Databases
What makes Big Data?
- Volume
- Variety
- Velocity
- Veracity
Log-Structured Merge Trees
Record data and changes in immutable segments.
When data is inserted/changed, the top level fills up and its data is copied into a new segment (compaction)
LevelDB, Apache HBase, Hyperdex, Cassandra, RocksDB, WiredTiger, Riak
Advantages:
- immutable storage segments easily cached+backed up
- writes are append-only, no need for read
- fragmentation not an issue (or replaced by simpler problems)
- less wear on SSDs (can’t update data in-place)
Jepsen
DBA Responsibilities
- installs/configures/updates/migrates databases
- develops and implements a backup strategy
- responsible for ensuring that the data can be recovered without losing transactions
- responsible for securing the databases
- authentication (setting up user accounts)
- authorization (ensuring permissions are correct)
- auditing (who did what to the database)
- storage and capacity planning
- performance monitoring and tuning
- troubleshooting
- high availability
- online backups
- clustering
- replication
- standby servers
- Handling large DBs
- table partitioning (Oracle)
- federated databases (SQL Server)
- replication (MySQL)
- an understanding of relational theory
- data extraction/transformation/loading (ETL)
Interesting DBA problems
ODBC Open Database Connectivity
https://docs.microsoft.com/en-us/sql/odbc/microsoft-open-database-connectivity-odbc
Cassandra
Based on the Amazon Dynamo paper
- Ring topology
- peer-to-peer, gossip protocol
- no special nodes
- Distributed Hash Table
- eventually consistent, tunable
Use vnodes whenever possible to avoid issues with topology changes, node rebuilds, hotspots, and heterogeneous clusters.
vnodes mean more ranges, which makes it easier to give a range to a new node
Keyspaces
List keyspaces
DESCRIBE KEYSPACES;
Create a keyspace
CREATE KEYSPACE "my_space"
WITH REPLICATION = {
'class': 'SimpleStrategy', 'replication_factor': 1
};
Use a keyspace
USE "my_space";
Tables
List tables
DESCRIBE TABLES;
Create table
CREATE TABLE "users" (
"user" text PRIMARY KEY,
"email" text,
"avatar" blob
);
Insert into table
INSERT INTO "users"
("avatar", "email", "avatar")
VALUES (
'admin',
'admin@example.org',
0xf00badc0ff33
);
Partitioners
- Murmur3Partitioner: even distribution of data across the cluster using the MurmurHash algorithm.
- RandomPartitioner: default prior to 1.2. Uses MD5 hashes. If you don’t use vnodes, you have to calculate the tokens.
- ByteOrderedPartitioner: orders rows lexically by key bytes. Not recommended, since it’s hard to load balance and can have hot spots.
ports
| desc | port |
|---|---|
| cqlsh | 9042 |
nodetool
https://docs.datastax.com/en/cassandra/2.1/cassandra/tools/toolsNodetool_r.html
# Checking node repair
nodetool netstats
nodetool compactionstats
Topologies
Uses ‘snitches’
- SimpleSnitch (default, good for 1 DC)
- RackInferringSnitch (infers from IP addr: 10.DC.RACK.NODE)
- PropertyFileSnitch (a KV file of IP=DC:RACK)
- EC2Snitch (discovers AWS AZ/regions)
Per-Query Consistency
ANY/ONE/QUORUM/LOCAL_QUORUM/ALL
Quick Docker dev setup
# Run server instance
docker run --name some-cassandra -v ~/my/own/datadir:/var/lib/cassandra -d cassandra:latest
Types
| name | desc |
|---|---|
| text | UTF8 |
| ascii | ASCII |
| int | 32-bit integer |
| bigint | 64-bit integer |
| varint | arbitrary size |
| float | 32-bit float |
| double | 64-bit float |
| decimal | variable-precision decimal |
| boolean | boolean |
| timestamp | ‘yyyy-mm-dd HH:mm:ssZ’ |
| uuid | UUID v1 and v4 |
| blob | binary (prefix with 0x) |
Resources
- Cassandra High Availability by X
Links
CouchDB
get list of databases
curl localhost:5984/_all_dbs
get database details
curl localhost:5984/db_name
access futon
curl localhost:5984/_utils
Links
Couchbase
Links
Elasticsearch
Running in Docker
docker pull docker.elastic.co/elasticsearch/elasticsearch:7.10.0
# single node
docker run -p 9200:9200 -p 9300:9300 -e "discovery.type=single-node" docker.elastic.co/elasticsearch/elasticsearch:7.10.0
Links
- https://www.elastic.co/guide/en/elasticsearch/reference/current/docker.html
- Kibana docker image
- Elasticsearch docker image
MongoDB
Queries
// Show current operations
db.currentOp()
// Show long running queries
db.currentOp()['inprog'].filter(function(x) {return x.secs_running > 10})
// Kill a query
db.killOp(12345)
Administration
// Show database list
show dbs
// Create/switch to database
use myDb
// Drop database
db.dropDatabase();
Rotate logs
kill -SIGUSR1
Documentation / Links
OrientDB
Time-series database
influxdb
Prometheus
Official docker container: prom/prometheus