-first rules

SELECT * FROM events WHERE user_id=?;★
Read a whole partition — the fast path.
WHERE user_id=? AND ts > ?★
Range on the clustering key — always cheap.
ORDER BY ts DESC LIMIT 20;
Only along the clustering order; LIMIT caps rows.
SELECT token(user_id), * FROM …
Inspect the token that placed a partition.
SELECT … ALLOW FILTERING;care
Scans/filters off-key columns — unpredictable at scale.

07Delete & Truncatewrites tombstones

DELETE email FROM users WHERE id=?;
Delete one column (sets it NULL).
DELETE FROM users WHERE id=?;★
Delete a whole row by primary key.
DELETE FROM events WHERE user_id=?;tombstones
Deletes a partition → many tombstones; hurts later reads.
TRUNCATE users;destroys
Wipes every row instantly, all nodes.

08Data Typescolumn types

text · varchar · ascii
UTF-8 / ASCII strings.
int · bigint · varint · decimal · double
Numbers; counter for distributed counts.
uuid · timeuuid★
timeuuid embeds time & sorts — great clustering key.
timestamp · date · time · duration
Temporal types (ms precision on timestamp).
boolean · blob · inet
Flags, raw bytes, IP addresses.

09Collections & UDTnested data

tags set<text>, log list<text>, attrs map<text,text>
Three collection types; keep them small (< a few KB).
UPDATE t SET tags = tags + {'vip'} …★
Add to a set/list without reading first.
UPDATE t SET attrs['k'] = 'v' …
Set / update a single map entry.
CREATE TYPE address (city text, zip text);
User-defined type; embed as frozen<address>.

10Batchatomic, not faster

BEGIN BATCH INSERT INTO … ; UPDATE … ; APPLY BATCH;
Logged batch = atomic across statements.
BEGIN UNLOGGED BATCH … APPLY BATCH;
No atomicity guarantee — only for one partition.
multi-partition batchanti
Spanning partitions to "save round-trips" hurts — don't.

11Indexes & Viewsquery other columns

CREATE INDEX ON users (email);
2i secondary index — OK for low cardinality only.
CREATE CUSTOM INDEX ON t (col) USING 'StorageAttachedIndex';★
SAI (Cassandra 5.0) — the modern, faster index.
CREATE MATERIALIZED VIEW mv AS SELECT * FROM base WHERE … PRIMARY KEY (…);
Auto-maintained table keyed differently for a 2nd query.

12Consistency Levelstunable per query

13Roles & SecurityDCL

CREATE ROLE app WITH PASSWORD='x' AND LOGIN=true;
Create a login role (replaces CREATE USER).
GRANT SELECT ON app.users TO app;
Grant a permission on a resource.
LIST ROLES; LIST PERMISSIONS;
Audit who can do what.
DROP ROLE app;care
Removes the role & its grants.

14Ops · nodetoolrun on a node

★The WHERE Ruleswhy queries fail

1. give the full partition key
Equality on every partition-key column — required.
2. clustering keys left-to-right
Restrict them in definition order, no gaps.
3. range only on the last one used
Earlier clustering cols need =; one may use < >.
4. ORDER BY follows clustering
You can only sort along the defined order (or reversed).
else → ALLOW FILTERINGsmell
If you need it, the table is modelled for the wrong query.

cassandra cheat sheet