Configuring Persistence with RDB

1. Understanding RDB Snapshots

Property	Detail
Format	Compact binary point-in-time dump
Mechanism	fork() + write to temp + rename
Recovery	Fast bulk load on restart

save 3600 1
save 300 100
save 60 10000
save ""              # disable RDB

Pattern	Meaning
`save SECONDS CHANGES`	Snapshot if N keys changed in M seconds

Command	Behavior
`SAVE`	Synchronous; blocks server AVOID
`BGSAVE [SCHEDULE]`	Background fork-based snapshot
`LASTSAVE`	Unix ts of last successful save

Directive	Default
`dbfilename`	`dump.rdb`

Directive	Notes
`dir /var/lib/redis`	RDB + AOF written here

Command	Returns
`LASTSAVE`	Unix timestamp
`INFO persistence`	Status, last bgsave duration, error

Directive	Default	Effect
`rdbcompression yes`	yes	LZF compression — small CPU cost

Directive	Default
`rdbchecksum yes`	CRC64 footer, detects corruption

Directive	Default
`stop-writes-on-bgsave-error yes`	Reject writes if persistence fails

Warning: Forking large datasets causes COW page faults — disable Transparent Huge Pages (THP).

Mitigation	Detail
Disable THP	`echo never > /sys/kernel/mm/transparent_hugepage/enabled`
vm.overcommit_memory=1	Required to prevent OOM during fork

Strategy	Detail
Hourly snapshot	Copy `dump.rdb` to S3
`BGSAVE` + copy	Trigger via cron

Workload	Recommendation
High write	Larger thresholds to reduce fork frequency
Cache only	`save ""` (no persistence)
Critical	Combine RDB + AOF